{"cases": [{"id": "case00", "pair_id": "pair00", "split": "val", "start": 493, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 85, "task": "Arrange the numbers from left to right to form the largest possible number, and place them on the pad.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000085", "start_seconds": 19.72, "starts": [493], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nSubtask context (ordered by onset): Place the number 3 on the end of the row on the pad.\nAtomic instruction: right arm: Place the number 3 on the end of the row on the pad."], "file": "case00.npz", "horizons": [1], "history_indices": [0, 318, 343, 368, 393, 418, 443, 468, 493], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-003.mp4", "start": 262.91999999999996, "end": 292.79999999999995, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-002.mp4", "start": 111.16, "end": 141.04, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-002.mp4", "start": 132.88, "end": 162.76, "fps": 25}}, "source_npz_sha256": "9ba9362f92cf9ac456ec8b643cac744a7e299d62e83a87333f54aecb7391cf5a", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nSubtask context (ordered by onset): Place the number 3 on the end of the row on the pad.\nAtomic instruction: right arm: Place the number 3 on the end of the row on the pad.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nSubtask context (ordered by onset): Place the number 3 on the end of the row on the pad.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nAtomic instruction: right arm: Place the number 3 on the end of the row on the pad.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the number 3 on the end of the row on the pad.\nAtomic instruction: right arm: Place the number 3 on the end of the row on the pad.", "l3": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the number 3 on the end of the row on the pad."}, "input_packs": [{"chunk": 0, "start": 493, "path": "inputs/case00-chunk0.npz", "sha256": "e8350a34f292fd789f511060d9b080b76524ff651dc2e3c19da57fb0430b050a"}], "input_pack": "inputs/case00-chunk0.npz", "input_image": "inputs/case00-input.png", "gt_videos": {"1": "inputs/case00-gt-1.mp4"}, "history_images": ["inputs/case00-history-0.png", "inputs/case00-history-1.png", "inputs/case00-history-2.png", "inputs/case00-history-3.png", "inputs/case00-history-4.png", "inputs/case00-history-5.png", "inputs/case00-history-6.png", "inputs/case00-history-7.png", "inputs/case00-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 0*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.", "official_native_context": "official/texts/1741a03e7604b7db00a31b035dba3f07dbc455b6fdcbac1cc6038de3ff64b339.npy"}, {"id": "case01", "pair_id": "pair00", "split": "train", "start": 329, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 0, "task": "Arrange the numbers from left to right to form the largest possible number, and place them on the pad.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000000", "start_seconds": 13.16, "starts": [329], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nSubtask context (ordered by onset): Move the number 2 block to a temporary position on the table.\nAtomic instruction: right arm: Place the number 2 block to the left of the number 5 block."], "file": "case01.npz", "horizons": [1], "history_indices": [0, 154, 179, 204, 229, 254, 279, 304, 329], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-000.mp4", "start": 0.0, "end": 23.12, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-000.mp4", "start": 0.0, "end": 23.12, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-000.mp4", "start": 0.0, "end": 23.12, "fps": 25}}, "source_npz_sha256": "581098e397ebf7da309f8bbbceba248888674e6d44dbffacd7e522b4ae9108f1", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nSubtask context (ordered by onset): Move the number 2 block to a temporary position on the table.\nAtomic instruction: right arm: Place the number 2 block to the left of the number 5 block.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nSubtask context (ordered by onset): Move the number 2 block to a temporary position on the table.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.\nAtomic instruction: right arm: Place the number 2 block to the left of the number 5 block.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Move the number 2 block to a temporary position on the table.\nAtomic instruction: right arm: Place the number 2 block to the left of the number 5 block.", "l3": "Robot: robodojo_sim_arx_x5. Task: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the number 2 block to the left of the number 5 block."}, "input_packs": [{"chunk": 0, "start": 329, "path": "inputs/case01-chunk0.npz", "sha256": "28ea46409ce1412d0213cb3614f7f67ed13066eb068498b001140bbf6c678884"}], "input_pack": "inputs/case01-chunk0.npz", "input_image": "inputs/case01-input.png", "gt_videos": {"1": "inputs/case01-gt-1.mp4"}, "history_images": ["inputs/case01-history-0.png", "inputs/case01-history-1.png", "inputs/case01-history-2.png", "inputs/case01-history-3.png", "inputs/case01-history-4.png", "inputs/case01-history-5.png", "inputs/case01-history-6.png", "inputs/case01-history-7.png", "inputs/case01-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 1*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Arrange the numbers from left to right to form the largest possible number, and place them on the pad.", "official_native_context": "official/texts/1741a03e7604b7db00a31b035dba3f07dbc455b6fdcbac1cc6038de3ff64b339.npy"}, {"id": "case02", "pair_id": "pair01", "split": "val", "start": 439, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 163, "task": "Build a tower using the wooden blocks and wooden boards.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000163", "start_seconds": 17.56, "starts": [439], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nSubtask context (ordered by onset): Assemble the tower by placing the wooden boards onto the blocks.\nAtomic instruction: left arm: Grasp the wooden board on the table. right arm: Grasp the right edge of the lower wooden board."], "file": "case02.npz", "horizons": [1], "history_indices": [0, 264, 289, 314, 339, 364, 389, 414, 439], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-007.mp4", "start": 144.51999999999998, "end": 173.67999999999998, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-004.mp4", "start": 494.32000000000005, "end": 523.48, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-004.mp4", "start": 580.8799999999999, "end": 610.0399999999998, "fps": 25}}, "source_npz_sha256": "df60187e92dc136d71573b1a5133a2d1b98a909f5f9183c26335f3579d6312de", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nSubtask context (ordered by onset): Assemble the tower by placing the wooden boards onto the blocks.\nAtomic instruction: left arm: Grasp the wooden board on the table. right arm: Grasp the right edge of the lower wooden board.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nSubtask context (ordered by onset): Assemble the tower by placing the wooden boards onto the blocks.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nAtomic instruction: left arm: Grasp the wooden board on the table. right arm: Grasp the right edge of the lower wooden board.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Assemble the tower by placing the wooden boards onto the blocks.\nAtomic instruction: left arm: Grasp the wooden board on the table. right arm: Grasp the right edge of the lower wooden board.", "l3": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the wooden board on the table. right arm: Grasp the right edge of the lower wooden board."}, "input_packs": [{"chunk": 0, "start": 439, "path": "inputs/case02-chunk0.npz", "sha256": "cfba8b2218f64845488f5e1c4a1405c5b826146026cdc0589e15e1ecc1a97b80"}], "input_pack": "inputs/case02-chunk0.npz", "input_image": "inputs/case02-input.png", "gt_videos": {"1": "inputs/case02-gt-1.mp4"}, "history_images": ["inputs/case02-history-0.png", "inputs/case02-history-1.png", "inputs/case02-history-2.png", "inputs/case02-history-3.png", "inputs/case02-history-4.png", "inputs/case02-history-5.png", "inputs/case02-history-6.png", "inputs/case02-history-7.png", "inputs/case02-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 2*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Build a tower using the wooden blocks and wooden boards.", "official_native_context": "official/texts/9db10cd4522d16e5f507175db9796f62f564e3ebd0936d86e0c0d7c4a92391cf.npy"}, {"id": "case03", "pair_id": "pair01", "split": "train", "start": 437, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 100, "task": "Build a tower using the wooden blocks and wooden boards.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000100", "start_seconds": 17.48, "starts": [437], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nSubtask context (ordered by onset): Stack the two wooden boards together.\nAtomic instruction: left arm: Grasp the left side of the wooden board. right arm: Grasp the white wooden block on the right side of the table."], "file": "case03.npz", "horizons": [1], "history_indices": [0, 262, 287, 312, 337, 362, 387, 412, 437], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-004.mp4", "start": 66.8, "end": 95.67999999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-002.mp4", "start": 477.6, "end": 506.48, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-002.mp4", "start": 499.31999999999994, "end": 528.1999999999999, "fps": 25}}, "source_npz_sha256": "912317ef49a21cd9592c01c165348a8dcdee49d9b24bae57b4c61aa127ad7ff8", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nSubtask context (ordered by onset): Stack the two wooden boards together.\nAtomic instruction: left arm: Grasp the left side of the wooden board. right arm: Grasp the white wooden block on the right side of the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nSubtask context (ordered by onset): Stack the two wooden boards together.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.\nAtomic instruction: left arm: Grasp the left side of the wooden board. right arm: Grasp the white wooden block on the right side of the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Stack the two wooden boards together.\nAtomic instruction: left arm: Grasp the left side of the wooden board. right arm: Grasp the white wooden block on the right side of the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Build a tower using the wooden blocks and wooden boards.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the left side of the wooden board. right arm: Grasp the white wooden block on the right side of the table."}, "input_packs": [{"chunk": 0, "start": 437, "path": "inputs/case03-chunk0.npz", "sha256": "09a2e6e0c07036d326715c5b6904a7136253c4b9635252701da5bc04679d6680"}], "input_pack": "inputs/case03-chunk0.npz", "input_image": "inputs/case03-input.png", "gt_videos": {"1": "inputs/case03-gt-1.mp4"}, "history_images": ["inputs/case03-history-0.png", "inputs/case03-history-1.png", "inputs/case03-history-2.png", "inputs/case03-history-3.png", "inputs/case03-history-4.png", "inputs/case03-history-5.png", "inputs/case03-history-6.png", "inputs/case03-history-7.png", "inputs/case03-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 3*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Build a tower using the wooden blocks and wooden boards.", "official_native_context": "official/texts/9db10cd4522d16e5f507175db9796f62f564e3ebd0936d86e0c0d7c4a92391cf.npy"}, {"id": "case04", "pair_id": "pair02", "split": "val", "start": 359, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 251, "task": "Sort the objects by category into the three baskets.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000251", "start_seconds": 14.36, "starts": [359], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nSubtask context (ordered by onset): Put the white rabbit toy into the blue basket.\nAtomic instruction: right arm: Grasp the white rabbit toy on the right side of the table."], "file": "case04.npz", "horizons": [1], "history_indices": [0, 184, 209, 234, 259, 284, 309, 334, 359], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-011.mp4", "start": 411.3999999999999, "end": 437.4799999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-007.mp4", "start": 336.15999999999997, "end": 362.23999999999995, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-007.mp4", "start": 502.52000000000004, "end": 528.6, "fps": 25}}, "source_npz_sha256": "282b27496a2a74ea9ad4e90536484234e869eba8ad752064ac5d3f81faf4495d", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nSubtask context (ordered by onset): Put the white rabbit toy into the blue basket.\nAtomic instruction: right arm: Grasp the white rabbit toy on the right side of the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nSubtask context (ordered by onset): Put the white rabbit toy into the blue basket.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nAtomic instruction: right arm: Grasp the white rabbit toy on the right side of the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Put the white rabbit toy into the blue basket.\nAtomic instruction: right arm: Grasp the white rabbit toy on the right side of the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the white rabbit toy on the right side of the table."}, "input_packs": [{"chunk": 0, "start": 359, "path": "inputs/case04-chunk0.npz", "sha256": "9047c5a1f82ba7769425d5c059df5cbb4968b1b98927ff9c4adc03c273ec6483"}], "input_pack": "inputs/case04-chunk0.npz", "input_image": "inputs/case04-input.png", "gt_videos": {"1": "inputs/case04-gt-1.mp4"}, "history_images": ["inputs/case04-history-0.png", "inputs/case04-history-1.png", "inputs/case04-history-2.png", "inputs/case04-history-3.png", "inputs/case04-history-4.png", "inputs/case04-history-5.png", "inputs/case04-history-6.png", "inputs/case04-history-7.png", "inputs/case04-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 4*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Sort the objects by category into the three baskets.", "official_native_context": "official/texts/58d29cbe871f0d0de8e26ca852e3281866575cd57fdbb85f138c23e68cf6ce70.npy"}, {"id": "case05", "pair_id": "pair02", "split": "train", "start": 467, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 200, "task": "Sort the objects by category into the three baskets.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000200", "start_seconds": 18.68, "starts": [467], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nSubtask context (ordered by onset): Put the beige rectangular object into the blue basket.\nAtomic instruction: left arm: Place the toy car into the blue basket."], "file": "case05.npz", "horizons": [1], "history_indices": [0, 292, 317, 342, 367, 392, 417, 442, 467], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-009.mp4", "start": 58.2, "end": 88.08, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-005.mp4", "start": 668.3199999999999, "end": 698.1999999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-005.mp4", "start": 754.9200000000001, "end": 784.8000000000001, "fps": 25}}, "source_npz_sha256": "b040959d52af284c8b2ffed8192fd627edfafccf710706c44cabd7fa5e03804d", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nSubtask context (ordered by onset): Put the beige rectangular object into the blue basket.\nAtomic instruction: left arm: Place the toy car into the blue basket.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nSubtask context (ordered by onset): Put the beige rectangular object into the blue basket.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.\nAtomic instruction: left arm: Place the toy car into the blue basket.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Put the beige rectangular object into the blue basket.\nAtomic instruction: left arm: Place the toy car into the blue basket.", "l3": "Robot: robodojo_sim_arx_x5. Task: Sort the objects by category into the three baskets.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the toy car into the blue basket."}, "input_packs": [{"chunk": 0, "start": 467, "path": "inputs/case05-chunk0.npz", "sha256": "99eea0570d179f2432cd2fd6b23a1f9525cdac3aeef1cb383526f6159388ef09"}], "input_pack": "inputs/case05-chunk0.npz", "input_image": "inputs/case05-input.png", "gt_videos": {"1": "inputs/case05-gt-1.mp4"}, "history_images": ["inputs/case05-history-0.png", "inputs/case05-history-1.png", "inputs/case05-history-2.png", "inputs/case05-history-3.png", "inputs/case05-history-4.png", "inputs/case05-history-5.png", "inputs/case05-history-6.png", "inputs/case05-history-7.png", "inputs/case05-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 5*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Sort the objects by category into the three baskets.", "official_native_context": "official/texts/58d29cbe871f0d0de8e26ca852e3281866575cd57fdbb85f138c23e68cf6ce70.npy"}, {"id": "case06", "pair_id": "pair03", "split": "val", "start": 354, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 326, "task": "Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000326", "start_seconds": 14.16, "starts": [354], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nSubtask context (ordered by onset): Uncover the red block by removing its yellow cup.; Uncover the green block by removing its yellow cup.\nAtomic instruction: right arm: Grasp the yellow cup covering the green block."], "file": "case06.npz", "horizons": [1], "history_indices": [0, 179, 204, 229, 254, 279, 304, 329, 354], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-015.mp4", "start": 131.04, "end": 153.04, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-009.mp4", "start": 598.4399999999999, "end": 620.4399999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-009.mp4", "start": 815.08, "end": 837.08, "fps": 25}}, "source_npz_sha256": "4c57fab5b26193cafba07519925adad9187f273ee9b6e9d435fc4ed6f57836a3", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nSubtask context (ordered by onset): Uncover the red block by removing its yellow cup.; Uncover the green block by removing its yellow cup.\nAtomic instruction: right arm: Grasp the yellow cup covering the green block.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nSubtask context (ordered by onset): Uncover the red block by removing its yellow cup.; Uncover the green block by removing its yellow cup.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nAtomic instruction: right arm: Grasp the yellow cup covering the green block.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Uncover the red block by removing its yellow cup.; Uncover the green block by removing its yellow cup.\nAtomic instruction: right arm: Grasp the yellow cup covering the green block.", "l3": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the yellow cup covering the green block."}, "input_packs": [{"chunk": 0, "start": 354, "path": "inputs/case06-chunk0.npz", "sha256": "e0f5b92c5cfab2b5173ec52eb158ced6ab852f63a02f4534f7587be5f14e823a"}], "input_pack": "inputs/case06-chunk0.npz", "input_image": "inputs/case06-input.png", "gt_videos": {"1": "inputs/case06-gt-1.mp4"}, "history_images": ["inputs/case06-history-0.png", "inputs/case06-history-1.png", "inputs/case06-history-2.png", "inputs/case06-history-3.png", "inputs/case06-history-4.png", "inputs/case06-history-5.png", "inputs/case06-history-6.png", "inputs/case06-history-7.png", "inputs/case06-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 6*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.", "official_native_context": "official/texts/88d940ae91d28e7505208769ed176e72da76b1629d03dd9498d85185d88942c7.npy"}, {"id": "case07", "pair_id": "pair03", "split": "train", "start": 309, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 300, "task": "Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000300", "start_seconds": 12.36, "starts": [309], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nSubtask context (ordered by onset): Uncover the blue block by removing its yellow cup with the right arm.\nAtomic instruction: right arm: Place the yellow cup over the blue block."], "file": "case07.npz", "horizons": [1], "history_indices": [0, 134, 159, 184, 209, 234, 259, 284, 309], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-014.mp4", "start": 146.36, "end": 168.56, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-009.mp4", "start": 32.16, "end": 54.36, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-009.mp4", "start": 248.80000000000004, "end": 271.00000000000006, "fps": 25}}, "source_npz_sha256": "1df2505cd4085ec5e03ae42c94d1e80a810a9df9ec98951e2e638fe64b9a6023", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nSubtask context (ordered by onset): Uncover the blue block by removing its yellow cup with the right arm.\nAtomic instruction: right arm: Place the yellow cup over the blue block.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nSubtask context (ordered by onset): Uncover the blue block by removing its yellow cup with the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.\nAtomic instruction: right arm: Place the yellow cup over the blue block.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Uncover the blue block by removing its yellow cup with the right arm.\nAtomic instruction: right arm: Place the yellow cup over the blue block.", "l3": "Robot: robodojo_sim_arx_x5. Task: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the yellow cup over the blue block."}, "input_packs": [{"chunk": 0, "start": 309, "path": "inputs/case07-chunk0.npz", "sha256": "dbb709f572b6c0c9cd9e487dbc22b95462535416866a30ff96a5b6761245b8f7"}], "input_pack": "inputs/case07-chunk0.npz", "input_image": "inputs/case07-input.png", "gt_videos": {"1": "inputs/case07-gt-1.mp4"}, "history_images": ["inputs/case07-history-0.png", "inputs/case07-history-1.png", "inputs/case07-history-2.png", "inputs/case07-history-3.png", "inputs/case07-history-4.png", "inputs/case07-history-5.png", "inputs/case07-history-6.png", "inputs/case07-history-7.png", "inputs/case07-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 7*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue.", "official_native_context": "official/texts/88d940ae91d28e7505208769ed176e72da76b1629d03dd9498d85185d88942c7.npy"}, {"id": "case08", "pair_id": "pair04", "split": "val", "start": 61, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 430, "task": "Pick up the coin from the holder and insert it precisely into the coin bank.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000430", "start_seconds": 2.44, "starts": [61], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.\nAtomic instruction: right arm: Insert the held coin into the slot of the pink piggy bank."], "file": "case08.npz", "horizons": [1], "history_indices": [0, 0, 0, 0, 0, 0, 11, 36, 61], "distinct_history_frames": 4, "history_padding": [false, true, true, true, true, true, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-018.mp4", "start": 178.16, "end": 182.72, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-011.mp4", "start": 193.79999999999995, "end": 198.35999999999996, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-011.mp4", "start": 717.0799999999999, "end": 721.6399999999999, "fps": 25}}, "source_npz_sha256": "86a5cf70b74fde73f46684c20429e06ec0dc5d470141e64b06d042d1f5a66146", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.\nAtomic instruction: right arm: Insert the held coin into the slot of the pink piggy bank.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nAtomic instruction: right arm: Insert the held coin into the slot of the pink piggy bank.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.\nAtomic instruction: right arm: Insert the held coin into the slot of the pink piggy bank.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Insert the held coin into the slot of the pink piggy bank."}, "input_packs": [{"chunk": 0, "start": 61, "path": "inputs/case08-chunk0.npz", "sha256": "2ed2a43534fdaa7a14638273d88a6def54f0a3dc03df2bc91b815d00bbafb8c5"}], "input_pack": "inputs/case08-chunk0.npz", "input_image": "inputs/case08-input.png", "gt_videos": {"1": "inputs/case08-gt-1.mp4"}, "history_images": ["inputs/case08-history-0.png", "inputs/case08-history-1.png", "inputs/case08-history-2.png", "inputs/case08-history-3.png", "inputs/case08-history-4.png", "inputs/case08-history-5.png", "inputs/case08-history-6.png", "inputs/case08-history-7.png", "inputs/case08-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 8*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the coin from the holder and insert it precisely into the coin bank.", "official_native_context": "official/texts/99b200c4a98d87dd293d47ac9cb81b4fdd724ee9aecb35237d16c891248ee562.npy"}, {"id": "case09", "pair_id": "pair04", "split": "train", "start": 135, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 400, "task": "Pick up the coin from the holder and insert it precisely into the coin bank.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000400", "start_seconds": 5.4, "starts": [135], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.\nAtomic instruction: left arm: Place the coin into the slot of the pink piggy bank."], "file": "case09.npz", "horizons": [1], "history_indices": [0, 0, 0, 10, 35, 60, 85, 110, 135], "distinct_history_frames": 7, "history_padding": [false, true, true, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-017.mp4", "start": 564.96, "end": 572.6, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-010.mp4", "start": 1106.7999999999995, "end": 1114.4399999999996, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-011.mp4", "start": 499.9200000000001, "end": 507.56000000000006, "fps": 25}}, "source_npz_sha256": "b9fe3fde433b762ee39ad8d86bc82985fcae6e636cb046ba24a947371380f096", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.\nAtomic instruction: left arm: Place the coin into the slot of the pink piggy bank.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.\nAtomic instruction: left arm: Place the coin into the slot of the pink piggy bank.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Insert the coin into the slot of the pink piggy bank.\nAtomic instruction: left arm: Place the coin into the slot of the pink piggy bank.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the coin from the holder and insert it precisely into the coin bank.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the coin into the slot of the pink piggy bank."}, "input_packs": [{"chunk": 0, "start": 135, "path": "inputs/case09-chunk0.npz", "sha256": "b9100b77bc4c6684bbb27d18b1aa9300f42964ca0ce88408d333b5fb35740ad6"}], "input_pack": "inputs/case09-chunk0.npz", "input_image": "inputs/case09-input.png", "gt_videos": {"1": "inputs/case09-gt-1.mp4"}, "history_images": ["inputs/case09-history-0.png", "inputs/case09-history-1.png", "inputs/case09-history-2.png", "inputs/case09-history-3.png", "inputs/case09-history-4.png", "inputs/case09-history-5.png", "inputs/case09-history-6.png", "inputs/case09-history-7.png", "inputs/case09-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 9*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the coin from the holder and insert it precisely into the coin bank.", "official_native_context": "official/texts/99b200c4a98d87dd293d47ac9cb81b4fdd724ee9aecb35237d16c891248ee562.npy"}, {"id": "case10", "pair_id": "pair05", "split": "val", "start": 593, "max_chunks": 8, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 604, "task": "Insert and tighten each screw into the nut of the same color.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000604", "start_seconds": 23.72, "starts": [593, 625, 657, 689, 721, 753, 785, 817], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again."], "file": "case10.npz", "horizons": [1, 3, 8], "history_indices": [0, 418, 443, 468, 493, 518, 543, 568, 593], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-029.mp4", "start": 0.0, "end": 57.56, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-017.mp4", "start": 111.6, "end": 169.16, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-017.mp4", "start": 434.68, "end": 492.24, "fps": 25}}, "source_npz_sha256": "7e79a23a603c505a96c837aa39a7e05e3c7d5ab7509dfb98300444b3dc2ce56b", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nAtomic instruction: left arm: Grasp the white nut again.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Attempt to grasp and manipulate the white nut with the left arm.\nAtomic instruction: left arm: Grasp the white nut again.", "l3": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the white nut again."}, "input_packs": [{"chunk": 0, "start": 593, "path": "inputs/case10-chunk0.npz", "sha256": "3adcbd953c5a329a949f28683ae3efd5ac61e2e878dd4f7522feec9c76abe3fb"}, {"chunk": 1, "start": 625, "path": "inputs/case10-chunk1.npz", "sha256": "8d46b80958b285bbc6fa24852b83512d505ea28e14ba2e7eeca7b2f9c4a6ec04"}, {"chunk": 2, "start": 657, "path": "inputs/case10-chunk2.npz", "sha256": "53b2686e87941819599bed9a0e2f14e64caf8d1ab740b1d123e50f030929a8c5"}, {"chunk": 3, "start": 689, "path": "inputs/case10-chunk3.npz", "sha256": "31e95cbe0119c2785fe02bd17284588c039a3656d582b74dcb0da61b30da5aeb"}, {"chunk": 4, "start": 721, "path": "inputs/case10-chunk4.npz", "sha256": "178a85187e052a76820602b72d437366f1578eab1f41f4d5a61a16056c218fe3"}, {"chunk": 5, "start": 753, "path": "inputs/case10-chunk5.npz", "sha256": "439b395b5f5774af1f713529c65e1cb645460e0da307e1e0722db1e9804063a0"}, {"chunk": 6, "start": 785, "path": "inputs/case10-chunk6.npz", "sha256": "e55b95bb698eb6c73427e7a03a69f1f3d75a7d5934263a768b393b0bb29b5283"}, {"chunk": 7, "start": 817, "path": "inputs/case10-chunk7.npz", "sha256": "6667a2717297733a75d73f73cec95474ac0c62209a7cce280ffd8e4ca47f4f7a"}], "input_pack": "inputs/case10-chunk0.npz", "input_image": "inputs/case10-input.png", "gt_videos": {"1": "inputs/case10-gt-1.mp4", "3": "inputs/case10-gt-3.mp4", "8": "inputs/case10-gt-8.mp4"}, "history_images": ["inputs/case10-history-0.png", "inputs/case10-history-1.png", "inputs/case10-history-2.png", "inputs/case10-history-3.png", "inputs/case10-history-4.png", "inputs/case10-history-5.png", "inputs/case10-history-6.png", "inputs/case10-history-7.png", "inputs/case10-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 10*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Insert and tighten each screw into the nut of the same color.", "official_native_context": "official/texts/3eef46639926d06016897ef78385e2a02d56d6a58753088cb1d40dba69a5c882.npy"}, {"id": "case11", "pair_id": "pair05", "split": "train", "start": 674, "max_chunks": 8, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 600, "task": "Insert and tighten each screw into the nut of the same color.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000600", "start_seconds": 26.96, "starts": [674, 706, 738, 770, 802, 834, 866, 898], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again."], "file": "case11.npz", "horizons": [1, 3, 8], "history_indices": [0, 499, 524, 549, 574, 599, 624, 649, 674], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-028.mp4", "start": 157.48000000000002, "end": 217.88000000000002, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-016.mp4", "start": 524.84, "end": 585.24, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-017.mp4", "start": 210.16000000000003, "end": 270.56, "fps": 25}}, "source_npz_sha256": "9f5b2ff1de999dce5e8bd1642b824fc55c9638d0aa98e57471f1f119d16cf0c5", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nSubtask context (ordered by onset): Tighten the red screw into the white nut.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.\nAtomic instruction: right arm: Grasp the red screw again.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Tighten the red screw into the white nut.\nAtomic instruction: right arm: Grasp the red screw again.", "l3": "Robot: robodojo_sim_arx_x5. Task: Insert and tighten each screw into the nut of the same color.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the red screw again."}, "input_packs": [{"chunk": 0, "start": 674, "path": "inputs/case11-chunk0.npz", "sha256": "ccbf9b5f576079eaa0c61b64ed503d1e2680d5894abc0f4dab2e842f75adefe2"}, {"chunk": 1, "start": 706, "path": "inputs/case11-chunk1.npz", "sha256": "d60c3f6be6c7f21edf5dced799c6f5023d84fb4fd53f2fee1b0004f315a9a249"}, {"chunk": 2, "start": 738, "path": "inputs/case11-chunk2.npz", "sha256": "eb24d570345e329a82062990f9df2303a6cdec1d59e5c38bef63026b8a92ebbf"}, {"chunk": 3, "start": 770, "path": "inputs/case11-chunk3.npz", "sha256": "000c87ce765a94e67a9a821862f7bacbd9d58318c7d7155e216abeab02ee8ed6"}, {"chunk": 4, "start": 802, "path": "inputs/case11-chunk4.npz", "sha256": "b7c58fd59c710434acb91a4fad855bd7627723db7b8d7c08f256dcc77336d7d7"}, {"chunk": 5, "start": 834, "path": "inputs/case11-chunk5.npz", "sha256": "5ec011ad1667fc23a6ea640456a0648f61b7d0f6a7bf34362a7e859d789567d2"}, {"chunk": 6, "start": 866, "path": "inputs/case11-chunk6.npz", "sha256": "a974f2d8b9b345c1c8525a07d074938a164d9a8d71ab29647240171b3404c349"}, {"chunk": 7, "start": 898, "path": "inputs/case11-chunk7.npz", "sha256": "b290e563b57a65564fe8b53540c10c5ad8ab7fbc1517b1f6540c5f9e0777c6a3"}], "input_pack": "inputs/case11-chunk0.npz", "input_image": "inputs/case11-input.png", "gt_videos": {"1": "inputs/case11-gt-1.mp4", "3": "inputs/case11-gt-3.mp4", "8": "inputs/case11-gt-8.mp4"}, "history_images": ["inputs/case11-history-0.png", "inputs/case11-history-1.png", "inputs/case11-history-2.png", "inputs/case11-history-3.png", "inputs/case11-history-4.png", "inputs/case11-history-5.png", "inputs/case11-history-6.png", "inputs/case11-history-7.png", "inputs/case11-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 11*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Insert and tighten each screw into the nut of the same color.", "official_native_context": "official/texts/3eef46639926d06016897ef78385e2a02d56d6a58753088cb1d40dba69a5c882.npy"}, {"id": "case12", "pair_id": "pair06", "split": "val", "start": 276, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 707, "task": "Place the four eggs from the basket into the egg holder, then close the lid.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000707", "start_seconds": 11.04, "starts": [276], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nSubtask context (ordered by onset): Place the fourth egg from the basket into the egg holder.\nAtomic instruction: left arm: Grasp the last egg from the basket."], "file": "case12.npz", "horizons": [1], "history_indices": [0, 101, 126, 151, 176, 201, 226, 251, 276], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-039.mp4", "start": 54.519999999999996, "end": 70.8, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-023.mp4", "start": 279.71999999999997, "end": 296.0, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-024.mp4", "start": 54.519999999999996, "end": 70.8, "fps": 25}}, "source_npz_sha256": "b1ca0767a103ed7840f7e3252f33752015427580bdc61b17d7344fa837aa6619", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nSubtask context (ordered by onset): Place the fourth egg from the basket into the egg holder.\nAtomic instruction: left arm: Grasp the last egg from the basket.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nSubtask context (ordered by onset): Place the fourth egg from the basket into the egg holder.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nAtomic instruction: left arm: Grasp the last egg from the basket.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the fourth egg from the basket into the egg holder.\nAtomic instruction: left arm: Grasp the last egg from the basket.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the last egg from the basket."}, "input_packs": [{"chunk": 0, "start": 276, "path": "inputs/case12-chunk0.npz", "sha256": "8e075014b93da5303b12580a704b678a9fb9b23c8635bae2df3c78e2cd285565"}], "input_pack": "inputs/case12-chunk0.npz", "input_image": "inputs/case12-input.png", "gt_videos": {"1": "inputs/case12-gt-1.mp4"}, "history_images": ["inputs/case12-history-0.png", "inputs/case12-history-1.png", "inputs/case12-history-2.png", "inputs/case12-history-3.png", "inputs/case12-history-4.png", "inputs/case12-history-5.png", "inputs/case12-history-6.png", "inputs/case12-history-7.png", "inputs/case12-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 12*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place the four eggs from the basket into the egg holder, then close the lid.", "official_native_context": "official/texts/998c60052b154604aea63f16019a0fdccd1681cdc6685c86f8be3f2421558229.npy"}, {"id": "case13", "pair_id": "pair06", "split": "train", "start": 295, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 700, "task": "Place the four eggs from the basket into the egg holder, then close the lid.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000700", "start_seconds": 11.8, "starts": [295], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nSubtask context (ordered by onset): Place the third egg from the basket into the egg holder.\nAtomic instruction: right arm: Place the egg into the next empty slot of the egg holder."], "file": "case13.npz", "horizons": [1], "history_indices": [0, 120, 145, 170, 195, 220, 245, 270, 295], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-038.mp4", "start": 497.68, "end": 517.2, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-023.mp4", "start": 162.72, "end": 182.24, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-023.mp4", "start": 769.44, "end": 788.96, "fps": 25}}, "source_npz_sha256": "3e3262c3affbdaa5e6a7e47abb62fdc96830473ff51b004169dc84311179584e", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nSubtask context (ordered by onset): Place the third egg from the basket into the egg holder.\nAtomic instruction: right arm: Place the egg into the next empty slot of the egg holder.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nSubtask context (ordered by onset): Place the third egg from the basket into the egg holder.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.\nAtomic instruction: right arm: Place the egg into the next empty slot of the egg holder.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the third egg from the basket into the egg holder.\nAtomic instruction: right arm: Place the egg into the next empty slot of the egg holder.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place the four eggs from the basket into the egg holder, then close the lid.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the egg into the next empty slot of the egg holder."}, "input_packs": [{"chunk": 0, "start": 295, "path": "inputs/case13-chunk0.npz", "sha256": "ae726a6ac632e19e0d0abc9a405a3fdec6a9b41c64087a61e3f37d7dbe0aa780"}], "input_pack": "inputs/case13-chunk0.npz", "input_image": "inputs/case13-input.png", "gt_videos": {"1": "inputs/case13-gt-1.mp4"}, "history_images": ["inputs/case13-history-0.png", "inputs/case13-history-1.png", "inputs/case13-history-2.png", "inputs/case13-history-3.png", "inputs/case13-history-4.png", "inputs/case13-history-5.png", "inputs/case13-history-6.png", "inputs/case13-history-7.png", "inputs/case13-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 13*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place the four eggs from the basket into the egg holder, then close the lid.", "official_native_context": "official/texts/998c60052b154604aea63f16019a0fdccd1681cdc6685c86f8be3f2421558229.npy"}, {"id": "case14", "pair_id": "pair07", "split": "val", "start": 442, "max_chunks": 8, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 801, "task": "Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000801", "start_seconds": 17.68, "starts": [442, 474, 506, 538, 570, 602, 634, 666], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the second blue marker into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the second blue marker into the pen holder.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the second blue marker into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the second blue marker into the pen holder.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the second blue marker into the pen holder.; Place the white pen into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the white pen into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Grasp the white pen on the table.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the white pen into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Grasp the white pen on the table.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the white pen into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Grasp the white pen on the table.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the white pen into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the white pen into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the white pen into the pen holder."], "file": "case14.npz", "horizons": [1, 3, 8], "history_indices": [0, 267, 292, 317, 342, 367, 392, 417, 442], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-042.mp4", "start": 79.52000000000001, "end": 121.52000000000001, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-026.mp4", "start": 108.03999999999999, "end": 150.04, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-026.mp4", "start": 511.71999999999997, "end": 553.72, "fps": 25}}, "source_npz_sha256": "822c2cb78154fc5abc0e49a060c3c22276cf41b83b5898b8b9e66730fba5db9a", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the second blue marker into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the second blue marker into the pen holder.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the second blue marker into the pen holder.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the second blue marker into the pen holder.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the second blue marker into the pen holder.\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the second blue marker into the pen holder.", "l3": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hold the pen holder steady in the air while the right arm places pens inside it. right arm: Place the second blue marker into the pen holder."}, "input_packs": [{"chunk": 0, "start": 442, "path": "inputs/case14-chunk0.npz", "sha256": "f099ea20becd54e746f63ccb4df7507ea1a9a335be6b7315bf053aef81798f2a"}, {"chunk": 1, "start": 474, "path": "inputs/case14-chunk1.npz", "sha256": "aec4127702b93b93941a77cccce5939ad8a9bb000e3c626449b235b76f87660c"}, {"chunk": 2, "start": 506, "path": "inputs/case14-chunk2.npz", "sha256": "dedb33923af06ab60c43f1279d23dee7fe6947342d8dc88666c6e186d428c54a"}, {"chunk": 3, "start": 538, "path": "inputs/case14-chunk3.npz", "sha256": "08f28314d4bf2d82964e0ecbd9b48c500f33df398b925dea4f2c55341f69e9a2"}, {"chunk": 4, "start": 570, "path": "inputs/case14-chunk4.npz", "sha256": "c28f5d42449277f99ea54426f5e02dcb42809b097a7b7d56f15df252a711ab5a"}, {"chunk": 5, "start": 602, "path": "inputs/case14-chunk5.npz", "sha256": "9cee81cd18952ed7379ae39b0e0f115d877d08d57d56eb8c0b1fd70b0a8a99a7"}, {"chunk": 6, "start": 634, "path": "inputs/case14-chunk6.npz", "sha256": "61e9fbfe48f0751f12ffa39d1709060d192ef129653930784be161167e9b09c0"}, {"chunk": 7, "start": 666, "path": "inputs/case14-chunk7.npz", "sha256": "102ea35843c3c68dc41953baa97485aaee5a801b48fc83f2e77ee68322a63cea"}], "input_pack": "inputs/case14-chunk0.npz", "input_image": "inputs/case14-input.png", "gt_videos": {"1": "inputs/case14-gt-1.mp4", "3": "inputs/case14-gt-3.mp4", "8": "inputs/case14-gt-8.mp4"}, "history_images": ["inputs/case14-history-0.png", "inputs/case14-history-1.png", "inputs/case14-history-2.png", "inputs/case14-history-3.png", "inputs/case14-history-4.png", "inputs/case14-history-5.png", "inputs/case14-history-6.png", "inputs/case14-history-7.png", "inputs/case14-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 14*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.", "official_native_context": "official/texts/40aaa4b612700b480c910cf98885a13a95e0d68f46cd37d5adef195555f60c0b.npy"}, {"id": "case15", "pair_id": "pair07", "split": "train", "start": 381, "max_chunks": 8, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 800, "task": "Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000800", "start_seconds": 15.24, "starts": [381, 413, 445, 477, 509, 541, 573, 605], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air. right arm: Place the purple marker into the pen holder held by the left arm.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air. right arm: Place the purple marker into the pen holder held by the left arm.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.; Place the gold pen into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the gold pen into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air. right arm: Grasp the gold pen on the table.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the gold pen into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air. right arm: Grasp the gold pen on the table.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the gold pen into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air.", "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the gold pen into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air. right arm: Place the gold pen into the pen holder held by the left arm."], "file": "case15.npz", "horizons": [1, 3, 8], "history_indices": [0, 206, 231, 256, 281, 306, 331, 356, 381], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-042.mp4", "start": 42.2, "end": 79.52000000000001, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-026.mp4", "start": 70.72, "end": 108.03999999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-026.mp4", "start": 474.4, "end": 511.71999999999997, "fps": 25}}, "source_npz_sha256": "672e387b1700fa3c3ef6d209ff05691e91f1c98cffa81968f84c08ed34f79985", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.\nAtomic instruction: left arm: Hold the pen holder steady in the air.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the purple marker into the pen holder held by the left arm.\nAtomic instruction: left arm: Hold the pen holder steady in the air.", "l3": "Robot: robodojo_sim_arx_x5. Task: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hold the pen holder steady in the air."}, "input_packs": [{"chunk": 0, "start": 381, "path": "inputs/case15-chunk0.npz", "sha256": "aef283f92c12901ff26009568d62ae0766039c76f31822515fee7909f60d5fe2"}, {"chunk": 1, "start": 413, "path": "inputs/case15-chunk1.npz", "sha256": "948085cd1c33a58f97e6b508a8fb9a250c613da6403fb5ff53a81fff87e936e0"}, {"chunk": 2, "start": 445, "path": "inputs/case15-chunk2.npz", "sha256": "fc698208a8e284d80aeff0a0a134250e5b3747d796df72c836ae771b93300122"}, {"chunk": 3, "start": 477, "path": "inputs/case15-chunk3.npz", "sha256": "2fe7211a3a337cf28d0fd49d01541e7e6ae2905281364db598568dd878d45c97"}, {"chunk": 4, "start": 509, "path": "inputs/case15-chunk4.npz", "sha256": "7dd7d56e18c9f231882d9b5eded0aab922100843781cc4523b32941762935470"}, {"chunk": 5, "start": 541, "path": "inputs/case15-chunk5.npz", "sha256": "1b6b36f7223d10716969819e7cafa7d4168acba26f69b8ee6defc6cf3f83811b"}, {"chunk": 6, "start": 573, "path": "inputs/case15-chunk6.npz", "sha256": "608c52374e6131119874e2b3c57e1b96224a3f1d5d4570abc948ee758f108b76"}, {"chunk": 7, "start": 605, "path": "inputs/case15-chunk7.npz", "sha256": "af42694369c6ff9ee15b7948aea99e74cb01a79f11396c3a51cc463f9f7e1f7c"}], "input_pack": "inputs/case15-chunk0.npz", "input_image": "inputs/case15-input.png", "gt_videos": {"1": "inputs/case15-gt-1.mp4", "3": "inputs/case15-gt-3.mp4", "8": "inputs/case15-gt-8.mp4"}, "history_images": ["inputs/case15-history-0.png", "inputs/case15-history-1.png", "inputs/case15-history-2.png", "inputs/case15-history-3.png", "inputs/case15-history-4.png", "inputs/case15-history-5.png", "inputs/case15-history-6.png", "inputs/case15-history-7.png", "inputs/case15-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 15*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down.", "official_native_context": "official/texts/40aaa4b612700b480c910cf98885a13a95e0d68f46cd37d5adef195555f60c0b.npy"}, {"id": "case16", "pair_id": "pair08", "split": "val", "start": 198, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 917, "task": "Fold the clothes neatly.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000917", "start_seconds": 7.92, "starts": [198], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top."], "file": "case16.npz", "horizons": [1], "history_indices": [0, 23, 48, 73, 98, 123, 148, 173, 198], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-047.mp4", "start": 278.31999999999994, "end": 290.23999999999995, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-029.mp4", "start": 434.52000000000004, "end": 446.44000000000005, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-030.mp4", "start": 48.080000000000005, "end": 60.00000000000001, "fps": 25}}, "source_npz_sha256": "771aa03756f05e6a7d01d40361fc2cab9ab396b40fedefec563d8949ea8699a2", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top.", "l3": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top."}, "input_packs": [{"chunk": 0, "start": 198, "path": "inputs/case16-chunk0.npz", "sha256": "bf83bc3fb1fdcd9af9899c1a81523f67af3787ee462658becdbb24506ed66870"}], "input_pack": "inputs/case16-chunk0.npz", "input_image": "inputs/case16-input.png", "gt_videos": {"1": "inputs/case16-gt-1.mp4"}, "history_images": ["inputs/case16-history-0.png", "inputs/case16-history-1.png", "inputs/case16-history-2.png", "inputs/case16-history-3.png", "inputs/case16-history-4.png", "inputs/case16-history-5.png", "inputs/case16-history-6.png", "inputs/case16-history-7.png", "inputs/case16-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 16*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Fold the clothes neatly.", "official_native_context": "official/texts/ada1599c4ce85e0d89f99b7f8e6e4404bc41b1e024c5f425ae1394cb740ad3b2.npy"}, {"id": "case17", "pair_id": "pair08", "split": "train", "start": 199, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 900, "task": "Fold the clothes neatly.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_000900", "start_seconds": 7.96, "starts": [199], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top."], "file": "case17.npz", "horizons": [1], "history_indices": [0, 24, 49, 74, 99, 124, 149, 174, 199], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-047.mp4", "start": 72.75999999999999, "end": 84.75999999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-029.mp4", "start": 228.95999999999998, "end": 240.95999999999998, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-029.mp4", "start": 703.4799999999999, "end": 715.4799999999999, "fps": 25}}, "source_npz_sha256": "4176f5bfad1047a46851ba5d0e31cad0aa65d285dd99f0a2b83f472170bf827f", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Fold the bottom edge of the long-sleeve top upwards towards the collar.\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top.", "l3": "Robot: robodojo_sim_arx_x5. Task: Fold the clothes neatly.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the bottom edge of the long-sleeve top. right arm: Grasp the bottom edge of the long-sleeve top."}, "input_packs": [{"chunk": 0, "start": 199, "path": "inputs/case17-chunk0.npz", "sha256": "2699751b77ed48b8d6076364ce06a578ac923d17148fa91e10fb4cd7f5fa9ba6"}], "input_pack": "inputs/case17-chunk0.npz", "input_image": "inputs/case17-input.png", "gt_videos": {"1": "inputs/case17-gt-1.mp4"}, "history_images": ["inputs/case17-history-0.png", "inputs/case17-history-1.png", "inputs/case17-history-2.png", "inputs/case17-history-3.png", "inputs/case17-history-4.png", "inputs/case17-history-5.png", "inputs/case17-history-6.png", "inputs/case17-history-7.png", "inputs/case17-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 17*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Fold the clothes neatly.", "official_native_context": "official/texts/ada1599c4ce85e0d89f99b7f8e6e4404bc41b1e024c5f425ae1394cb740ad3b2.npy"}, {"id": "case18", "pair_id": "pair09", "split": "val", "start": 252, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 1081, "task": "Hang all the mugs on the mug rack.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001081", "start_seconds": 10.08, "starts": [252, 284, 316], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the wooden mug rack.\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack.", "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the wooden mug rack.\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack.", "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the wooden mug rack.\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack."], "file": "case18.npz", "horizons": [1, 3], "history_indices": [0, 77, 102, 127, 152, 177, 202, 227, 252], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-052.mp4", "start": 399.9999999999999, "end": 420.7599999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-033.mp4", "start": 422.52, "end": 443.28, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-034.mp4", "start": 58.92, "end": 79.68, "fps": 25}}, "source_npz_sha256": "87c0f9f8f6ad8e1f1e0a3538f2f24ef92e975cc99a2e31c578f3abc58174327f", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the wooden mug rack.\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the wooden mug rack.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Hang the purple mug on a hook of the wooden mug rack.\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack.", "l3": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the purple mug onto a hook of the wooden mug rack."}, "input_packs": [{"chunk": 0, "start": 252, "path": "inputs/case18-chunk0.npz", "sha256": "735d7259b45b3887fa4e3c80148b7a793810aca68e4830ac21b7e950b7d1b2a3"}, {"chunk": 1, "start": 284, "path": "inputs/case18-chunk1.npz", "sha256": "b07aea5768309bb39d2ea644c561426231c179e7dc030baf9d545dbc30ec085a"}, {"chunk": 2, "start": 316, "path": "inputs/case18-chunk2.npz", "sha256": "61a1a3667d93f19ebe1d79ba17247207630c40c6b49ae1d73df350600c3c84e1"}], "input_pack": "inputs/case18-chunk0.npz", "input_image": "inputs/case18-input.png", "gt_videos": {"1": "inputs/case18-gt-1.mp4", "3": "inputs/case18-gt-3.mp4"}, "history_images": ["inputs/case18-history-0.png", "inputs/case18-history-1.png", "inputs/case18-history-2.png", "inputs/case18-history-3.png", "inputs/case18-history-4.png", "inputs/case18-history-5.png", "inputs/case18-history-6.png", "inputs/case18-history-7.png", "inputs/case18-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 18*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Hang all the mugs on the mug rack.", "official_native_context": "official/texts/0d7bdbfb65bf4aed4938916dd1b9a4af9905e244aa677619d05f72489f9c8dae.npy"}, {"id": "case19", "pair_id": "pair09", "split": "train", "start": 407, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 1000, "task": "Hang all the mugs on the mug rack.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001000", "start_seconds": 16.28, "starts": [407, 439, 471], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the mug rack.\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack.", "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the mug rack.\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack.", "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the mug rack.\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack."], "file": "case19.npz", "horizons": [1, 3], "history_indices": [0, 232, 257, 282, 307, 332, 357, 382, 407], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-049.mp4", "start": 231.51999999999998, "end": 251.99999999999997, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-031.mp4", "start": 267.79999999999995, "end": 288.28, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-032.mp4", "start": 0.0, "end": 20.48, "fps": 25}}, "source_npz_sha256": "dea46ce864ad2db549408836628be68ddf5681e5593b4fbcaf2c3a457c5964f2", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the mug rack.\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nSubtask context (ordered by onset): Hang the purple mug on a hook of the mug rack.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Hang the purple mug on a hook of the mug rack.\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack.", "l3": "Robot: robodojo_sim_arx_x5. Task: Hang all the mugs on the mug rack.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hang the purple mug on the hook of the mug rack."}, "input_packs": [{"chunk": 0, "start": 407, "path": "inputs/case19-chunk0.npz", "sha256": "9429229e9da0691ca9bfb59e92729f876d9c14957eeddc4a507c0f5aba005b13"}, {"chunk": 1, "start": 439, "path": "inputs/case19-chunk1.npz", "sha256": "44176e3403378fd9ecd014f5d65481a0355e566005f66c474c7f57f76d8af2e2"}, {"chunk": 2, "start": 471, "path": "inputs/case19-chunk2.npz", "sha256": "df3ee8a4e3504d1daf2827157881826e69b5e0c28a9aff058afb80c41bc77c45"}], "input_pack": "inputs/case19-chunk0.npz", "input_image": "inputs/case19-input.png", "gt_videos": {"1": "inputs/case19-gt-1.mp4", "3": "inputs/case19-gt-3.mp4"}, "history_images": ["inputs/case19-history-0.png", "inputs/case19-history-1.png", "inputs/case19-history-2.png", "inputs/case19-history-3.png", "inputs/case19-history-4.png", "inputs/case19-history-5.png", "inputs/case19-history-6.png", "inputs/case19-history-7.png", "inputs/case19-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 19*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Hang all the mugs on the mug rack.", "official_native_context": "official/texts/0d7bdbfb65bf4aed4938916dd1b9a4af9905e244aa677619d05f72489f9c8dae.npy"}, {"id": "case20", "pair_id": "pair10", "split": "val", "start": 701, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 1164, "task": "Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001164", "start_seconds": 28.04, "starts": [701], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nSubtask context (ordered by onset): Place the doll into the white basket.\nAtomic instruction: left arm: Grasp the doll on the table."], "file": "case20.npz", "horizons": [1], "history_indices": [0, 526, 551, 576, 601, 626, 651, 676, 701], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-060.mp4", "start": 100.32, "end": 143.48, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-037.mp4", "start": 685.6800000000001, "end": 728.84, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-039.mp4", "start": 48.96, "end": 92.12, "fps": 25}}, "source_npz_sha256": "7adb7b0e740610d7b73724e013e37670b8a2874ddd30fc0e1e9e3a6d7d21ef43", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nSubtask context (ordered by onset): Place the doll into the white basket.\nAtomic instruction: left arm: Grasp the doll on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nSubtask context (ordered by onset): Place the doll into the white basket.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nAtomic instruction: left arm: Grasp the doll on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the doll into the white basket.\nAtomic instruction: left arm: Grasp the doll on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the doll on the table."}, "input_packs": [{"chunk": 0, "start": 701, "path": "inputs/case20-chunk0.npz", "sha256": "32d3a97038a2c37abe1f42e37d500af883e02c7b774e2becdc815b17e4378148"}], "input_pack": "inputs/case20-chunk0.npz", "input_image": "inputs/case20-input.png", "gt_videos": {"1": "inputs/case20-gt-1.mp4"}, "history_images": ["inputs/case20-history-0.png", "inputs/case20-history-1.png", "inputs/case20-history-2.png", "inputs/case20-history-3.png", "inputs/case20-history-4.png", "inputs/case20-history-5.png", "inputs/case20-history-6.png", "inputs/case20-history-7.png", "inputs/case20-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 20*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.", "official_native_context": "official/texts/e4d8fdede875781adcc11014a9e18295f5584f11c7295e798a9cd66d8d6c9dcf.npy"}, {"id": "case21", "pair_id": "pair10", "split": "train", "start": 822, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 1100, "task": "Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001100", "start_seconds": 32.88, "starts": [822], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nSubtask context (ordered by onset): Place the green object into the basket.; Place the green toy car into the white basket on the left.\nAtomic instruction: left arm: Grasp the green toy car on the table."], "file": "case21.npz", "horizons": [1], "history_indices": [0, 647, 672, 697, 722, 747, 772, 797, 822], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-053.mp4", "start": 196.07999999999998, "end": 243.55999999999997, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-033.mp4", "start": 774.12, "end": 821.6, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-034.mp4", "start": 410.52000000000004, "end": 458.00000000000006, "fps": 25}}, "source_npz_sha256": "671e91ce176cf62cf2910251e952fb92f9e7f76f99610c8b737ced94b64254f7", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nSubtask context (ordered by onset): Place the green object into the basket.; Place the green toy car into the white basket on the left.\nAtomic instruction: left arm: Grasp the green toy car on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nSubtask context (ordered by onset): Place the green object into the basket.; Place the green toy car into the white basket on the left.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.\nAtomic instruction: left arm: Grasp the green toy car on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the green object into the basket.; Place the green toy car into the white basket on the left.\nAtomic instruction: left arm: Grasp the green toy car on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the green toy car on the table."}, "input_packs": [{"chunk": 0, "start": 822, "path": "inputs/case21-chunk0.npz", "sha256": "47e9006c636d96f053d510ea4d687b43bdb804d9ebfd04f3d423b885b63cdda9"}], "input_pack": "inputs/case21-chunk0.npz", "input_image": "inputs/case21-input.png", "gt_videos": {"1": "inputs/case21-gt-1.mp4"}, "history_images": ["inputs/case21-history-0.png", "inputs/case21-history-1.png", "inputs/case21-history-2.png", "inputs/case21-history-3.png", "inputs/case21-history-4.png", "inputs/case21-history-5.png", "inputs/case21-history-6.png", "inputs/case21-history-7.png", "inputs/case21-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 21*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order.", "official_native_context": "official/texts/e4d8fdede875781adcc11014a9e18295f5584f11c7295e798a9cd66d8d6c9dcf.npy"}, {"id": "case22", "pair_id": "pair11", "split": "val", "start": 139, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1291, "task": "Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001291", "start_seconds": 5.56, "starts": [139], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nSubtask context (ordered by onset): Insert the key into the keyhole and turn it.\nAtomic instruction: left arm: Hand over the key to the right arm."], "file": "case22.npz", "horizons": [1], "history_indices": [0, 0, 0, 14, 39, 64, 89, 114, 139], "distinct_history_frames": 7, "history_padding": [false, true, true, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-065.mp4", "start": 78.84, "end": 85.88000000000001, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-040.mp4", "start": 648.4400000000003, "end": 655.4800000000002, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-042.mp4", "start": 285.2800000000001, "end": 292.3200000000001, "fps": 25}}, "source_npz_sha256": "5191e5c4a93f7fe2003d4077631c5a3b33a6bae04c0634d8d57b669b71a02905", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nSubtask context (ordered by onset): Insert the key into the keyhole and turn it.\nAtomic instruction: left arm: Hand over the key to the right arm.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nSubtask context (ordered by onset): Insert the key into the keyhole and turn it.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nAtomic instruction: left arm: Hand over the key to the right arm.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Insert the key into the keyhole and turn it.\nAtomic instruction: left arm: Hand over the key to the right arm.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hand over the key to the right arm."}, "input_packs": [{"chunk": 0, "start": 139, "path": "inputs/case22-chunk0.npz", "sha256": "5daab1343dce8e7c2861d981d0b7ebca6694fe7590fd619687069f4bb00926d4"}], "input_pack": "inputs/case22-chunk0.npz", "input_image": "inputs/case22-input.png", "gt_videos": {"1": "inputs/case22-gt-1.mp4"}, "history_images": ["inputs/case22-history-0.png", "inputs/case22-history-1.png", "inputs/case22-history-2.png", "inputs/case22-history-3.png", "inputs/case22-history-4.png", "inputs/case22-history-5.png", "inputs/case22-history-6.png", "inputs/case22-history-7.png", "inputs/case22-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 22*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "official_native_context": "official/texts/b5a8e422bd7715906ae0a8a4eccac354e19480d6003a0118a73efaf658b9b174.npy"}, {"id": "case23", "pair_id": "pair11", "split": "train", "start": 140, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f000", "episode": 1200, "task": "Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001200", "start_seconds": 5.6, "starts": [140], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nSubtask context (ordered by onset): Hand the key back from the left gripper to the right gripper.\nAtomic instruction: left arm: Hand over the key to the right arm."], "file": "case23.npz", "horizons": [1], "history_indices": [0, 0, 0, 15, 40, 65, 90, 115, 140], "distinct_history_frames": 7, "history_padding": [false, true, true, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-064.mp4", "start": 48.24, "end": 55.24, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-040.mp4", "start": 0.0, "end": 7.0, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-041.mp4", "start": 428.28000000000003, "end": 435.28000000000003, "fps": 25}}, "source_npz_sha256": "34bb0ab8f382c0c28af3b9fdb0a04c18542fcde8c802bb8e59f3f3c9495127a1", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nSubtask context (ordered by onset): Hand the key back from the left gripper to the right gripper.\nAtomic instruction: left arm: Hand over the key to the right arm.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nSubtask context (ordered by onset): Hand the key back from the left gripper to the right gripper.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.\nAtomic instruction: left arm: Hand over the key to the right arm.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Hand the key back from the left gripper to the right gripper.\nAtomic instruction: left arm: Hand over the key to the right arm.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hand over the key to the right arm."}, "input_packs": [{"chunk": 0, "start": 140, "path": "inputs/case23-chunk0.npz", "sha256": "ce94a6881e562b2666f5dd64befe3188e3d6df0f98bfc64252050ff1577fe6e1"}], "input_pack": "inputs/case23-chunk0.npz", "input_image": "inputs/case23-input.png", "gt_videos": {"1": "inputs/case23-gt-1.mp4"}, "history_images": ["inputs/case23-history-0.png", "inputs/case23-history-1.png", "inputs/case23-history-2.png", "inputs/case23-history-3.png", "inputs/case23-history-4.png", "inputs/case23-history-5.png", "inputs/case23-history-6.png", "inputs/case23-history-7.png", "inputs/case23-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 23*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it.", "official_native_context": "official/texts/b5a8e422bd7715906ae0a8a4eccac354e19480d6003a0118a73efaf658b9b174.npy"}, {"id": "case24", "pair_id": "pair12", "split": "val", "start": 238, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1350, "task": "Insert the three tubes into the rack one by one.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001350", "start_seconds": 9.52, "starts": [238], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nSubtask context (ordered by onset): Insert the third test tube into the rack using the left arm.\nAtomic instruction: left arm: Insert the held test tube into the next available slot in the tube rack."], "file": "case24.npz", "horizons": [1], "history_indices": [0, 63, 88, 113, 138, 163, 188, 213, 238], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-066.mp4", "start": 206.4, "end": 219.20000000000002, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-041.mp4", "start": 386.8399999999999, "end": 399.63999999999993, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-043.mp4", "start": 90.75999999999999, "end": 103.55999999999999, "fps": 25}}, "source_npz_sha256": "af1935623ff616f2ebfecb8c5f9295ecc4076bd8eb60eb5a39d26768ea0c1e39", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nSubtask context (ordered by onset): Insert the third test tube into the rack using the left arm.\nAtomic instruction: left arm: Insert the held test tube into the next available slot in the tube rack.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nSubtask context (ordered by onset): Insert the third test tube into the rack using the left arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nAtomic instruction: left arm: Insert the held test tube into the next available slot in the tube rack.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Insert the third test tube into the rack using the left arm.\nAtomic instruction: left arm: Insert the held test tube into the next available slot in the tube rack.", "l3": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Insert the held test tube into the next available slot in the tube rack."}, "input_packs": [{"chunk": 0, "start": 238, "path": "inputs/case24-chunk0.npz", "sha256": "cd0be19b4ad4ad700e40a3588a64fefb9601a2d28a71c5841d4bf0ea9f7dc4ae"}], "input_pack": "inputs/case24-chunk0.npz", "input_image": "inputs/case24-input.png", "gt_videos": {"1": "inputs/case24-gt-1.mp4"}, "history_images": ["inputs/case24-history-0.png", "inputs/case24-history-1.png", "inputs/case24-history-2.png", "inputs/case24-history-3.png", "inputs/case24-history-4.png", "inputs/case24-history-5.png", "inputs/case24-history-6.png", "inputs/case24-history-7.png", "inputs/case24-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 24*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Insert the three tubes into the rack one by one.", "official_native_context": "official/texts/89836fc02c0423bc732d7fb93d4ce940594d3499b4e787e9889242712a4a9b78.npy"}, {"id": "case25", "pair_id": "pair12", "split": "train", "start": 237, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1300, "task": "Insert the three tubes into the rack one by one.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001300", "start_seconds": 9.48, "starts": [237], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nSubtask context (ordered by onset): Insert the third test tube into the blue tube rack.\nAtomic instruction: left arm: Place the test tube into the blue tube rack."], "file": "case25.npz", "horizons": [1], "history_indices": [0, 62, 87, 112, 137, 162, 187, 212, 237], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-065.mp4", "start": 143.64000000000004, "end": 156.40000000000003, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-040.mp4", "start": 713.2400000000002, "end": 726.0000000000002, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-042.mp4", "start": 350.0800000000001, "end": 362.8400000000001, "fps": 25}}, "source_npz_sha256": "03783ffd2ed62dd15ed123db6bbbd6301880ce658732411fcfb25d2ada32495d", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nSubtask context (ordered by onset): Insert the third test tube into the blue tube rack.\nAtomic instruction: left arm: Place the test tube into the blue tube rack.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nSubtask context (ordered by onset): Insert the third test tube into the blue tube rack.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.\nAtomic instruction: left arm: Place the test tube into the blue tube rack.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Insert the third test tube into the blue tube rack.\nAtomic instruction: left arm: Place the test tube into the blue tube rack.", "l3": "Robot: robodojo_sim_arx_x5. Task: Insert the three tubes into the rack one by one.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the test tube into the blue tube rack."}, "input_packs": [{"chunk": 0, "start": 237, "path": "inputs/case25-chunk0.npz", "sha256": "a52d69e12e8d007bea70b30bdb81ee011cee7189083acacf4cb14879423570cf"}], "input_pack": "inputs/case25-chunk0.npz", "input_image": "inputs/case25-input.png", "gt_videos": {"1": "inputs/case25-gt-1.mp4"}, "history_images": ["inputs/case25-history-0.png", "inputs/case25-history-1.png", "inputs/case25-history-2.png", "inputs/case25-history-3.png", "inputs/case25-history-4.png", "inputs/case25-history-5.png", "inputs/case25-history-6.png", "inputs/case25-history-7.png", "inputs/case25-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 25*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Insert the three tubes into the rack one by one.", "official_native_context": "official/texts/89836fc02c0423bc732d7fb93d4ce940594d3499b4e787e9889242712a4a9b78.npy"}, {"id": "case26", "pair_id": "pair13", "split": "val", "start": 315, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1460, "task": "Wait for the opponent to discard a tile, then declare a kong with the matching tiles.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001460", "start_seconds": 12.6, "starts": [315], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nSubtask context (ordered by onset): Transport the held mahjong tiles to the center of the table and place them down.; Grasp the discarded mahjong tile with the right arm and place it next to the other tiles to declare a kong.\nAtomic instruction: right arm: Move the held mahjong tile to the right and place it down on the table."], "file": "case26.npz", "horizons": [1], "history_indices": [0, 140, 165, 190, 215, 240, 265, 290, 315], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-069.mp4", "start": 104.03999999999999, "end": 118.72, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-043.mp4", "start": 312.0, "end": 326.68, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-045.mp4", "start": 74.56, "end": 89.24000000000001, "fps": 25}}, "source_npz_sha256": "67cab54fb9620b4ce3d414fce64cdcf9d9665f7c37a6edb9e807010986de8cc8", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nSubtask context (ordered by onset): Transport the held mahjong tiles to the center of the table and place them down.; Grasp the discarded mahjong tile with the right arm and place it next to the other tiles to declare a kong.\nAtomic instruction: right arm: Move the held mahjong tile to the right and place it down on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nSubtask context (ordered by onset): Transport the held mahjong tiles to the center of the table and place them down.; Grasp the discarded mahjong tile with the right arm and place it next to the other tiles to declare a kong.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nAtomic instruction: right arm: Move the held mahjong tile to the right and place it down on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Transport the held mahjong tiles to the center of the table and place them down.; Grasp the discarded mahjong tile with the right arm and place it next to the other tiles to declare a kong.\nAtomic instruction: right arm: Move the held mahjong tile to the right and place it down on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Move the held mahjong tile to the right and place it down on the table."}, "input_packs": [{"chunk": 0, "start": 315, "path": "inputs/case26-chunk0.npz", "sha256": "26140085864057c0296b5ba5847291073ba764d341f44a044c4f0276f41f0629"}], "input_pack": "inputs/case26-chunk0.npz", "input_image": "inputs/case26-input.png", "gt_videos": {"1": "inputs/case26-gt-1.mp4"}, "history_images": ["inputs/case26-history-0.png", "inputs/case26-history-1.png", "inputs/case26-history-2.png", "inputs/case26-history-3.png", "inputs/case26-history-4.png", "inputs/case26-history-5.png", "inputs/case26-history-6.png", "inputs/case26-history-7.png", "inputs/case26-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 26*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.", "official_native_context": "official/texts/f674cda4b29a45595029faed4656c3e814112d79252b6ccf06a5e1eef636700c.npy"}, {"id": "case27", "pair_id": "pair13", "split": "train", "start": 297, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1400, "task": "Wait for the opponent to discard a tile, then declare a kong with the matching tiles.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001400", "start_seconds": 11.88, "starts": [297], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nSubtask context (ordered by onset): Grasp another mahjong tile from the center and place it onto the table surface.\nAtomic instruction: left arm: Place the mahjong tile onto the table surface. right arm: Grasp another mahjong tile from the center area."], "file": "case27.npz", "horizons": [1], "history_indices": [0, 122, 147, 172, 197, 222, 247, 272, 297], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-067.mp4", "start": 282.64, "end": 297.91999999999996, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-042.mp4", "start": 192.87999999999997, "end": 208.15999999999997, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-043.mp4", "start": 732.56, "end": 747.8399999999999, "fps": 25}}, "source_npz_sha256": "9b9f0082a6c37ef2c567daa5c914930cb4af51447ebf1da20410baa840d088fe", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nSubtask context (ordered by onset): Grasp another mahjong tile from the center and place it onto the table surface.\nAtomic instruction: left arm: Place the mahjong tile onto the table surface. right arm: Grasp another mahjong tile from the center area.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nSubtask context (ordered by onset): Grasp another mahjong tile from the center and place it onto the table surface.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.\nAtomic instruction: left arm: Place the mahjong tile onto the table surface. right arm: Grasp another mahjong tile from the center area.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Grasp another mahjong tile from the center and place it onto the table surface.\nAtomic instruction: left arm: Place the mahjong tile onto the table surface. right arm: Grasp another mahjong tile from the center area.", "l3": "Robot: robodojo_sim_arx_x5. Task: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the mahjong tile onto the table surface. right arm: Grasp another mahjong tile from the center area."}, "input_packs": [{"chunk": 0, "start": 297, "path": "inputs/case27-chunk0.npz", "sha256": "128db4cd42bee6eb0f9b15ef7ebf8e8231f1a6d364d3883e02581d5befaecf14"}], "input_pack": "inputs/case27-chunk0.npz", "input_image": "inputs/case27-input.png", "gt_videos": {"1": "inputs/case27-gt-1.mp4"}, "history_images": ["inputs/case27-history-0.png", "inputs/case27-history-1.png", "inputs/case27-history-2.png", "inputs/case27-history-3.png", "inputs/case27-history-4.png", "inputs/case27-history-5.png", "inputs/case27-history-6.png", "inputs/case27-history-7.png", "inputs/case27-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 27*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Wait for the opponent to discard a tile, then declare a kong with the matching tiles.", "official_native_context": "official/texts/f674cda4b29a45595029faed4656c3e814112d79252b6ccf06a5e1eef636700c.npy"}, {"id": "case28", "pair_id": "pair14", "split": "val", "start": 209, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1597, "task": "Pick up two slices of bread, place them into the toaster, and press the lever down.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001597", "start_seconds": 8.36, "starts": [209, 241, 273], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.; Pick up the second slice of bread and hand it to the right arm.\nAtomic instruction: right arm: Place the bread slice into the toaster slot.", "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the second slice of bread and hand it to the right arm.\nAtomic instruction: right arm: Place the bread slice into the toaster slot.", "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the second slice of bread and hand it to the right arm.\nAtomic instruction: right arm: Place the bread slice into the toaster slot."], "file": "case28.npz", "horizons": [1, 3], "history_indices": [0, 34, 59, 84, 109, 134, 159, 184, 209], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-075.mp4", "start": 54.0, "end": 77.44, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-047.mp4", "start": 602.1999999999999, "end": 625.64, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-049.mp4", "start": 456.12, "end": 479.56, "fps": 25}}, "source_npz_sha256": "c95888b9f1cc77bd889152152f115ab76c7e1223512c65fc4d40679394d06f6b", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.; Pick up the second slice of bread and hand it to the right arm.\nAtomic instruction: right arm: Place the bread slice into the toaster slot.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.; Pick up the second slice of bread and hand it to the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nAtomic instruction: right arm: Place the bread slice into the toaster slot.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.; Pick up the second slice of bread and hand it to the right arm.\nAtomic instruction: right arm: Place the bread slice into the toaster slot.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the bread slice into the toaster slot."}, "input_packs": [{"chunk": 0, "start": 209, "path": "inputs/case28-chunk0.npz", "sha256": "ad143cbcac3c01352a1e9d227d35e16c1d16e0f33888c61a2f0793d68ef17983"}, {"chunk": 1, "start": 241, "path": "inputs/case28-chunk1.npz", "sha256": "0482ec0db23006e0b109f26b9e33b64428598ab06d4080025fbaa1c5b7ec750f"}, {"chunk": 2, "start": 273, "path": "inputs/case28-chunk2.npz", "sha256": "b246561ad609dae1b907cd26fc39a8d9193331ab1209186e198d7a0908b613f8"}], "input_pack": "inputs/case28-chunk0.npz", "input_image": "inputs/case28-input.png", "gt_videos": {"1": "inputs/case28-gt-1.mp4", "3": "inputs/case28-gt-3.mp4"}, "history_images": ["inputs/case28-history-0.png", "inputs/case28-history-1.png", "inputs/case28-history-2.png", "inputs/case28-history-3.png", "inputs/case28-history-4.png", "inputs/case28-history-5.png", "inputs/case28-history-6.png", "inputs/case28-history-7.png", "inputs/case28-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 28*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up two slices of bread, place them into the toaster, and press the lever down.", "official_native_context": "official/texts/bcde67d1c1f1c00faa15fb3f1478d8a9083c11044b664e8393148dad9fd63d4d.npy"}, {"id": "case29", "pair_id": "pair14", "split": "train", "start": 187, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1500, "task": "Pick up two slices of bread, place them into the toaster, and press the lever down.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001500", "start_seconds": 7.48, "starts": [187, 219, 251], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.\nAtomic instruction: left arm: Place the bread slice into the toaster slot.", "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.\nAtomic instruction: left arm: Place the bread slice into the toaster slot.", "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.\nAtomic instruction: left arm: Place the bread slice into the toaster slot."], "file": "case29.npz", "horizons": [1, 3], "history_indices": [0, 12, 37, 62, 87, 112, 137, 162, 187], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-070.mp4", "start": 178.32000000000002, "end": 201.64000000000001, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-044.mp4", "start": 148.76, "end": 172.07999999999998, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-045.mp4", "start": 667.72, "end": 691.0400000000001, "fps": 25}}, "source_npz_sha256": "dc7cd666655a5e7cfaaf941c1d99b08b85ec3e42c30fe83ce0c8f8f897bfeaa3", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.\nAtomic instruction: left arm: Place the bread slice into the toaster slot.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.\nAtomic instruction: left arm: Place the bread slice into the toaster slot.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Pick up the first slice of bread and place it into the toaster.\nAtomic instruction: left arm: Place the bread slice into the toaster slot.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up two slices of bread, place them into the toaster, and press the lever down.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the bread slice into the toaster slot."}, "input_packs": [{"chunk": 0, "start": 187, "path": "inputs/case29-chunk0.npz", "sha256": "ddc3d1eae06d7834d9d21979395c6667a0b65f9d7fc8941b79282238136883de"}, {"chunk": 1, "start": 219, "path": "inputs/case29-chunk1.npz", "sha256": "aaeda09b01d18ed59ffe4fbd83fcbd8f3ec7f36a86b24d08636ae9686c4b2868"}, {"chunk": 2, "start": 251, "path": "inputs/case29-chunk2.npz", "sha256": "6d5b1eb1af66eb8ba79c8c5e73872966ff8885b84a9c61e4dc4852435c764a7b"}], "input_pack": "inputs/case29-chunk0.npz", "input_image": "inputs/case29-input.png", "gt_videos": {"1": "inputs/case29-gt-1.mp4", "3": "inputs/case29-gt-3.mp4"}, "history_images": ["inputs/case29-history-0.png", "inputs/case29-history-1.png", "inputs/case29-history-2.png", "inputs/case29-history-3.png", "inputs/case29-history-4.png", "inputs/case29-history-5.png", "inputs/case29-history-6.png", "inputs/case29-history-7.png", "inputs/case29-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 29*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up two slices of bread, place them into the toaster, and press the lever down.", "official_native_context": "official/texts/bcde67d1c1f1c00faa15fb3f1478d8a9083c11044b664e8393148dad9fd63d4d.npy"}, {"id": "case30", "pair_id": "pair15", "split": "val", "start": 443, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1672, "task": "Remember the first object on the conveyor, then pick the matching object when it appears again.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001672", "start_seconds": 17.72, "starts": [443], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nSubtask context (ordered by onset): Pick the matching bottle from the conveyor.\nAtomic instruction: right arm: Grasp the bottle on the conveyor belt."], "file": "case30.npz", "horizons": [1], "history_indices": [0, 268, 293, 318, 343, 368, 393, 418, 443], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-077.mp4", "start": 261.2, "end": 281.59999999999997, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-049.mp4", "start": 199.12, "end": 219.52, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-050.mp4", "start": 949.96, "end": 970.36, "fps": 25}}, "source_npz_sha256": "75bf5d758ea631f7e6179f5dfe416d02ec8a0cb3099d7de49cba4921e8c79331", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nSubtask context (ordered by onset): Pick the matching bottle from the conveyor.\nAtomic instruction: right arm: Grasp the bottle on the conveyor belt.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nSubtask context (ordered by onset): Pick the matching bottle from the conveyor.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nAtomic instruction: right arm: Grasp the bottle on the conveyor belt.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Pick the matching bottle from the conveyor.\nAtomic instruction: right arm: Grasp the bottle on the conveyor belt.", "l3": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the bottle on the conveyor belt."}, "input_packs": [{"chunk": 0, "start": 443, "path": "inputs/case30-chunk0.npz", "sha256": "8a8ca54e6315d87943f9061e344a14be9d9fe45e658a430cebe440d0ea1c885f"}], "input_pack": "inputs/case30-chunk0.npz", "input_image": "inputs/case30-input.png", "gt_videos": {"1": "inputs/case30-gt-1.mp4"}, "history_images": ["inputs/case30-history-0.png", "inputs/case30-history-1.png", "inputs/case30-history-2.png", "inputs/case30-history-3.png", "inputs/case30-history-4.png", "inputs/case30-history-5.png", "inputs/case30-history-6.png", "inputs/case30-history-7.png", "inputs/case30-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 30*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Remember the first object on the conveyor, then pick the matching object when it appears again.", "official_native_context": "official/texts/cf0c42b0977a9f8ce83284566659e72622a220bb8246bfa1eab2f050a44570fa.npy"}, {"id": "case31", "pair_id": "pair15", "split": "train", "start": 413, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1600, "task": "Remember the first object on the conveyor, then pick the matching object when it appears again.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001600", "start_seconds": 16.52, "starts": [413], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nSubtask context (ordered by onset): Grasp the green bottle on the conveyor belt.\nAtomic instruction: right arm: Grasp the green bottle on the conveyor belt."], "file": "case31.npz", "horizons": [1], "history_indices": [0, 238, 263, 288, 313, 338, 363, 388, 413], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-075.mp4", "start": 136.64000000000001, "end": 155.60000000000002, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-047.mp4", "start": 684.84, "end": 703.8000000000001, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-049.mp4", "start": 538.76, "end": 557.72, "fps": 25}}, "source_npz_sha256": "eed107ebea24e48d179092bd25e5ee6fc15f5a25808560b336fb36089bef935c", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nSubtask context (ordered by onset): Grasp the green bottle on the conveyor belt.\nAtomic instruction: right arm: Grasp the green bottle on the conveyor belt.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nSubtask context (ordered by onset): Grasp the green bottle on the conveyor belt.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.\nAtomic instruction: right arm: Grasp the green bottle on the conveyor belt.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Grasp the green bottle on the conveyor belt.\nAtomic instruction: right arm: Grasp the green bottle on the conveyor belt.", "l3": "Robot: robodojo_sim_arx_x5. Task: Remember the first object on the conveyor, then pick the matching object when it appears again.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the green bottle on the conveyor belt."}, "input_packs": [{"chunk": 0, "start": 413, "path": "inputs/case31-chunk0.npz", "sha256": "4dfee5d32cb1aa62d72fbff06199edcbfcebdceb9ef80ef912ba05167103ac9a"}], "input_pack": "inputs/case31-chunk0.npz", "input_image": "inputs/case31-input.png", "gt_videos": {"1": "inputs/case31-gt-1.mp4"}, "history_images": ["inputs/case31-history-0.png", "inputs/case31-history-1.png", "inputs/case31-history-2.png", "inputs/case31-history-3.png", "inputs/case31-history-4.png", "inputs/case31-history-5.png", "inputs/case31-history-6.png", "inputs/case31-history-7.png", "inputs/case31-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 31*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Remember the first object on the conveyor, then pick the matching object when it appears again.", "official_native_context": "official/texts/cf0c42b0977a9f8ce83284566659e72622a220bb8246bfa1eab2f050a44570fa.npy"}, {"id": "case32", "pair_id": "pair16", "split": "val", "start": 403, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1763, "task": "Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001763", "start_seconds": 16.12, "starts": [403, 435, 467], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Place the mouse on the mouse pad.\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline.", "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Place the mouse on the mouse pad.\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline. right arm: Release the mouse onto the black mouse pad.", "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Place the mouse on the mouse pad.; Push the keyboard into the frame.\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline."], "file": "case32.npz", "horizons": [1, 3], "history_indices": [0, 228, 253, 278, 303, 328, 353, 378, 403], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-081.mp4", "start": 104.04, "end": 125.12, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-051.mp4", "start": 461.48, "end": 482.56, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-053.mp4", "start": 145.72, "end": 166.8, "fps": 25}}, "source_npz_sha256": "39fca419ade651189dc1790731104f1771e7a9ea85f976eddef034e4d7216399", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Place the mouse on the mouse pad.\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Place the mouse on the mouse pad.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the mouse on the mouse pad.\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Push the keyboard forward into the rectangular frame outline."}, "input_packs": [{"chunk": 0, "start": 403, "path": "inputs/case32-chunk0.npz", "sha256": "1ee55acd2dc67897c6795632e6b30843312507805922108d5870f9f0ea5a31d4"}, {"chunk": 1, "start": 435, "path": "inputs/case32-chunk1.npz", "sha256": "7f124a104893450ca2275b86f4ec2ae7b76bf19ee4d3d13a139de979ae477918"}, {"chunk": 2, "start": 467, "path": "inputs/case32-chunk2.npz", "sha256": "dc7694a3b2fb17b6820cee32271f21c987196e9817322a242fff2d91cf8c793b"}], "input_pack": "inputs/case32-chunk0.npz", "input_image": "inputs/case32-input.png", "gt_videos": {"1": "inputs/case32-gt-1.mp4", "3": "inputs/case32-gt-3.mp4"}, "history_images": ["inputs/case32-history-0.png", "inputs/case32-history-1.png", "inputs/case32-history-2.png", "inputs/case32-history-3.png", "inputs/case32-history-4.png", "inputs/case32-history-5.png", "inputs/case32-history-6.png", "inputs/case32-history-7.png", "inputs/case32-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 32*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.", "official_native_context": "official/texts/6c186499609d4e1fe5203e9a4723bd01d7badabc68f41d7c03e9399890728a1d.npy"}, {"id": "case33", "pair_id": "pair16", "split": "train", "start": 643, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1700, "task": "Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001700", "start_seconds": 25.72, "starts": [643, 675, 707], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Push the keyboard into the frame.\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area.", "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Push the keyboard into the frame.\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area.", "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Push the keyboard into the frame.\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area."], "file": "case33.npz", "horizons": [1, 3], "history_indices": [0, 468, 493, 518, 543, 568, 593, 618, 643], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-078.mp4", "start": 236.44, "end": 269.96, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-049.mp4", "start": 718.0, "end": 751.52, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-051.mp4", "start": 391.6, "end": 425.12, "fps": 25}}, "source_npz_sha256": "566af328d90eaa4b8203b85c195778bb2a16a1f17d5ff4ceb28d181642fb0db5", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Push the keyboard into the frame.\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nSubtask context (ordered by onset): Push the keyboard into the frame.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Push the keyboard into the frame.\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Release the keyboard and retract the arm. right arm: Continue pushing the keyboard into the designated frame area."}, "input_packs": [{"chunk": 0, "start": 643, "path": "inputs/case33-chunk0.npz", "sha256": "c561eb367a4e57c77dc44bfcf8a93de0a6d099497cfbc05c848183b4d6e557db"}, {"chunk": 1, "start": 675, "path": "inputs/case33-chunk1.npz", "sha256": "1e47269ed73b19c83c756775eacf9e537267d813489371a30726dadc91802179"}, {"chunk": 2, "start": 707, "path": "inputs/case33-chunk2.npz", "sha256": "f6a5cbf9b98cd8923b377661b19058a38988e82af0786457501571ff4963b5dc"}], "input_pack": "inputs/case33-chunk0.npz", "input_image": "inputs/case33-input.png", "gt_videos": {"1": "inputs/case33-gt-1.mp4", "3": "inputs/case33-gt-3.mp4"}, "history_images": ["inputs/case33-history-0.png", "inputs/case33-history-1.png", "inputs/case33-history-2.png", "inputs/case33-history-3.png", "inputs/case33-history-4.png", "inputs/case33-history-5.png", "inputs/case33-history-6.png", "inputs/case33-history-7.png", "inputs/case33-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 33*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame.", "official_native_context": "official/texts/6c186499609d4e1fe5203e9a4723bd01d7badabc68f41d7c03e9399890728a1d.npy"}, {"id": "case34", "pair_id": "pair17", "split": "val", "start": 445, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1854, "task": "Place all the objects into the box with their front sides facing left.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001854", "start_seconds": 17.8, "starts": [445], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nSubtask context (ordered by onset): Place the electric toothbrush into the box using the right arm.\nAtomic instruction: right arm: Place the electric toothbrush into the box."], "file": "case34.npz", "horizons": [1], "history_indices": [0, 270, 295, 320, 345, 370, 395, 420, 445], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-086.mp4", "start": 169.03999999999996, "end": 198.79999999999995, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-054.mp4", "start": 634.28, "end": 664.04, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-056.mp4", "start": 323.40000000000003, "end": 353.16, "fps": 25}}, "source_npz_sha256": "a8a3f729fc1945af21ed5c118a1f35d681774ec3b4e2cd08c1bbdb6f6dd55be9", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nSubtask context (ordered by onset): Place the electric toothbrush into the box using the right arm.\nAtomic instruction: right arm: Place the electric toothbrush into the box.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nSubtask context (ordered by onset): Place the electric toothbrush into the box using the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nAtomic instruction: right arm: Place the electric toothbrush into the box.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the electric toothbrush into the box using the right arm.\nAtomic instruction: right arm: Place the electric toothbrush into the box.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the electric toothbrush into the box."}, "input_packs": [{"chunk": 0, "start": 445, "path": "inputs/case34-chunk0.npz", "sha256": "0649c9a3ee832487a862e69f44424c1f7a79f73e26e5f0d3177085c8c8fd6e1b"}], "input_pack": "inputs/case34-chunk0.npz", "input_image": "inputs/case34-input.png", "gt_videos": {"1": "inputs/case34-gt-1.mp4"}, "history_images": ["inputs/case34-history-0.png", "inputs/case34-history-1.png", "inputs/case34-history-2.png", "inputs/case34-history-3.png", "inputs/case34-history-4.png", "inputs/case34-history-5.png", "inputs/case34-history-6.png", "inputs/case34-history-7.png", "inputs/case34-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 34*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place all the objects into the box with their front sides facing left.", "official_native_context": "official/texts/ebfa21e484270c7971cd5e471d1b4d21782d3ec40857ea625c206a92e00e120b.npy"}, {"id": "case35", "pair_id": "pair17", "split": "train", "start": 684, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1800, "task": "Place all the objects into the box with their front sides facing left.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001800", "start_seconds": 27.36, "starts": [684], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nSubtask context (ordered by onset): Place the hammer into the box.\nAtomic instruction: right arm: Place the hammer into the cardboard box."], "file": "case35.npz", "horizons": [1], "history_indices": [0, 509, 534, 559, 584, 609, 634, 659, 684], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-083.mp4", "start": 35.88, "end": 82.12, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-052.mp4", "start": 569.5200000000001, "end": 615.7600000000001, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-054.mp4", "start": 321.2, "end": 367.44, "fps": 25}}, "source_npz_sha256": "0ebe3fc9f59a625a099f5d31fe913eb43fbb5679ca0cbda04e89be788fdb9c16", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nSubtask context (ordered by onset): Place the hammer into the box.\nAtomic instruction: right arm: Place the hammer into the cardboard box.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nSubtask context (ordered by onset): Place the hammer into the box.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.\nAtomic instruction: right arm: Place the hammer into the cardboard box.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the hammer into the box.\nAtomic instruction: right arm: Place the hammer into the cardboard box.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place all the objects into the box with their front sides facing left.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the hammer into the cardboard box."}, "input_packs": [{"chunk": 0, "start": 684, "path": "inputs/case35-chunk0.npz", "sha256": "e2c3882f629632a83231cad41525369b24aa2eb316b48cd0036240f6180ba68b"}], "input_pack": "inputs/case35-chunk0.npz", "input_image": "inputs/case35-input.png", "gt_videos": {"1": "inputs/case35-gt-1.mp4"}, "history_images": ["inputs/case35-history-0.png", "inputs/case35-history-1.png", "inputs/case35-history-2.png", "inputs/case35-history-3.png", "inputs/case35-history-4.png", "inputs/case35-history-5.png", "inputs/case35-history-6.png", "inputs/case35-history-7.png", "inputs/case35-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 35*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place all the objects into the box with their front sides facing left.", "official_native_context": "official/texts/ebfa21e484270c7971cd5e471d1b4d21782d3ec40857ea625c206a92e00e120b.npy"}, {"id": "case36", "pair_id": "pair18", "split": "val", "start": 209, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1955, "task": "Pick up the mallet and strike all xylophone keys from left to right.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001955", "start_seconds": 8.36, "starts": [209, 241, 273], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right using the mallet held by the right arm.\nAtomic instruction: right arm: Lift the mallet off the xylophone keys.", "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Strike all xylophone keys from left to right using the mallet held by the right arm.\nAtomic instruction: right arm: Lift the mallet off the xylophone keys.", "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Strike all xylophone keys from left to right using the mallet held by the right arm.\nAtomic instruction: right arm: Lift the mallet off the xylophone keys."], "file": "case36.npz", "horizons": [1, 3], "history_indices": [0, 34, 59, 84, 109, 134, 159, 184, 209], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-090.mp4", "start": 196.96000000000004, "end": 209.36000000000004, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-057.mp4", "start": 246.72000000000003, "end": 259.12, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-058.mp4", "start": 702.4800000000001, "end": 714.8800000000001, "fps": 25}}, "source_npz_sha256": "ccefb6be2e7e4aab59ac604d86f24caeeb1cc116e18938e8d86b75eeddca0a3d", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right using the mallet held by the right arm.\nAtomic instruction: right arm: Lift the mallet off the xylophone keys.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right using the mallet held by the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nAtomic instruction: right arm: Lift the mallet off the xylophone keys.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right using the mallet held by the right arm.\nAtomic instruction: right arm: Lift the mallet off the xylophone keys.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Lift the mallet off the xylophone keys."}, "input_packs": [{"chunk": 0, "start": 209, "path": "inputs/case36-chunk0.npz", "sha256": "efdd06bb532a147d7b88058326ba67af41a3a5adf749b760f5e1d7fcb19e6bef"}, {"chunk": 1, "start": 241, "path": "inputs/case36-chunk1.npz", "sha256": "f19efc346fe38f196ec9a8c0b7ed9ab3dfbfc18021b695d452b56df277a1c53f"}, {"chunk": 2, "start": 273, "path": "inputs/case36-chunk2.npz", "sha256": "f265e58e0c12493bbcedd84adde0b272c4a5a6482b7a2cb4b5ce29347a7a8cb7"}], "input_pack": "inputs/case36-chunk0.npz", "input_image": "inputs/case36-input.png", "gt_videos": {"1": "inputs/case36-gt-1.mp4", "3": "inputs/case36-gt-3.mp4"}, "history_images": ["inputs/case36-history-0.png", "inputs/case36-history-1.png", "inputs/case36-history-2.png", "inputs/case36-history-3.png", "inputs/case36-history-4.png", "inputs/case36-history-5.png", "inputs/case36-history-6.png", "inputs/case36-history-7.png", "inputs/case36-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 36*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the mallet and strike all xylophone keys from left to right.", "official_native_context": "official/texts/3ca3fe513bbd5471ec14dc31fe66472cbf50f2ac199c853d15131dacfc7821d7.npy"}, {"id": "case37", "pair_id": "pair18", "split": "train", "start": 210, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 1900, "task": "Pick up the mallet and strike all xylophone keys from left to right.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_001900", "start_seconds": 8.4, "starts": [210, 242, 274], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right with the mallet in the right arm.\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet.", "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Strike all xylophone keys from left to right with the mallet in the right arm.\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet.", "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Strike all xylophone keys from left to right with the mallet in the right arm.\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet."], "file": "case37.npz", "horizons": [1, 3], "history_indices": [0, 35, 60, 85, 110, 135, 160, 185, 210], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-089.mp4", "start": 48.599999999999994, "end": 60.99999999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-056.mp4", "start": 359.88, "end": 372.28, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-058.mp4", "start": 24.72, "end": 37.12, "fps": 25}}, "source_npz_sha256": "2d0f683b17543b680ae17a1b931a7ac59596aa6e7a6995a90ad9a598d006b372", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right with the mallet in the right arm.\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right with the mallet in the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Hand the mallet from the left gripper to the right gripper.; Strike all xylophone keys from left to right with the mallet in the right arm.\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the mallet and strike all xylophone keys from left to right.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Strike the xylophone keys with the held mallet."}, "input_packs": [{"chunk": 0, "start": 210, "path": "inputs/case37-chunk0.npz", "sha256": "3b8055e50577c3ae620fecd219692299afbeab9891a2591d725fbef00dc80f06"}, {"chunk": 1, "start": 242, "path": "inputs/case37-chunk1.npz", "sha256": "8e7355c13000124374d2fe733927ec1276a4efc0cabf406543a78692002a4785"}, {"chunk": 2, "start": 274, "path": "inputs/case37-chunk2.npz", "sha256": "e068e698d66b333a724e85b4a0bc03ba208d1d29f61cf45f626d589df8db81fc"}], "input_pack": "inputs/case37-chunk0.npz", "input_image": "inputs/case37-input.png", "gt_videos": {"1": "inputs/case37-gt-1.mp4", "3": "inputs/case37-gt-3.mp4"}, "history_images": ["inputs/case37-history-0.png", "inputs/case37-history-1.png", "inputs/case37-history-2.png", "inputs/case37-history-3.png", "inputs/case37-history-4.png", "inputs/case37-history-5.png", "inputs/case37-history-6.png", "inputs/case37-history-7.png", "inputs/case37-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 37*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the mallet and strike all xylophone keys from left to right.", "official_native_context": "official/texts/3ca3fe513bbd5471ec14dc31fe66472cbf50f2ac199c853d15131dacfc7821d7.npy"}, {"id": "case38", "pair_id": "pair19", "split": "val", "start": 548, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2093, "task": "Place all stacking toy pieces onto the correct pegs.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002093", "start_seconds": 21.92, "starts": [548], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nSubtask context (ordered by onset): Place the second orange circular piece onto the base.\nAtomic instruction: left arm: Place the orange circular piece onto the base."], "file": "case38.npz", "horizons": [1], "history_indices": [0, 373, 398, 423, 448, 473, 498, 523, 548], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-097.mp4", "start": 326.8, "end": 363.0, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-062.mp4", "start": 216.96, "end": 253.16000000000003, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-064.mp4", "start": 181.88, "end": 218.07999999999998, "fps": 25}}, "source_npz_sha256": "123467ff09757a795d17bc4f238b58ca8f07c90935f60cf67ee0ace95bba373f", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nSubtask context (ordered by onset): Place the second orange circular piece onto the base.\nAtomic instruction: left arm: Place the orange circular piece onto the base.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nSubtask context (ordered by onset): Place the second orange circular piece onto the base.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nAtomic instruction: left arm: Place the orange circular piece onto the base.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the second orange circular piece onto the base.\nAtomic instruction: left arm: Place the orange circular piece onto the base.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the orange circular piece onto the base."}, "input_packs": [{"chunk": 0, "start": 548, "path": "inputs/case38-chunk0.npz", "sha256": "0fb314e88904e2b15bb5a8615edcf2e65363442801ae628c7205fc2f6b064f65"}], "input_pack": "inputs/case38-chunk0.npz", "input_image": "inputs/case38-input.png", "gt_videos": {"1": "inputs/case38-gt-1.mp4"}, "history_images": ["inputs/case38-history-0.png", "inputs/case38-history-1.png", "inputs/case38-history-2.png", "inputs/case38-history-3.png", "inputs/case38-history-4.png", "inputs/case38-history-5.png", "inputs/case38-history-6.png", "inputs/case38-history-7.png", "inputs/case38-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 38*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place all stacking toy pieces onto the correct pegs.", "official_native_context": "official/texts/285a24a5a7f0b6b4ba38747143b5a10ecc69eafe105c72fa6773d1a1a4e68f83.npy"}, {"id": "case39", "pair_id": "pair19", "split": "train", "start": 541, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2000, "task": "Place all stacking toy pieces onto the correct pegs.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002000", "start_seconds": 21.64, "starts": [541], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nSubtask context (ordered by onset): Place the second yellow square stacking toy piece onto the peg.\nAtomic instruction: right arm: Place the second yellow square stacking toy piece onto the peg."], "file": "case39.npz", "horizons": [1], "history_indices": [0, 366, 391, 416, 441, 466, 491, 516, 541], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-091.mp4", "start": 223.16, "end": 258.64, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-058.mp4", "start": 25.36, "end": 60.839999999999996, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-059.mp4", "start": 532.04, "end": 567.52, "fps": 25}}, "source_npz_sha256": "0efd8db3c5792152afb07021ff3e429724905c68c8c9e7f958eae1e49619d63a", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nSubtask context (ordered by onset): Place the second yellow square stacking toy piece onto the peg.\nAtomic instruction: right arm: Place the second yellow square stacking toy piece onto the peg.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nSubtask context (ordered by onset): Place the second yellow square stacking toy piece onto the peg.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.\nAtomic instruction: right arm: Place the second yellow square stacking toy piece onto the peg.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the second yellow square stacking toy piece onto the peg.\nAtomic instruction: right arm: Place the second yellow square stacking toy piece onto the peg.", "l3": "Robot: robodojo_sim_arx_x5. Task: Place all stacking toy pieces onto the correct pegs.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the second yellow square stacking toy piece onto the peg."}, "input_packs": [{"chunk": 0, "start": 541, "path": "inputs/case39-chunk0.npz", "sha256": "1d8396a05ba8037dfe477a6e159917fda005e15ec42566aded374e6a79747279"}], "input_pack": "inputs/case39-chunk0.npz", "input_image": "inputs/case39-input.png", "gt_videos": {"1": "inputs/case39-gt-1.mp4"}, "history_images": ["inputs/case39-history-0.png", "inputs/case39-history-1.png", "inputs/case39-history-2.png", "inputs/case39-history-3.png", "inputs/case39-history-4.png", "inputs/case39-history-5.png", "inputs/case39-history-6.png", "inputs/case39-history-7.png", "inputs/case39-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 39*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Place all stacking toy pieces onto the correct pegs.", "official_native_context": "official/texts/285a24a5a7f0b6b4ba38747143b5a10ecc69eafe105c72fa6773d1a1a4e68f83.npy"}, {"id": "case40", "pair_id": "pair20", "split": "val", "start": 612, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2127, "task": "Play tic-tac-toe as the first player and fill the board with the opponent.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002127", "start_seconds": 24.48, "starts": [612], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nSubtask context (ordered by onset): Place the first yellow ring game piece on an empty square of the tic-tac-toe board.\nAtomic instruction: right arm: Place the yellow ring game piece onto an empty square on the tic-tac-toe board."], "file": "case40.npz", "horizons": [1], "history_indices": [0, 437, 462, 487, 512, 537, 562, 587, 612], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-100.mp4", "start": 0.0, "end": 35.24, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-063.mp4", "start": 599.36, "end": 634.6, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-065.mp4", "start": 633.68, "end": 668.92, "fps": 25}}, "source_npz_sha256": "81cab4a01dc98dcf646bf36db9b3e1635f206fa17598b9fd3b076b3f89b935b2", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nSubtask context (ordered by onset): Place the first yellow ring game piece on an empty square of the tic-tac-toe board.\nAtomic instruction: right arm: Place the yellow ring game piece onto an empty square on the tic-tac-toe board.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nSubtask context (ordered by onset): Place the first yellow ring game piece on an empty square of the tic-tac-toe board.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nAtomic instruction: right arm: Place the yellow ring game piece onto an empty square on the tic-tac-toe board.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the first yellow ring game piece on an empty square of the tic-tac-toe board.\nAtomic instruction: right arm: Place the yellow ring game piece onto an empty square on the tic-tac-toe board.", "l3": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the yellow ring game piece onto an empty square on the tic-tac-toe board."}, "input_packs": [{"chunk": 0, "start": 612, "path": "inputs/case40-chunk0.npz", "sha256": "dfa7470fe34ee00c942762a9481d56a8e32e65dd84dc40b48c498c01b5b7b5a6"}], "input_pack": "inputs/case40-chunk0.npz", "input_image": "inputs/case40-input.png", "gt_videos": {"1": "inputs/case40-gt-1.mp4"}, "history_images": ["inputs/case40-history-0.png", "inputs/case40-history-1.png", "inputs/case40-history-2.png", "inputs/case40-history-3.png", "inputs/case40-history-4.png", "inputs/case40-history-5.png", "inputs/case40-history-6.png", "inputs/case40-history-7.png", "inputs/case40-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 40*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Play tic-tac-toe as the first player and fill the board with the opponent.", "official_native_context": "official/texts/0c0b0979256fb90a22a9bdc1d65044cb9e62f969a8b5a769c5e8e088de13b83c.npy"}, {"id": "case41", "pair_id": "pair20", "split": "train", "start": 587, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2100, "task": "Play tic-tac-toe as the first player and fill the board with the opponent.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002100", "start_seconds": 23.48, "starts": [587], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nSubtask context (ordered by onset): Place a yellow ring game piece on an empty square of the board.\nAtomic instruction: right arm: Grasp the yellow ring game piece on the right side of the table."], "file": "case41.npz", "horizons": [1], "history_indices": [0, 412, 437, 462, 487, 512, 537, 562, 587], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-098.mp4", "start": 35.84, "end": 71.16, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-062.mp4", "start": 474.79999999999995, "end": 510.11999999999995, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-064.mp4", "start": 439.7199999999999, "end": 475.0399999999999, "fps": 25}}, "source_npz_sha256": "8a5393c5b0c4a254be20041cc89ca489324ae1b9ebd335e676e65c5172f56bb2", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nSubtask context (ordered by onset): Place a yellow ring game piece on an empty square of the board.\nAtomic instruction: right arm: Grasp the yellow ring game piece on the right side of the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nSubtask context (ordered by onset): Place a yellow ring game piece on an empty square of the board.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.\nAtomic instruction: right arm: Grasp the yellow ring game piece on the right side of the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place a yellow ring game piece on an empty square of the board.\nAtomic instruction: right arm: Grasp the yellow ring game piece on the right side of the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Play tic-tac-toe as the first player and fill the board with the opponent.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the yellow ring game piece on the right side of the table."}, "input_packs": [{"chunk": 0, "start": 587, "path": "inputs/case41-chunk0.npz", "sha256": "7b4ed63c177f3cb4da6171e7c0ac5eb557eead2d950385b2fb3ee9652c08fe11"}], "input_pack": "inputs/case41-chunk0.npz", "input_image": "inputs/case41-input.png", "gt_videos": {"1": "inputs/case41-gt-1.mp4"}, "history_images": ["inputs/case41-history-0.png", "inputs/case41-history-1.png", "inputs/case41-history-2.png", "inputs/case41-history-3.png", "inputs/case41-history-4.png", "inputs/case41-history-5.png", "inputs/case41-history-6.png", "inputs/case41-history-7.png", "inputs/case41-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 41*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Play tic-tac-toe as the first player and fill the board with the opponent.", "official_native_context": "official/texts/0c0b0979256fb90a22a9bdc1d65044cb9e62f969a8b5a769c5e8e088de13b83c.npy"}, {"id": "case42", "pair_id": "pair21", "split": "val", "start": 134, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2281, "task": "Plug the charger into the power strip.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002281", "start_seconds": 5.36, "starts": [134], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nSubtask context (ordered by onset): Plug the charger into the power strip using the right arm.\nAtomic instruction: right arm: Move the charger to the right and place it on the table surface."], "file": "case42.npz", "horizons": [1], "history_indices": [0, 0, 0, 9, 34, 59, 84, 109, 134], "distinct_history_frames": 7, "history_padding": [false, true, true, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-106.mp4", "start": 274.71999999999997, "end": 282.55999999999995, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-067.mp4", "start": 643.5599999999996, "end": 651.3999999999996, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-069.mp4", "start": 800.3599999999996, "end": 808.1999999999996, "fps": 25}}, "source_npz_sha256": "f490150732d4fdfdb17fc17d38ac6c43329de9007515d1d84d5fc1b411186e0e", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nSubtask context (ordered by onset): Plug the charger into the power strip using the right arm.\nAtomic instruction: right arm: Move the charger to the right and place it on the table surface.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nSubtask context (ordered by onset): Plug the charger into the power strip using the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nAtomic instruction: right arm: Move the charger to the right and place it on the table surface.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Plug the charger into the power strip using the right arm.\nAtomic instruction: right arm: Move the charger to the right and place it on the table surface.", "l3": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Move the charger to the right and place it on the table surface."}, "input_packs": [{"chunk": 0, "start": 134, "path": "inputs/case42-chunk0.npz", "sha256": "0804f2069b4d755676d6f730f2c7393a40f565c4ffb097c04af99963b2f7b9ac"}], "input_pack": "inputs/case42-chunk0.npz", "input_image": "inputs/case42-input.png", "gt_videos": {"1": "inputs/case42-gt-1.mp4"}, "history_images": ["inputs/case42-history-0.png", "inputs/case42-history-1.png", "inputs/case42-history-2.png", "inputs/case42-history-3.png", "inputs/case42-history-4.png", "inputs/case42-history-5.png", "inputs/case42-history-6.png", "inputs/case42-history-7.png", "inputs/case42-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 42*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Plug the charger into the power strip.", "official_native_context": "official/texts/d5a30ed2dee778067044a7d2a5da03e887e072d45f8b17b364931e45a00470b7.npy"}, {"id": "case43", "pair_id": "pair21", "split": "train", "start": 136, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2200, "task": "Plug the charger into the power strip.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002200", "start_seconds": 5.44, "starts": [136], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nSubtask context (ordered by onset): Move the charger to the power strip location and place it down.\nAtomic instruction: left arm: Move the charger to the left and place it on the table."], "file": "case43.npz", "horizons": [1], "history_indices": [0, 0, 0, 11, 36, 61, 86, 111, 136], "distinct_history_frames": 7, "history_padding": [false, true, true, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-105.mp4", "start": 104.24000000000001, "end": 112.04, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-066.mp4", "start": 734.4, "end": 742.1999999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-069.mp4", "start": 34.28, "end": 42.08, "fps": 25}}, "source_npz_sha256": "2e52c00554a1e9eac3c3abca43ebb08fffa3cb88228cc83ec5daedf091aec756", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nSubtask context (ordered by onset): Move the charger to the power strip location and place it down.\nAtomic instruction: left arm: Move the charger to the left and place it on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nSubtask context (ordered by onset): Move the charger to the power strip location and place it down.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.\nAtomic instruction: left arm: Move the charger to the left and place it on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Move the charger to the power strip location and place it down.\nAtomic instruction: left arm: Move the charger to the left and place it on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Plug the charger into the power strip.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Move the charger to the left and place it on the table."}, "input_packs": [{"chunk": 0, "start": 136, "path": "inputs/case43-chunk0.npz", "sha256": "4cf2365f3c708853553138b04c2ad99d468d216a416b8fd56c140b5bfc05c394"}], "input_pack": "inputs/case43-chunk0.npz", "input_image": "inputs/case43-input.png", "gt_videos": {"1": "inputs/case43-gt-1.mp4"}, "history_images": ["inputs/case43-history-0.png", "inputs/case43-history-1.png", "inputs/case43-history-2.png", "inputs/case43-history-3.png", "inputs/case43-history-4.png", "inputs/case43-history-5.png", "inputs/case43-history-6.png", "inputs/case43-history-7.png", "inputs/case43-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 43*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Plug the charger into the power strip.", "official_native_context": "official/texts/d5a30ed2dee778067044a7d2a5da03e887e072d45f8b17b364931e45a00470b7.npy"}, {"id": "case44", "pair_id": "pair22", "split": "val", "start": 306, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2367, "task": "Pour all the balls from the cup into the vase.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002367", "start_seconds": 12.24, "starts": [306], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nSubtask context (ordered by onset): Place the cup on the table next to the vase and release it.\nAtomic instruction: right arm: Place the cup on the table next to the vase and release it."], "file": "case44.npz", "horizons": [1], "history_indices": [0, 131, 156, 181, 206, 231, 256, 281, 306], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-108.mp4", "start": 146.55999999999997, "end": 163.75999999999996, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-068.mp4", "start": 648.8400000000001, "end": 666.0400000000002, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-070.mp4", "start": 888.0000000000001, "end": 905.2000000000002, "fps": 25}}, "source_npz_sha256": "94cfd91c416fed0aa6b4a684b12f27d6c02933aa613ad1a9f958f47c37e9931b", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nSubtask context (ordered by onset): Place the cup on the table next to the vase and release it.\nAtomic instruction: right arm: Place the cup on the table next to the vase and release it.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nSubtask context (ordered by onset): Place the cup on the table next to the vase and release it.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nAtomic instruction: right arm: Place the cup on the table next to the vase and release it.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the cup on the table next to the vase and release it.\nAtomic instruction: right arm: Place the cup on the table next to the vase and release it.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the cup on the table next to the vase and release it."}, "input_packs": [{"chunk": 0, "start": 306, "path": "inputs/case44-chunk0.npz", "sha256": "53ba0cf6439405cff614494600471c7a62f2d44ec7a66353f7671733adf76e79"}], "input_pack": "inputs/case44-chunk0.npz", "input_image": "inputs/case44-input.png", "gt_videos": {"1": "inputs/case44-gt-1.mp4"}, "history_images": ["inputs/case44-history-0.png", "inputs/case44-history-1.png", "inputs/case44-history-2.png", "inputs/case44-history-3.png", "inputs/case44-history-4.png", "inputs/case44-history-5.png", "inputs/case44-history-6.png", "inputs/case44-history-7.png", "inputs/case44-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 44*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pour all the balls from the cup into the vase.", "official_native_context": "official/texts/6cdec038eb07cc9a975e74f4c1c2c429748cec7731cd6a5ae6f77fa8cfd93939.npy"}, {"id": "case45", "pair_id": "pair22", "split": "train", "start": 265, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2300, "task": "Pour all the balls from the cup into the vase.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002300", "start_seconds": 10.6, "starts": [265], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nSubtask context (ordered by onset): Re-grasp the yellow cup and place it on the table next to the vase.\nAtomic instruction: left arm: Place the yellow cup on the table next to the vase."], "file": "case45.npz", "horizons": [1], "history_indices": [0, 90, 115, 140, 165, 190, 215, 240, 265], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-106.mp4", "start": 433.2399999999999, "end": 447.4399999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-067.mp4", "start": 802.0799999999995, "end": 816.2799999999995, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-069.mp4", "start": 958.8799999999994, "end": 973.0799999999995, "fps": 25}}, "source_npz_sha256": "119faa0682428f1beea5a006da42c9e9fef83eb4d3049d001ae672a132117412", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nSubtask context (ordered by onset): Re-grasp the yellow cup and place it on the table next to the vase.\nAtomic instruction: left arm: Place the yellow cup on the table next to the vase.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nSubtask context (ordered by onset): Re-grasp the yellow cup and place it on the table next to the vase.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.\nAtomic instruction: left arm: Place the yellow cup on the table next to the vase.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Re-grasp the yellow cup and place it on the table next to the vase.\nAtomic instruction: left arm: Place the yellow cup on the table next to the vase.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pour all the balls from the cup into the vase.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the yellow cup on the table next to the vase."}, "input_packs": [{"chunk": 0, "start": 265, "path": "inputs/case45-chunk0.npz", "sha256": "6df7c8632978e23c648ec1b208c7c547e1f875141fab3d6b85a70d62ca6bd0a9"}], "input_pack": "inputs/case45-chunk0.npz", "input_image": "inputs/case45-input.png", "gt_videos": {"1": "inputs/case45-gt-1.mp4"}, "history_images": ["inputs/case45-history-0.png", "inputs/case45-history-1.png", "inputs/case45-history-2.png", "inputs/case45-history-3.png", "inputs/case45-history-4.png", "inputs/case45-history-5.png", "inputs/case45-history-6.png", "inputs/case45-history-7.png", "inputs/case45-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 45*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pour all the balls from the cup into the vase.", "official_native_context": "official/texts/6cdec038eb07cc9a975e74f4c1c2c429748cec7731cd6a5ae6f77fa8cfd93939.npy"}, {"id": "case46", "pair_id": "pair23", "split": "val", "start": 78, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2400, "task": "Pour the liquid from the bottle into the cup.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002400", "start_seconds": 3.12, "starts": [78, 110, 142], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup.", "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup.", "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup."], "file": "case46.npz", "horizons": [1, 3], "history_indices": [0, 0, 0, 0, 0, 3, 28, 53, 78], "distinct_history_frames": 5, "history_padding": [false, true, true, true, true, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-108.mp4", "start": 582.2799999999999, "end": 589.3199999999998, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-069.mp4", "start": 96.28, "end": 103.32000000000001, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-071.mp4", "start": 277.7599999999999, "end": 284.7999999999999, "fps": 25}}, "source_npz_sha256": "16fbf8384bab7690a0eeb4c4e66a9a36c2a0035bc798576da576c436abf1c9b2", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Tilt the bottle held in the left arm to pour liquid into the cup."}, "input_packs": [{"chunk": 0, "start": 78, "path": "inputs/case46-chunk0.npz", "sha256": "71dad6b3195948974d5192ccd0fb1a2a44df764fa00a7e29a9f6727daec891ea"}, {"chunk": 1, "start": 110, "path": "inputs/case46-chunk1.npz", "sha256": "e0019f507a413d576b658252582ae78cd0329a3994880a6148fbcaeca29e0bc3"}, {"chunk": 2, "start": 142, "path": "inputs/case46-chunk2.npz", "sha256": "5eab89885ee9117f2040f36587882e589d95f749c2afc092e26dd7931dd0a60e"}], "input_pack": "inputs/case46-chunk0.npz", "input_image": "inputs/case46-input.png", "gt_videos": {"1": "inputs/case46-gt-1.mp4", "3": "inputs/case46-gt-3.mp4"}, "history_images": ["inputs/case46-history-0.png", "inputs/case46-history-1.png", "inputs/case46-history-2.png", "inputs/case46-history-3.png", "inputs/case46-history-4.png", "inputs/case46-history-5.png", "inputs/case46-history-6.png", "inputs/case46-history-7.png", "inputs/case46-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 46*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pour the liquid from the bottle into the cup.", "official_native_context": "official/texts/9496ba509aa58af057c99af592970d3ef35bda18f1efdb19b3e77b0311a68012.npy"}, {"id": "case47", "pair_id": "pair23", "split": "train", "start": 99, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2401, "task": "Pour the liquid from the bottle into the cup.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002401", "start_seconds": 3.96, "starts": [99, 131, 163], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup.", "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup.", "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup."], "file": "case47.npz", "horizons": [1, 3], "history_indices": [0, 0, 0, 0, 0, 24, 49, 74, 99], "distinct_history_frames": 5, "history_padding": [false, true, true, true, true, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-108.mp4", "start": 589.3199999999998, "end": 597.2799999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-069.mp4", "start": 103.32000000000001, "end": 111.28, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-071.mp4", "start": 284.7999999999999, "end": 292.7599999999999, "fps": 25}}, "source_npz_sha256": "ef039d3d879cdb8cdd88daf6bdab45a88b7dab5cf9e712f1d5ad0b9547c8f3fb", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Pour the liquid from the bottle into the cup.\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pour the liquid from the bottle into the cup.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Pour the liquid from the bottle into the cup."}, "input_packs": [{"chunk": 0, "start": 99, "path": "inputs/case47-chunk0.npz", "sha256": "7668aac24382069f0edd13bc9e073f648de1f92bab067264ccdc4d95c644e39c"}, {"chunk": 1, "start": 131, "path": "inputs/case47-chunk1.npz", "sha256": "a343e5ae766949b14ba8341fd99a0f51007b1a2347e8c4755dab47f7ae8cd4c9"}, {"chunk": 2, "start": 163, "path": "inputs/case47-chunk2.npz", "sha256": "ced1dc70f056e0746afa7d1d3b16f5fe9a8e9020d670250aef304dcb31ab8e5a"}], "input_pack": "inputs/case47-chunk0.npz", "input_image": "inputs/case47-input.png", "gt_videos": {"1": "inputs/case47-gt-1.mp4", "3": "inputs/case47-gt-3.mp4"}, "history_images": ["inputs/case47-history-0.png", "inputs/case47-history-1.png", "inputs/case47-history-2.png", "inputs/case47-history-3.png", "inputs/case47-history-4.png", "inputs/case47-history-5.png", "inputs/case47-history-6.png", "inputs/case47-history-7.png", "inputs/case47-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 47*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pour the liquid from the bottle into the cup.", "official_native_context": "official/texts/9496ba509aa58af057c99af592970d3ef35bda18f1efdb19b3e77b0311a68012.npy"}, {"id": "case48", "pair_id": "pair24", "split": "val", "start": 168, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2539, "task": "Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002539", "start_seconds": 6.72, "starts": [168, 200, 232], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the right red button repeatedly with the left arm while the right arm moves to the blue button.\nAtomic instruction: left arm: Press the right red button repeatedly.", "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the right red button repeatedly with the left arm while the right arm moves to the blue button.\nAtomic instruction: left arm: Press the right red button repeatedly.", "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the right red button repeatedly with the left arm while the right arm moves to the blue button.\nAtomic instruction: left arm: Press the right red button repeatedly."], "file": "case48.npz", "horizons": [1, 3], "history_indices": [0, 0, 18, 43, 68, 93, 118, 143, 168], "distinct_history_frames": 8, "history_padding": [false, true, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-111.mp4", "start": 394.4399999999999, "end": 408.3199999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-070.mp4", "start": 252.84, "end": 266.72, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-072.mp4", "start": 833.0800000000002, "end": 846.9600000000002, "fps": 25}}, "source_npz_sha256": "951fa499156f3c6d112fad3e8d261d2533179ddb6a2a07325ebc5adcef009572", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the right red button repeatedly with the left arm while the right arm moves to the blue button.\nAtomic instruction: left arm: Press the right red button repeatedly.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the right red button repeatedly with the left arm while the right arm moves to the blue button.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nAtomic instruction: left arm: Press the right red button repeatedly.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Press the right red button repeatedly with the left arm while the right arm moves to the blue button.\nAtomic instruction: left arm: Press the right red button repeatedly.", "l3": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Press the right red button repeatedly."}, "input_packs": [{"chunk": 0, "start": 168, "path": "inputs/case48-chunk0.npz", "sha256": "6117753fb4af56ceca4271ae8c1358347b847099e104187f485d5b724ae59b85"}, {"chunk": 1, "start": 200, "path": "inputs/case48-chunk1.npz", "sha256": "825ea5406c2011794941d50f2497fd0b5a5bb245f08750541004d29167fa6d74"}, {"chunk": 2, "start": 232, "path": "inputs/case48-chunk2.npz", "sha256": "b389a277413f92e26f4a3dd1015f78d8fedd7ac1f6b3eaf9f1adeddc922bf214"}], "input_pack": "inputs/case48-chunk0.npz", "input_image": "inputs/case48-input.png", "gt_videos": {"1": "inputs/case48-gt-1.mp4", "3": "inputs/case48-gt-3.mp4"}, "history_images": ["inputs/case48-history-0.png", "inputs/case48-history-1.png", "inputs/case48-history-2.png", "inputs/case48-history-3.png", "inputs/case48-history-4.png", "inputs/case48-history-5.png", "inputs/case48-history-6.png", "inputs/case48-history-7.png", "inputs/case48-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 48*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.", "official_native_context": "official/texts/5ec7d5b650e976e945182ef53df6ec68cf66664d7df217685ae5ac68847c0c0c.npy"}, {"id": "case49", "pair_id": "pair24", "split": "train", "start": 87, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2500, "task": "Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002500", "start_seconds": 3.48, "starts": [87, 119, 151], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the left red button multiple times.\nAtomic instruction: left arm: Press the left red button again.", "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the left red button multiple times.\nAtomic instruction: left arm: Press the left red button again.", "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the left red button multiple times.\nAtomic instruction: left arm: Press the left red button again."], "file": "case49.npz", "horizons": [1, 3], "history_indices": [0, 0, 0, 0, 0, 12, 37, 62, 87], "distinct_history_frames": 5, "history_padding": [false, true, true, true, true, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-110.mp4", "start": 352.52000000000004, "end": 367.6, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-069.mp4", "start": 1063.28, "end": 1078.36, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-072.mp4", "start": 204.88000000000002, "end": 219.96000000000004, "fps": 25}}, "source_npz_sha256": "0a7668c67a556bd3e241079592191db8a94dbf38d98ff15bb41a5807fb0a945f", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the left red button multiple times.\nAtomic instruction: left arm: Press the left red button again.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nSubtask context (ordered by onset): Press the left red button multiple times.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.\nAtomic instruction: left arm: Press the left red button again.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Press the left red button multiple times.\nAtomic instruction: left arm: Press the left red button again.", "l3": "Robot: robodojo_sim_arx_x5. Task: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Press the left red button again."}, "input_packs": [{"chunk": 0, "start": 87, "path": "inputs/case49-chunk0.npz", "sha256": "1a20b82f0bb86c087834761fc51078f2602baa11c44905a0d01a132ba5220f92"}, {"chunk": 1, "start": 119, "path": "inputs/case49-chunk1.npz", "sha256": "4942c0978eb7b3c40d1237d1ed3c3ae7622a8f1132a51b87511b5ad41f8edd5b"}, {"chunk": 2, "start": 151, "path": "inputs/case49-chunk2.npz", "sha256": "ba0d1228290f35e03e60f173eae88f2fd8ea0866ddb0b32bb6945e4636a6f1d5"}], "input_pack": "inputs/case49-chunk0.npz", "input_image": "inputs/case49-input.png", "gt_videos": {"1": "inputs/case49-gt-1.mp4", "3": "inputs/case49-gt-3.mp4"}, "history_images": ["inputs/case49-history-0.png", "inputs/case49-history-1.png", "inputs/case49-history-2.png", "inputs/case49-history-3.png", "inputs/case49-history-4.png", "inputs/case49-history-5.png", "inputs/case49-history-6.png", "inputs/case49-history-7.png", "inputs/case49-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 49*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm.", "official_native_context": "official/texts/5ec7d5b650e976e945182ef53df6ec68cf66664d7df217685ae5ac68847c0c0c.npy"}, {"id": "case50", "pair_id": "pair25", "split": "val", "start": 230, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2685, "task": "Push the T-shaped block to align it precisely with the gray T-shaped pad.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002685", "start_seconds": 9.2, "starts": [230, 262, 294], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block with the right arm and push it to align precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Release the red T-shaped block onto the table.", "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block with the right arm and push it to align precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Release the red T-shaped block onto the table.", "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block with the right arm and push it to align precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Release the red T-shaped block onto the table."], "file": "case50.npz", "horizons": [1, 3], "history_indices": [0, 55, 80, 105, 130, 155, 180, 205, 230], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-115.mp4", "start": 9.04, "end": 23.84, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-072.mp4", "start": 155.88000000000002, "end": 170.68000000000004, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-074.mp4", "start": 921.8400000000001, "end": 936.6400000000001, "fps": 25}}, "source_npz_sha256": "49d109ba862717b98655efe9205e11a911ddb071b1b847de25b4f22a7f4fe18a", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block with the right arm and push it to align precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Release the red T-shaped block onto the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block with the right arm and push it to align precisely with the gray T-shaped pad.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Release the red T-shaped block onto the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Grasp the red T-shaped block with the right arm and push it to align precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Release the red T-shaped block onto the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Release the red T-shaped block onto the table."}, "input_packs": [{"chunk": 0, "start": 230, "path": "inputs/case50-chunk0.npz", "sha256": "78d84489342552a69c5814617cd75a975bccb0a8175004cf2cf18e5db0296fe7"}, {"chunk": 1, "start": 262, "path": "inputs/case50-chunk1.npz", "sha256": "0c42791d46fe1b038e8be9ff3e01beac9b8bd83a9627ce591ed2fd628e2f3406"}, {"chunk": 2, "start": 294, "path": "inputs/case50-chunk2.npz", "sha256": "1728f72ebb71d63d1989f3ac5e7b29a60fc80532cd2d0c2d66815f5f290bb952"}], "input_pack": "inputs/case50-chunk0.npz", "input_image": "inputs/case50-input.png", "gt_videos": {"1": "inputs/case50-gt-1.mp4", "3": "inputs/case50-gt-3.mp4"}, "history_images": ["inputs/case50-history-0.png", "inputs/case50-history-1.png", "inputs/case50-history-2.png", "inputs/case50-history-3.png", "inputs/case50-history-4.png", "inputs/case50-history-5.png", "inputs/case50-history-6.png", "inputs/case50-history-7.png", "inputs/case50-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 50*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Push the T-shaped block to align it precisely with the gray T-shaped pad.", "official_native_context": "official/texts/9d1bc1fdf3d72f3f5fb0e36f68db5358b8392c21cc2e3bd497192a20715318fd.npy"}, {"id": "case51", "pair_id": "pair25", "split": "train", "start": 209, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2600, "task": "Push the T-shaped block to align it precisely with the gray T-shaped pad.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002600", "start_seconds": 8.36, "starts": [209, 241, 273], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block and push it to align with the gray T-shaped pad.\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad.", "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block and push it to align with the gray T-shaped pad.\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad.", "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block and push it to align with the gray T-shaped pad.\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad."], "file": "case51.npz", "horizons": [1, 3], "history_indices": [0, 34, 59, 84, 109, 134, 159, 184, 209], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-113.mp4", "start": 124.64000000000001, "end": 139.0, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-071.mp4", "start": 114.24000000000001, "end": 128.60000000000002, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-073.mp4", "start": 779.2799999999997, "end": 793.6399999999998, "fps": 25}}, "source_npz_sha256": "438206f73b429ca324b1c8370ab5cfac45580a2f9bd6338997e545d69eb7ee07", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block and push it to align with the gray T-shaped pad.\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nSubtask context (ordered by onset): Grasp the red T-shaped block and push it to align with the gray T-shaped pad.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Grasp the red T-shaped block and push it to align with the gray T-shaped pad.\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad.", "l3": "Robot: robodojo_sim_arx_x5. Task: Push the T-shaped block to align it precisely with the gray T-shaped pad.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Push the held T-shaped block across the table to align it with the gray T-shaped pad."}, "input_packs": [{"chunk": 0, "start": 209, "path": "inputs/case51-chunk0.npz", "sha256": "b743ffa740877343dca2e250bb0a2864e8741a9a29d9406c8ef76c0bdd821735"}, {"chunk": 1, "start": 241, "path": "inputs/case51-chunk1.npz", "sha256": "cf648c89cf43714c8f1a5f606c6752352513ac560669a16e86391ee5a790035e"}, {"chunk": 2, "start": 273, "path": "inputs/case51-chunk2.npz", "sha256": "b22b50aa388aa35cf3466b132eac3426cf68cb297989573817800d9b6300e53c"}], "input_pack": "inputs/case51-chunk0.npz", "input_image": "inputs/case51-input.png", "gt_videos": {"1": "inputs/case51-gt-1.mp4", "3": "inputs/case51-gt-3.mp4"}, "history_images": ["inputs/case51-history-0.png", "inputs/case51-history-1.png", "inputs/case51-history-2.png", "inputs/case51-history-3.png", "inputs/case51-history-4.png", "inputs/case51-history-5.png", "inputs/case51-history-6.png", "inputs/case51-history-7.png", "inputs/case51-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 51*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Push the T-shaped block to align it precisely with the gray T-shaped pad.", "official_native_context": "official/texts/9d1bc1fdf3d72f3f5fb0e36f68db5358b8392c21cc2e3bd497192a20715318fd.npy"}, {"id": "case52", "pair_id": "pair26", "split": "val", "start": 269, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2755, "task": "Pick up the bottles and throw them into the dustbin, using handover when needed.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002755", "start_seconds": 10.76, "starts": [269], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nSubtask context (ordered by onset): Throw the blue-capped bottle into the dustbin.\nAtomic instruction: right arm: Retreat the empty gripper away from the handover location."], "file": "case52.npz", "horizons": [1], "history_indices": [0, 94, 119, 144, 169, 194, 219, 244, 269], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-116.mp4", "start": 436.32000000000005, "end": 448.84000000000003, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-073.mp4", "start": 49.239999999999995, "end": 61.75999999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-076.mp4", "start": 16.12, "end": 28.64, "fps": 25}}, "source_npz_sha256": "7f3f6d6c61dc6d1b95fc5f284bd26bc27b97b592123d4724666889cd9dac24fa", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nSubtask context (ordered by onset): Throw the blue-capped bottle into the dustbin.\nAtomic instruction: right arm: Retreat the empty gripper away from the handover location.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nSubtask context (ordered by onset): Throw the blue-capped bottle into the dustbin.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nAtomic instruction: right arm: Retreat the empty gripper away from the handover location.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Throw the blue-capped bottle into the dustbin.\nAtomic instruction: right arm: Retreat the empty gripper away from the handover location.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Retreat the empty gripper away from the handover location."}, "input_packs": [{"chunk": 0, "start": 269, "path": "inputs/case52-chunk0.npz", "sha256": "ea829d3e81ddfa96f448c2f7b89f91c7f757cbfa16a50b7af5372e80298bb958"}], "input_pack": "inputs/case52-chunk0.npz", "input_image": "inputs/case52-input.png", "gt_videos": {"1": "inputs/case52-gt-1.mp4"}, "history_images": ["inputs/case52-history-0.png", "inputs/case52-history-1.png", "inputs/case52-history-2.png", "inputs/case52-history-3.png", "inputs/case52-history-4.png", "inputs/case52-history-5.png", "inputs/case52-history-6.png", "inputs/case52-history-7.png", "inputs/case52-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 52*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the bottles and throw them into the dustbin, using handover when needed.", "official_native_context": "official/texts/8467bc83bf13fff6dd5a4cf4ee956df7a2649f9ef728f00718ccac4c84b3e934.npy"}, {"id": "case53", "pair_id": "pair26", "split": "train", "start": 230, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2700, "task": "Pick up the bottles and throw them into the dustbin, using handover when needed.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002700", "start_seconds": 9.2, "starts": [230], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nSubtask context (ordered by onset): Throw the remaining bottle into the dustbin.\nAtomic instruction: left arm: Grasp the remaining bottle on the table."], "file": "case53.npz", "horizons": [1], "history_indices": [0, 55, 80, 105, 130, 155, 180, 205, 230], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-115.mp4", "start": 176.8, "end": 191.28, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-072.mp4", "start": 323.63999999999993, "end": 338.11999999999995, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-075.mp4", "start": 62.2, "end": 76.68, "fps": 25}}, "source_npz_sha256": "d0cbef48533dd039a3a221bb467e2058976553cec4f0ab6f76b790a00ef1bede", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nSubtask context (ordered by onset): Throw the remaining bottle into the dustbin.\nAtomic instruction: left arm: Grasp the remaining bottle on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nSubtask context (ordered by onset): Throw the remaining bottle into the dustbin.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.\nAtomic instruction: left arm: Grasp the remaining bottle on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Throw the remaining bottle into the dustbin.\nAtomic instruction: left arm: Grasp the remaining bottle on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the bottles and throw them into the dustbin, using handover when needed.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the remaining bottle on the table."}, "input_packs": [{"chunk": 0, "start": 230, "path": "inputs/case53-chunk0.npz", "sha256": "dde0197925ba63c2d044e73d0f85d07b35c18b542b3f261035123469765663c6"}], "input_pack": "inputs/case53-chunk0.npz", "input_image": "inputs/case53-input.png", "gt_videos": {"1": "inputs/case53-gt-1.mp4"}, "history_images": ["inputs/case53-history-0.png", "inputs/case53-history-1.png", "inputs/case53-history-2.png", "inputs/case53-history-3.png", "inputs/case53-history-4.png", "inputs/case53-history-5.png", "inputs/case53-history-6.png", "inputs/case53-history-7.png", "inputs/case53-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 53*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the bottles and throw them into the dustbin, using handover when needed.", "official_native_context": "official/texts/8467bc83bf13fff6dd5a4cf4ee956df7a2649f9ef728f00718ccac4c84b3e934.npy"}, {"id": "case54", "pair_id": "pair27", "split": "val", "start": 327, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2803, "task": "Arrange the five nesting dolls in a row from left to right, from smallest to largest.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002803", "start_seconds": 13.08, "starts": [327], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nSubtask context (ordered by onset): Place the fourth medium-sized nesting doll to continue the line.\nAtomic instruction: left arm: Place the medium nesting doll on the table to continue the line from smallest to largest."], "file": "case54.npz", "horizons": [1], "history_indices": [0, 152, 177, 202, 227, 252, 277, 302, 327], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-118.mp4", "start": 76.84, "end": 99.24000000000001, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-073.mp4", "start": 914.9200000000001, "end": 937.32, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-076.mp4", "start": 881.8000000000001, "end": 904.2, "fps": 25}}, "source_npz_sha256": "e6c9942356e033289db95bab329717109265f788e7a29feabce8da4ad3bca440", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nSubtask context (ordered by onset): Place the fourth medium-sized nesting doll to continue the line.\nAtomic instruction: left arm: Place the medium nesting doll on the table to continue the line from smallest to largest.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nSubtask context (ordered by onset): Place the fourth medium-sized nesting doll to continue the line.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nAtomic instruction: left arm: Place the medium nesting doll on the table to continue the line from smallest to largest.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the fourth medium-sized nesting doll to continue the line.\nAtomic instruction: left arm: Place the medium nesting doll on the table to continue the line from smallest to largest.", "l3": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the medium nesting doll on the table to continue the line from smallest to largest."}, "input_packs": [{"chunk": 0, "start": 327, "path": "inputs/case54-chunk0.npz", "sha256": "fcb9afd46c9ce0b5915fd6372b7bfc63d0de4c3b8673c915e6501ee459d0c731"}], "input_pack": "inputs/case54-chunk0.npz", "input_image": "inputs/case54-input.png", "gt_videos": {"1": "inputs/case54-gt-1.mp4"}, "history_images": ["inputs/case54-history-0.png", "inputs/case54-history-1.png", "inputs/case54-history-2.png", "inputs/case54-history-3.png", "inputs/case54-history-4.png", "inputs/case54-history-5.png", "inputs/case54-history-6.png", "inputs/case54-history-7.png", "inputs/case54-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 54*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Arrange the five nesting dolls in a row from left to right, from smallest to largest.", "official_native_context": "official/texts/aa4621cb19e7bd491e0154e7fe173b0690330ef791d1b6f1e62a022e5de0357d.npy"}, {"id": "case55", "pair_id": "pair27", "split": "train", "start": 456, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2800, "task": "Arrange the five nesting dolls in a row from left to right, from smallest to largest.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002800", "start_seconds": 18.24, "starts": [456], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nSubtask context (ordered by onset): Place the fifth nesting doll at the end of the current arrangement.\nAtomic instruction: right arm: Move the right arm towards the nesting dolls on the table."], "file": "case55.npz", "horizons": [1], "history_indices": [0, 281, 306, 331, 356, 381, 406, 431, 456], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-118.mp4", "start": 0.0, "end": 29.92, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-073.mp4", "start": 838.08, "end": 868.0, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-076.mp4", "start": 804.96, "end": 834.88, "fps": 25}}, "source_npz_sha256": "28233d52e95625f4c8bcf882dede63b60f5760c6eac14b1f518ac7d8dc64ecd9", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nSubtask context (ordered by onset): Place the fifth nesting doll at the end of the current arrangement.\nAtomic instruction: right arm: Move the right arm towards the nesting dolls on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nSubtask context (ordered by onset): Place the fifth nesting doll at the end of the current arrangement.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.\nAtomic instruction: right arm: Move the right arm towards the nesting dolls on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the fifth nesting doll at the end of the current arrangement.\nAtomic instruction: right arm: Move the right arm towards the nesting dolls on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Arrange the five nesting dolls in a row from left to right, from smallest to largest.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Move the right arm towards the nesting dolls on the table."}, "input_packs": [{"chunk": 0, "start": 456, "path": "inputs/case55-chunk0.npz", "sha256": "95b18b5a3b7d25d5b28ba0242996dd81b2302537593e690ebbe366f49853e2a8"}], "input_pack": "inputs/case55-chunk0.npz", "input_image": "inputs/case55-input.png", "gt_videos": {"1": "inputs/case55-gt-1.mp4"}, "history_images": ["inputs/case55-history-0.png", "inputs/case55-history-1.png", "inputs/case55-history-2.png", "inputs/case55-history-3.png", "inputs/case55-history-4.png", "inputs/case55-history-5.png", "inputs/case55-history-6.png", "inputs/case55-history-7.png", "inputs/case55-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 55*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Arrange the five nesting dolls in a row from left to right, from smallest to largest.", "official_native_context": "official/texts/aa4621cb19e7bd491e0154e7fe173b0690330ef791d1b6f1e62a022e5de0357d.npy"}, {"id": "case56", "pair_id": "pair28", "split": "val", "start": 183, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2976, "task": "Stack the three blocks with different textures.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002976", "start_seconds": 7.32, "starts": [183], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nSubtask context (ordered by onset): Grasp the white block and stack it on top of the wooden block.\nAtomic instruction: right arm: Place the white block on top of the stack to complete it."], "file": "case56.npz", "horizons": [1], "history_indices": [0, 8, 33, 58, 83, 108, 133, 158, 183], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-123.mp4", "start": 243.76, "end": 253.64, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-077.mp4", "start": 116.16000000000001, "end": 126.04, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-080.mp4", "start": 719.0, "end": 728.88, "fps": 25}}, "source_npz_sha256": "9f5d708b53757614f1e2de65c1ade784357c2767c7e2b077e048bf5753d61a87", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nSubtask context (ordered by onset): Grasp the white block and stack it on top of the wooden block.\nAtomic instruction: right arm: Place the white block on top of the stack to complete it.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nSubtask context (ordered by onset): Grasp the white block and stack it on top of the wooden block.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nAtomic instruction: right arm: Place the white block on top of the stack to complete it.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Grasp the white block and stack it on top of the wooden block.\nAtomic instruction: right arm: Place the white block on top of the stack to complete it.", "l3": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the white block on top of the stack to complete it."}, "input_packs": [{"chunk": 0, "start": 183, "path": "inputs/case56-chunk0.npz", "sha256": "9f7f35078a1fc33a3d58478acd1a01b312ead30949048d936f1c122bc67fd551"}], "input_pack": "inputs/case56-chunk0.npz", "input_image": "inputs/case56-input.png", "gt_videos": {"1": "inputs/case56-gt-1.mp4"}, "history_images": ["inputs/case56-history-0.png", "inputs/case56-history-1.png", "inputs/case56-history-2.png", "inputs/case56-history-3.png", "inputs/case56-history-4.png", "inputs/case56-history-5.png", "inputs/case56-history-6.png", "inputs/case56-history-7.png", "inputs/case56-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 56*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Stack the three blocks with different textures.", "official_native_context": "official/texts/112516ec7bba4989ae0d2745f5dfcd60773f180eccf84082eb8f5405091d9e04.npy"}, {"id": "case57", "pair_id": "pair28", "split": "train", "start": 200, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 2900, "task": "Stack the three blocks with different textures.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_002900", "start_seconds": 8.0, "starts": [200], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nSubtask context (ordered by onset): Stack the orange block onto the existing stack.\nAtomic instruction: left arm: Place the orange block onto the stack of blocks in the center."], "file": "case57.npz", "horizons": [1], "history_indices": [0, 25, 50, 75, 100, 125, 150, 175, 200], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-122.mp4", "start": 57.88, "end": 68.56, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-076.mp4", "start": 325.64, "end": 336.32, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-079.mp4", "start": 765.72, "end": 776.4, "fps": 25}}, "source_npz_sha256": "2c309d962667e202bc402beee2d8dbf2260073bf98afdcf954fbfd21534c555f", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nSubtask context (ordered by onset): Stack the orange block onto the existing stack.\nAtomic instruction: left arm: Place the orange block onto the stack of blocks in the center.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nSubtask context (ordered by onset): Stack the orange block onto the existing stack.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.\nAtomic instruction: left arm: Place the orange block onto the stack of blocks in the center.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Stack the orange block onto the existing stack.\nAtomic instruction: left arm: Place the orange block onto the stack of blocks in the center.", "l3": "Robot: robodojo_sim_arx_x5. Task: Stack the three blocks with different textures.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the orange block onto the stack of blocks in the center."}, "input_packs": [{"chunk": 0, "start": 200, "path": "inputs/case57-chunk0.npz", "sha256": "4f384b838d1e8e54d93f2c9dc5c77ca8ae5c49c0a0604ceaf59df26c7c194e44"}], "input_pack": "inputs/case57-chunk0.npz", "input_image": "inputs/case57-input.png", "gt_videos": {"1": "inputs/case57-gt-1.mp4"}, "history_images": ["inputs/case57-history-0.png", "inputs/case57-history-1.png", "inputs/case57-history-2.png", "inputs/case57-history-3.png", "inputs/case57-history-4.png", "inputs/case57-history-5.png", "inputs/case57-history-6.png", "inputs/case57-history-7.png", "inputs/case57-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 57*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Stack the three blocks with different textures.", "official_native_context": "official/texts/112516ec7bba4989ae0d2745f5dfcd60773f180eccf84082eb8f5405091d9e04.npy"}, {"id": "case58", "pair_id": "pair29", "split": "val", "start": 307, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3072, "task": "Stack the three bowls together.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003072", "start_seconds": 12.28, "starts": [307, 339, 371], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Pick up the remaining bowl from the left and place it on top of the stack.\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack.", "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Pick up the remaining bowl from the left and place it on top of the stack.\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack.", "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Pick up the remaining bowl from the left and place it on top of the stack.\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack."], "file": "case58.npz", "horizons": [1, 3], "history_indices": [0, 132, 157, 182, 207, 232, 257, 282, 307], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-125.mp4", "start": 547.1999999999999, "end": 564.4799999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-079.mp4", "start": 48.28, "end": 65.56, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-082.mp4", "start": 528.56, "end": 545.8399999999999, "fps": 25}}, "source_npz_sha256": "4cbb4c953fc77efd9d6701df9a6181fff5ea8dde8d2b4d3498be4dc02ae7f2dc", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Pick up the remaining bowl from the left and place it on top of the stack.\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Pick up the remaining bowl from the left and place it on top of the stack.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Pick up the remaining bowl from the left and place it on top of the stack.\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack.", "l3": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the held bowl onto the top of the existing stack."}, "input_packs": [{"chunk": 0, "start": 307, "path": "inputs/case58-chunk0.npz", "sha256": "70813f9de9dc63c1c221af8757a1a980d9a9dc746125303d7c550d6ad31f0c10"}, {"chunk": 1, "start": 339, "path": "inputs/case58-chunk1.npz", "sha256": "24580c49333dc9d36bbff01f3f96c6bfbd24e4aa828e682c9a5d421380056390"}, {"chunk": 2, "start": 371, "path": "inputs/case58-chunk2.npz", "sha256": "121d444287f0834faa7834d516619ea23b75f7eebbb839871788da0e0f8c3a1a"}], "input_pack": "inputs/case58-chunk0.npz", "input_image": "inputs/case58-input.png", "gt_videos": {"1": "inputs/case58-gt-1.mp4", "3": "inputs/case58-gt-3.mp4"}, "history_images": ["inputs/case58-history-0.png", "inputs/case58-history-1.png", "inputs/case58-history-2.png", "inputs/case58-history-3.png", "inputs/case58-history-4.png", "inputs/case58-history-5.png", "inputs/case58-history-6.png", "inputs/case58-history-7.png", "inputs/case58-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 58*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Stack the three bowls together.", "official_native_context": "official/texts/2c89a2177d0dd9a785e1a9dab69d94b4d8817adb3683217ad0b286903285a847.npy"}, {"id": "case59", "pair_id": "pair29", "split": "train", "start": 149, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f001", "episode": 3016, "task": "Stack the three bowls together.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003016", "start_seconds": 5.96, "starts": [149, 181, 213], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Grasp the single bowl on the left side and stack it onto the existing pile in the center.\nAtomic instruction: left arm: Place the held bowl onto the stack of bowls in the center. right arm: Remain stationary while the left arm stacks the final bowl.", "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Grasp the single bowl on the left side and stack it onto the existing pile in the center.; Retreat the arms from the stacked bowls.\nAtomic instruction: right arm: Remain stationary while the left arm stacks the final bowl.", "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Retreat the arms from the stacked bowls.\nAtomic instruction: left arm: Retreat the arm away from the stacked bowls. right arm: Remain stationary while the left arm stacks the final bowl."], "file": "case59.npz", "horizons": [1, 3], "history_indices": [0, 0, 0, 24, 49, 74, 99, 124, 149], "distinct_history_frames": 7, "history_padding": [false, true, true, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-124.mp4", "start": 67.08, "end": 76.96, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-077.mp4", "start": 559.3599999999999, "end": 569.2399999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-081.mp4", "start": 164.64000000000001, "end": 174.52, "fps": 25}}, "source_npz_sha256": "35782a22a3e22475e5f51d04a7ff237152d5ab01cd83f9ace7a6f7d40f3eb9bf", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Grasp the single bowl on the left side and stack it onto the existing pile in the center.\nAtomic instruction: left arm: Place the held bowl onto the stack of bowls in the center. right arm: Remain stationary while the left arm stacks the final bowl.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nSubtask context (ordered by onset): Grasp the single bowl on the left side and stack it onto the existing pile in the center.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.\nAtomic instruction: left arm: Place the held bowl onto the stack of bowls in the center. right arm: Remain stationary while the left arm stacks the final bowl.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Grasp the single bowl on the left side and stack it onto the existing pile in the center.\nAtomic instruction: left arm: Place the held bowl onto the stack of bowls in the center. right arm: Remain stationary while the left arm stacks the final bowl.", "l3": "Robot: robodojo_sim_arx_x5. Task: Stack the three bowls together.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Place the held bowl onto the stack of bowls in the center. right arm: Remain stationary while the left arm stacks the final bowl."}, "input_packs": [{"chunk": 0, "start": 149, "path": "inputs/case59-chunk0.npz", "sha256": "3ea3305533eecd5df47ce267d20a670833ea2c119fb1f4ad367894db79d0f192"}, {"chunk": 1, "start": 181, "path": "inputs/case59-chunk1.npz", "sha256": "eaa431e6fdf25ca0c2d110e929af483667b730681010d02c18e0454f8598b429"}, {"chunk": 2, "start": 213, "path": "inputs/case59-chunk2.npz", "sha256": "66dbe5f705bc9492899dd76ba81951cad382fb2fd6d7752ec383a22e9b880edd"}], "input_pack": "inputs/case59-chunk0.npz", "input_image": "inputs/case59-input.png", "gt_videos": {"1": "inputs/case59-gt-1.mp4", "3": "inputs/case59-gt-3.mp4"}, "history_images": ["inputs/case59-history-0.png", "inputs/case59-history-1.png", "inputs/case59-history-2.png", "inputs/case59-history-3.png", "inputs/case59-history-4.png", "inputs/case59-history-5.png", "inputs/case59-history-6.png", "inputs/case59-history-7.png", "inputs/case59-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 59*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Stack the three bowls together.", "official_native_context": "official/texts/2c89a2177d0dd9a785e1a9dab69d94b4d8817adb3683217ad0b286903285a847.npy"}, {"id": "case60", "pair_id": "pair30", "split": "val", "start": 496, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3184, "task": "Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003184", "start_seconds": 19.84, "starts": [496, 528, 560], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical slot of the stand.\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand.", "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical slot of the stand.\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand.", "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical slot of the stand.\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand."], "file": "case60.npz", "horizons": [1, 3], "history_indices": [0, 321, 346, 371, 396, 421, 446, 471, 496], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-130.mp4", "start": 23.92, "end": 49.08, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-081.mp4", "start": 772.1999999999998, "end": 797.3599999999998, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-085.mp4", "start": 439.00000000000006, "end": 464.1600000000001, "fps": 25}}, "source_npz_sha256": "25079061e68e6ffe8234914d33c418c8dec8ffbbcbc1ae15a677b910e98897bf", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical slot of the stand.\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical slot of the stand.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the closed laptop into the vertical slot of the stand.\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand.", "l3": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the laptop into the vertical slot on the right side of the stand."}, "input_packs": [{"chunk": 0, "start": 496, "path": "inputs/case60-chunk0.npz", "sha256": "e829a588eec8a1728515d2b89d5635394e2486321aa0fe09593e1bfc057c09eb"}, {"chunk": 1, "start": 528, "path": "inputs/case60-chunk1.npz", "sha256": "7f0c4781cf3dbebcece7acad5757f4a3665047ac954a646f6b517dc245923842"}, {"chunk": 2, "start": 560, "path": "inputs/case60-chunk2.npz", "sha256": "6b714974299e251b6a526d42282f1cbfd314f7a5ec1cbf046bc5802d2ad4f469"}], "input_pack": "inputs/case60-chunk0.npz", "input_image": "inputs/case60-input.png", "gt_videos": {"1": "inputs/case60-gt-1.mp4", "3": "inputs/case60-gt-3.mp4"}, "history_images": ["inputs/case60-history-0.png", "inputs/case60-history-1.png", "inputs/case60-history-2.png", "inputs/case60-history-3.png", "inputs/case60-history-4.png", "inputs/case60-history-5.png", "inputs/case60-history-6.png", "inputs/case60-history-7.png", "inputs/case60-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 60*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.", "official_native_context": "official/texts/d89781d7ef33ee3aa43f360fd29e9f8b768bf84ed56e7f0dcb0e33df45c39bba.npy"}, {"id": "case61", "pair_id": "pair30", "split": "train", "start": 425, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3100, "task": "Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003100", "start_seconds": 17.0, "starts": [425, 457, 489], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table.", "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table.", "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table."], "file": "case61.npz", "horizons": [1, 3], "history_indices": [0, 250, 275, 300, 325, 350, 375, 400, 425], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-126.mp4", "start": 517.52, "end": 539.52, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-079.mp4", "start": 583.0800000000002, "end": 605.0800000000002, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-083.mp4", "start": 299.96, "end": 321.96, "fps": 25}}, "source_npz_sha256": "1b6eeacf7a34b98ef4d430da60cf38d5c41ec4b8c29635ae7178407c560e22fb", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nSubtask context (ordered by onset): Place the closed laptop into the vertical laptop stand.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the closed laptop into the vertical laptop stand.\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the laptop into the black vertical stand on the table."}, "input_packs": [{"chunk": 0, "start": 425, "path": "inputs/case61-chunk0.npz", "sha256": "27fe80cbb5ac84113ffddcb33adbdc69d2da0cc1d0521efab95bb691e78ab3c7"}, {"chunk": 1, "start": 457, "path": "inputs/case61-chunk1.npz", "sha256": "eef92191ac4937cd55abe5366b44fc737a970ab7346689430b6968e17a67d305"}, {"chunk": 2, "start": 489, "path": "inputs/case61-chunk2.npz", "sha256": "acaed7fbc47ef38b6a79ea276efbac1e36ae68f24a05a7c5f13d2459466e9776"}], "input_pack": "inputs/case61-chunk0.npz", "input_image": "inputs/case61-input.png", "gt_videos": {"1": "inputs/case61-gt-1.mp4", "3": "inputs/case61-gt-3.mp4"}, "history_images": ["inputs/case61-history-0.png", "inputs/case61-history-1.png", "inputs/case61-history-2.png", "inputs/case61-history-3.png", "inputs/case61-history-4.png", "inputs/case61-history-5.png", "inputs/case61-history-6.png", "inputs/case61-history-7.png", "inputs/case61-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 61*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand.", "official_native_context": "official/texts/d89781d7ef33ee3aa43f360fd29e9f8b768bf84ed56e7f0dcb0e33df45c39bba.npy"}, {"id": "case62", "pair_id": "pair31", "split": "val", "start": 208, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3217, "task": "Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003217", "start_seconds": 8.32, "starts": [208], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nSubtask context (ordered by onset): Release the blue T-shaped block onto the table.\nAtomic instruction: right arm: Release the blue T-shaped block onto the table."], "file": "case62.npz", "horizons": [1], "history_indices": [0, 33, 58, 83, 108, 133, 158, 183, 208], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-131.mp4", "start": 77.16000000000001, "end": 88.20000000000002, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-082.mp4", "start": 505.2, "end": 516.24, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-086.mp4", "start": 254.95999999999995, "end": 265.99999999999994, "fps": 25}}, "source_npz_sha256": "280d2fb3b3210edcd1fc945aafcf09730fc82b7a5cb5ef8948cef95424d0ba8c", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nSubtask context (ordered by onset): Release the blue T-shaped block onto the table.\nAtomic instruction: right arm: Release the blue T-shaped block onto the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nSubtask context (ordered by onset): Release the blue T-shaped block onto the table.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nAtomic instruction: right arm: Release the blue T-shaped block onto the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Release the blue T-shaped block onto the table.\nAtomic instruction: right arm: Release the blue T-shaped block onto the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Release the blue T-shaped block onto the table."}, "input_packs": [{"chunk": 0, "start": 208, "path": "inputs/case62-chunk0.npz", "sha256": "656ca6005d6d7d523bf0639ba451a9ceaa6fae4ff79a5ce2498105301714d41d"}], "input_pack": "inputs/case62-chunk0.npz", "input_image": "inputs/case62-input.png", "gt_videos": {"1": "inputs/case62-gt-1.mp4"}, "history_images": ["inputs/case62-history-0.png", "inputs/case62-history-1.png", "inputs/case62-history-2.png", "inputs/case62-history-3.png", "inputs/case62-history-4.png", "inputs/case62-history-5.png", "inputs/case62-history-6.png", "inputs/case62-history-7.png", "inputs/case62-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 62*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.", "official_native_context": "official/texts/98e62cde53dd1978f833a737bcb6641067ebdf77cf7763b8379bfaf022b2b212.npy"}, {"id": "case63", "pair_id": "pair31", "split": "train", "start": 206, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3200, "task": "Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003200", "start_seconds": 8.24, "starts": [206], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nSubtask context (ordered by onset): Place the blue T-shaped block on the table at the original red block's position.\nAtomic instruction: right arm: Place the blue T-shaped block onto the table surface."], "file": "case63.npz", "horizons": [1], "history_indices": [0, 31, 56, 81, 106, 131, 156, 181, 206], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-130.mp4", "start": 438.28000000000003, "end": 449.16, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-082.mp4", "start": 318.12, "end": 329.0, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-086.mp4", "start": 67.88, "end": 78.75999999999999, "fps": 25}}, "source_npz_sha256": "ac1130e0619cea7b13675138b9bb196706e4d872a4b4809d88780fdd10956898", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nSubtask context (ordered by onset): Place the blue T-shaped block on the table at the original red block's position.\nAtomic instruction: right arm: Place the blue T-shaped block onto the table surface.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nSubtask context (ordered by onset): Place the blue T-shaped block on the table at the original red block's position.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.\nAtomic instruction: right arm: Place the blue T-shaped block onto the table surface.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Place the blue T-shaped block on the table at the original red block's position.\nAtomic instruction: right arm: Place the blue T-shaped block onto the table surface.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Place the blue T-shaped block onto the table surface."}, "input_packs": [{"chunk": 0, "start": 206, "path": "inputs/case63-chunk0.npz", "sha256": "b09e1036e0357476b30011e0b115244f8e6e392411caa2e4b361b1ada5132ba3"}], "input_pack": "inputs/case63-chunk0.npz", "input_image": "inputs/case63-input.png", "gt_videos": {"1": "inputs/case63-gt-1.mp4"}, "history_images": ["inputs/case63-history-0.png", "inputs/case63-history-1.png", "inputs/case63-history-2.png", "inputs/case63-history-3.png", "inputs/case63-history-4.png", "inputs/case63-history-5.png", "inputs/case63-history-6.png", "inputs/case63-history-7.png", "inputs/case63-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 63*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations.", "official_native_context": "official/texts/98e62cde53dd1978f833a737bcb6641067ebdf77cf7763b8379bfaf022b2b212.npy"}, {"id": "case64", "pair_id": "pair32", "split": "val", "start": 277, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3308, "task": "Swap the two blocks using the empty mat, pressing the button after each move.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003308", "start_seconds": 11.08, "starts": [277], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nSubtask context (ordered by onset): Move the red block from the center mat to the left mat and press the button.\nAtomic instruction: right arm: Grasp the red block currently resting on the right mat."], "file": "case64.npz", "horizons": [1], "history_indices": [0, 102, 127, 152, 177, 202, 227, 252, 277], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-132.mp4", "start": 514.1199999999999, "end": 531.6399999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-083.mp4", "start": 524.9999999999999, "end": 542.5199999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-087.mp4", "start": 249.55999999999997, "end": 267.08, "fps": 25}}, "source_npz_sha256": "22ee7dc631e7ac24398936ad444611546e965507ee15da46cc90776fa3757b7e", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nSubtask context (ordered by onset): Move the red block from the center mat to the left mat and press the button.\nAtomic instruction: right arm: Grasp the red block currently resting on the right mat.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nSubtask context (ordered by onset): Move the red block from the center mat to the left mat and press the button.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nAtomic instruction: right arm: Grasp the red block currently resting on the right mat.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Move the red block from the center mat to the left mat and press the button.\nAtomic instruction: right arm: Grasp the red block currently resting on the right mat.", "l3": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: right arm: Grasp the red block currently resting on the right mat."}, "input_packs": [{"chunk": 0, "start": 277, "path": "inputs/case64-chunk0.npz", "sha256": "28edfeb8a4c4fce083e12eb78146e753c000c6edf56151981825e1ea1029018b"}], "input_pack": "inputs/case64-chunk0.npz", "input_image": "inputs/case64-input.png", "gt_videos": {"1": "inputs/case64-gt-1.mp4"}, "history_images": ["inputs/case64-history-0.png", "inputs/case64-history-1.png", "inputs/case64-history-2.png", "inputs/case64-history-3.png", "inputs/case64-history-4.png", "inputs/case64-history-5.png", "inputs/case64-history-6.png", "inputs/case64-history-7.png", "inputs/case64-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 64*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Swap the two blocks using the empty mat, pressing the button after each move.", "official_native_context": "official/texts/936cbfea04a30520a4e7f60bdc1804e09321a73397bafd6e68ec53b837c5610e.npy"}, {"id": "case65", "pair_id": "pair32", "split": "train", "start": 277, "max_chunks": 1, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3300, "task": "Swap the two blocks using the empty mat, pressing the button after each move.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003300", "start_seconds": 11.08, "starts": [277], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nSubtask context (ordered by onset): Move the white block from the left mat to the right mat and press the button.; Move the red block from the left mat to the empty mat and press the button.\nAtomic instruction: left arm: Grasp the red block on the left side of the mat."], "file": "case65.npz", "horizons": [1], "history_indices": [0, 102, 127, 152, 177, 202, 227, 252, 277], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-132.mp4", "start": 373.9999999999999, "end": 391.7599999999999, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-083.mp4", "start": 384.8799999999999, "end": 402.6399999999999, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-087.mp4", "start": 109.44, "end": 127.2, "fps": 25}}, "source_npz_sha256": "a3bb6675aed54895a7f2aa4f9dfc70e2fa0c24c5efdf0f09f7ad0d591afd63b8", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nSubtask context (ordered by onset): Move the white block from the left mat to the right mat and press the button.; Move the red block from the left mat to the empty mat and press the button.\nAtomic instruction: left arm: Grasp the red block on the left side of the mat.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nSubtask context (ordered by onset): Move the white block from the left mat to the right mat and press the button.; Move the red block from the left mat to the empty mat and press the button.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.\nAtomic instruction: left arm: Grasp the red block on the left side of the mat.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Move the white block from the left mat to the right mat and press the button.; Move the red block from the left mat to the empty mat and press the button.\nAtomic instruction: left arm: Grasp the red block on the left side of the mat.", "l3": "Robot: robodojo_sim_arx_x5. Task: Swap the two blocks using the empty mat, pressing the button after each move.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Grasp the red block on the left side of the mat."}, "input_packs": [{"chunk": 0, "start": 277, "path": "inputs/case65-chunk0.npz", "sha256": "a758afbb03fd2d21ce39f3c836292c838662b28650253dbff1159c195a6ff98f"}], "input_pack": "inputs/case65-chunk0.npz", "input_image": "inputs/case65-input.png", "gt_videos": {"1": "inputs/case65-gt-1.mp4"}, "history_images": ["inputs/case65-history-0.png", "inputs/case65-history-1.png", "inputs/case65-history-2.png", "inputs/case65-history-3.png", "inputs/case65-history-4.png", "inputs/case65-history-5.png", "inputs/case65-history-6.png", "inputs/case65-history-7.png", "inputs/case65-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 65*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Swap the two blocks using the empty mat, pressing the button after each move.", "official_native_context": "official/texts/936cbfea04a30520a4e7f60bdc1804e09321a73397bafd6e68ec53b837c5610e.npy"}, {"id": "case66", "pair_id": "pair33", "split": "val", "start": 253, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3486, "task": "Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003486", "start_seconds": 10.12, "starts": [253, 285, 317], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Sweep the blocks into the dustpan using the broom held by the right arm.\nAtomic instruction: left arm: Hold the dustpan in place for sweeping. right arm: Release the broom onto the table.", "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Sweep the blocks into the dustpan using the broom held by the right arm.\nAtomic instruction: left arm: Hold the dustpan in place for sweeping. right arm: Release the broom onto the table.", "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Sweep the blocks into the dustpan using the broom held by the right arm.\nAtomic instruction: right arm: Release the broom onto the table."], "file": "case66.npz", "horizons": [1, 3], "history_indices": [0, 78, 103, 128, 153, 178, 203, 228, 253], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-139.mp4", "start": 241.72, "end": 256.76, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-087.mp4", "start": 191.16, "end": 206.2, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-090.mp4", "start": 830.8000000000001, "end": 845.84, "fps": 25}}, "source_npz_sha256": "f4efc85a871316d1cf9f6df337450afff5bef23354bc73b03025f268d15e8c69", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Sweep the blocks into the dustpan using the broom held by the right arm.\nAtomic instruction: left arm: Hold the dustpan in place for sweeping. right arm: Release the broom onto the table.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Sweep the blocks into the dustpan using the broom held by the right arm.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nAtomic instruction: left arm: Hold the dustpan in place for sweeping. right arm: Release the broom onto the table.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Sweep the blocks into the dustpan using the broom held by the right arm.\nAtomic instruction: left arm: Hold the dustpan in place for sweeping. right arm: Release the broom onto the table.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hold the dustpan in place for sweeping. right arm: Release the broom onto the table."}, "input_packs": [{"chunk": 0, "start": 253, "path": "inputs/case66-chunk0.npz", "sha256": "1362ed76a19ad1f7cda291dbefeeb157ba21675461b89cbc046810d089372264"}, {"chunk": 1, "start": 285, "path": "inputs/case66-chunk1.npz", "sha256": "d799d6a999c2869c539c0b99dd5b766bfb32acdbeb2808438362ed426c63bbce"}, {"chunk": 2, "start": 317, "path": "inputs/case66-chunk2.npz", "sha256": "cc6a64eeed391a4418c4f9d1b637c3c17478a9fe381e2bfc692f5273b49d6ef5"}], "input_pack": "inputs/case66-chunk0.npz", "input_image": "inputs/case66-input.png", "gt_videos": {"1": "inputs/case66-gt-1.mp4", "3": "inputs/case66-gt-3.mp4"}, "history_images": ["inputs/case66-history-0.png", "inputs/case66-history-1.png", "inputs/case66-history-2.png", "inputs/case66-history-3.png", "inputs/case66-history-4.png", "inputs/case66-history-5.png", "inputs/case66-history-6.png", "inputs/case66-history-7.png", "inputs/case66-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 66*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.", "official_native_context": "official/texts/fece4665d541aba0f92f8ee150141c47026bc39bdbc19b939b297d64f0b17fb9.npy"}, {"id": "case67", "pair_id": "pair33", "split": "train", "start": 451, "max_chunks": 3, "tree": "robodojo_lerobot_v3/robodojo_sim_arx_x5", "shard": "c000-f002", "episode": 3400, "task": "Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.", "source_id": "RoboDojo_ee_lerobot_v30_video/episode_003400", "start_seconds": 18.04, "starts": [451, 483, 515], "prompts": ["Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Release the broom onto the table and retract both arms.\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface.", "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Release the broom onto the table and retract both arms.\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface.", "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Release the broom onto the table and retract both arms.\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface."], "file": "case67.npz", "horizons": [1, 3], "history_indices": [0, 276, 301, 326, 351, 376, 401, 426, 451], "distinct_history_frames": 9, "history_padding": [false, false, false, false, false, false, false, false, false], "cameras": {"head": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_high/chunk-000/file-135.mp4", "start": 329.71999999999997, "end": 357.79999999999995, "fps": 25}, "wrist_left": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_left_wrist/chunk-000/file-085.mp4", "start": 208.07999999999998, "end": 236.15999999999997, "fps": 25}, "wrist_right": {"path": "oss://stellaredge-sz/datasets_train/robodojo_lerobot_v3/robodojo_sim_arx_x5/videos/observation.images.cam_right_wrist/chunk-000/file-088.mp4", "start": 886.44, "end": 914.5200000000001, "fps": 25}}, "source_npz_sha256": "6d9a7c135278654125e4dc93a7e0814dbfd263433c4d66ea119c47c9c9e94226", "level_prompts": {"l3l2l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Release the broom onto the table and retract both arms.\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface.", "l3l2": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nSubtask context (ordered by onset): Release the broom onto the table and retract both arms.", "l3l1": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface.", "l2l1": "Robot: robodojo_sim_arx_x5\nSubtask context (ordered by onset): Release the broom onto the table and retract both arms.\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface.", "l3": "Robot: robodojo_sim_arx_x5. Task: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.", "l1": "Robot: robodojo_sim_arx_x5\nAtomic instruction: left arm: Hold the dustpan steady on the table. right arm: Release the broom onto the table surface."}, "input_packs": [{"chunk": 0, "start": 451, "path": "inputs/case67-chunk0.npz", "sha256": "8031b6877cd2bd123f576fc76c6e097136e6e514c4eafbe581d62013bd20c938"}, {"chunk": 1, "start": 483, "path": "inputs/case67-chunk1.npz", "sha256": "76814d0006f48bad1c1ee05050d32dd1ea7067f3db2d3fbe71009112ef06498b"}, {"chunk": 2, "start": 515, "path": "inputs/case67-chunk2.npz", "sha256": "bfd082285241c01cfff7cbad834ad4f157d9ef4e787cf7f8a6a5773a19c72752"}], "input_pack": "inputs/case67-chunk0.npz", "input_image": "inputs/case67-input.png", "gt_videos": {"1": "inputs/case67-gt-1.mp4", "3": "inputs/case67-gt-3.mp4"}, "history_images": ["inputs/case67-history-0.png", "inputs/case67-history-1.png", "inputs/case67-history-2.png", "inputs/case67-history-3.png", "inputs/case67-history-4.png", "inputs/case67-history-5.png", "inputs/case67-history-6.png", "inputs/case67-history-7.png", "inputs/case67-history-8.png"], "input_contract": {"native_fps": 25, "output_fps": 6.25, "video_shape": [3, 9, 240, 416], "pixel_range": [-1, 1], "proprio_normalization": "(state-mean)/std; invalid dims zero", "action_format": "two arms x (xyz metres, quaternion wxyz, gripper)", "text_shape": [128, 4096], "text_dtype": "bfloat16 stored as uint16 bits", "context_mask": "all true", "video_sampling_steps": 10, "action_sampling_steps": 10, "sigma_shift": 3, "history_masks": "legacy matched inference API: no explicit padding mask", "generated_feedback": "fixed GT anchor and past <=initial current; later history uses preceding generated frames; proprio previous predicted action last; prompts follow recorded timeline", "rollout_seed": "42 + 67*100 + chunk_index", "official_baselines": {"matched": "same RGB, state, text embeddings and action/video seeds; sigma_shift3", "native": "same RGB/state/action seeds; official task-only DEFAULT_PROMPT, sigma_shift1, video_seed=action_seed+1000003", "normalizer": "official/dataset_stats.json", "pipeline": "official noisy video KV cache -> Action10 -> post-action VideoFull10; generated video does not condition the current action"}}, "official_native_prompt": "A video recorded from a robot's point of view executing the following instruction: Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks.", "official_native_context": "official/texts/fece4665d541aba0f92f8ee150141c47026bc39bdbc19b939b297d64f0b17fb9.npy"}], "selection": "All34 official heldout episodes plus the lowest-ID eligible training episode of the same task; equal maximum valid prompt-chain horizon within each task pair. Prefer full nonpadded history and take middle eligible start; otherwise choose latest eligible start. No selection using model outputs.", "test_definition": "The official34-episode validation split used during fine-tuning monitoring; held out from gradient updates, not a new unseen benchmark.", "horizon_counts": {"train": {"1": 34, "3": 12, "8": 2}, "val": {"1": 34, "3": 12, "8": 2}}, "summary": [{"step": 10000, "split": "train", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 21.33669008750405, "xyz_rmse_mm": 15.133828541044803}, {"step": 10000, "split": "train", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 21.064191188304942, "xyz_rmse_mm": 20.536909384342533}, {"step": 10000, "split": "train", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 23.537836257073025, "xyz_rmse_mm": 14.427791233174503}, {"step": 10000, "split": "train", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 21.33669008750405, "xyz_rmse_mm": 15.133828541044803}, {"step": 10000, "split": "train", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 19.80295646141452, "xyz_rmse_mm": 32.58383254675815}, {"step": 10000, "split": "train", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 21.71461435577991, "xyz_rmse_mm": 33.8516877964139}, {"step": 10000, "split": "val", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 20.819714411609688, "xyz_rmse_mm": 19.041115568731637}, {"step": 10000, "split": "val", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 20.415568619767413, "xyz_rmse_mm": 20.88351840696608}, {"step": 10000, "split": "val", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 23.815286892072876, "xyz_rmse_mm": 15.303531487006694}, {"step": 10000, "split": "val", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 20.819714411609688, "xyz_rmse_mm": 19.041115568731637}, {"step": 10000, "split": "val", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 18.52736062399529, "xyz_rmse_mm": 36.59400047035888}, {"step": 10000, "split": "val", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 20.72993804577335, "xyz_rmse_mm": 32.86315419245511}, {"step": 20000, "split": "train", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 21.6440501078807, "xyz_rmse_mm": 14.156131888739765}, {"step": 20000, "split": "train", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 21.246885062304262, "xyz_rmse_mm": 20.01365846566235}, {"step": 20000, "split": "train", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 23.79180179108015, "xyz_rmse_mm": 13.676226662937552}, {"step": 20000, "split": "train", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 21.6440501078807, "xyz_rmse_mm": 14.156131888739765}, {"step": 20000, "split": "train", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 19.869273699007298, "xyz_rmse_mm": 33.94922815884153}, {"step": 20000, "split": "train", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 21.77367557950864, "xyz_rmse_mm": 33.057000138796866}, {"step": 20000, "split": "val", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 21.1407357065561, "xyz_rmse_mm": 17.075198219732982}, {"step": 20000, "split": "val", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 20.588337192409497, "xyz_rmse_mm": 20.564945220636826}, {"step": 20000, "split": "val", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 23.941285045460695, "xyz_rmse_mm": 15.56595490546897}, {"step": 20000, "split": "val", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 21.1407357065561, "xyz_rmse_mm": 17.075198219732982}, {"step": 20000, "split": "val", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 19.06583342712602, "xyz_rmse_mm": 32.72318766297152}, {"step": 20000, "split": "val", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 20.68793669548594, "xyz_rmse_mm": 29.44368263706565}, {"step": "official", "split": "train", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 20.555423279326423, "xyz_rmse_mm": 3.543583587195505}, {"step": "official", "split": "train", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 20.859725404674972, "xyz_rmse_mm": 3.6852161817175024}, {"step": "official", "split": "train", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 20.975600896564274, "xyz_rmse_mm": 4.806815704796463}, {"step": "official", "split": "train", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 20.555423279326423, "xyz_rmse_mm": 3.543583587195505}, {"step": "official", "split": "train", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 18.819774809290767, "xyz_rmse_mm": 8.449725428363308}, {"step": "official", "split": "train", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 17.757704513828976, "xyz_rmse_mm": 13.854759279638529}, {"step": "official", "split": "val", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 19.881263826750136, "xyz_rmse_mm": 10.337274295606596}, {"step": "official", "split": "val", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 20.067673550593792, "xyz_rmse_mm": 11.334181239362806}, {"step": "official", "split": "val", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 20.696870985817842, "xyz_rmse_mm": 10.104542248882353}, {"step": "official", "split": "val", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 19.881263826750136, "xyz_rmse_mm": 10.337274295606596}, {"step": "official", "split": "val", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 18.185653423807167, "xyz_rmse_mm": 22.67717109255803}, {"step": "official", "split": "val", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 17.517859903743435, "xyz_rmse_mm": 34.96811690274626}, {"step": "official_native", "split": "train", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 20.03744152650523, "xyz_rmse_mm": 2.0722215079277864}, {"step": "official_native", "split": "train", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 20.300674681335128, "xyz_rmse_mm": 2.4578862818695293}, {"step": "official_native", "split": "train", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 20.41943057246241, "xyz_rmse_mm": 3.8068530411692336}, {"step": "official_native", "split": "train", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 20.03744152650523, "xyz_rmse_mm": 2.0722215079277864}, {"step": "official_native", "split": "train", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 18.35342074076301, "xyz_rmse_mm": 8.349402909516357}, {"step": "official_native", "split": "train", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 17.338235340445806, "xyz_rmse_mm": 11.212624842301011}, {"step": "official_native", "split": "val", "mode": "gt_feedback", "chunks": 1, "cases": 34, "psnr": 19.453166788178823, "xyz_rmse_mm": 9.751379512154552}, {"step": "official_native", "split": "val", "mode": "gt_feedback", "chunks": 3, "cases": 12, "psnr": 19.592896914218763, "xyz_rmse_mm": 11.167877508948246}, {"step": "official_native", "split": "val", "mode": "gt_feedback", "chunks": 8, "cases": 2, "psnr": 20.1125187239371, "xyz_rmse_mm": 9.604841499822214}, {"step": "official_native", "split": "val", "mode": "generated_feedback", "chunks": 1, "cases": 34, "psnr": 19.453166788178823, "xyz_rmse_mm": 9.751379512154552}, {"step": "official_native", "split": "val", "mode": "generated_feedback", "chunks": 3, "cases": 12, "psnr": 17.77169113231454, "xyz_rmse_mm": 22.884101238256942}, {"step": "official_native", "split": "val", "mode": "generated_feedback", "chunks": 8, "cases": 2, "psnr": 17.23761675034744, "xyz_rmse_mm": 31.70656447764486}], "responses": [{"step": 10000, "split": "train", "variant": "full_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "full_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 23.048556858317596, "median_motion_ratio": 0.9926021026014488, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "full_swap_l1", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 12.845362637616585, "median_motion_ratio": 1.0011850162318612, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l1_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 10.581205819428707, "median_motion_ratio": 0.999986279818855, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l1_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.929429910974303, "median_motion_ratio": 0.9967936819804106, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l1_only_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 20.636405025417094, "median_motion_ratio": 1.0078225303871142, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l1_only_swap", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 15.72106500637293, "median_motion_ratio": 1.0014816247816065, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l2_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 4.187104840050726, "median_motion_ratio": 1.0062242954904752, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l2l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.9747769217732745, "median_motion_ratio": 0.9973347778225026, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l3_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.660585550652981, "median_motion_ratio": 0.9995410823915234, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l3_only_edit", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 17.878941321053873, "median_motion_ratio": 1.0098886203294788, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l3_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 6.280154568872497, "median_motion_ratio": 0.9973788328251451, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l3l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.487813548417762, "median_motion_ratio": 0.9992311160427454, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "l3l2_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 4.174526539881138, "median_motion_ratio": 1.003232144484034, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "repeat", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 10000, "split": "train", "variant": "robot_only", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 14.51959094436675, "median_motion_ratio": 1.0119521142095103, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "full_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "full_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 25.98121474318973, "median_motion_ratio": 0.9597590810028926, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "full_swap_l1", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 15.836705035112304, "median_motion_ratio": 0.9998898080431333, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l1_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 9.667282817484674, "median_motion_ratio": 0.9894247848010012, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l1_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 4.314224397496539, "median_motion_ratio": 1.0020811913486116, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l1_only_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 18.69979396167978, "median_motion_ratio": 0.9784849539027913, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l1_only_swap", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 18.839669720310827, "median_motion_ratio": 0.998528659253461, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l2_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 8.454858616460115, "median_motion_ratio": 1.0019118507663516, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l2l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.175499383788145, "median_motion_ratio": 0.9998974067752205, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l3_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.5685290580868774, "median_motion_ratio": 0.9974230447859993, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l3_only_edit", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 17.977384126211025, "median_motion_ratio": 0.9984141181987037, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l3_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 14.49360021027079, "median_motion_ratio": 1.0011438460317437, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l3l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.857499709695249, "median_motion_ratio": 0.9972519320590787, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "l3l2_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 5.060470327033031, "median_motion_ratio": 1.0026348727845824, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "repeat", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 10000, "split": "val", "variant": "robot_only", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 18.86788961094092, "median_motion_ratio": 0.995189908355523, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "full_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "full_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 22.095437700045295, "median_motion_ratio": 0.9862858017934935, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "full_swap_l1", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 12.299910770653856, "median_motion_ratio": 1.0023641076356409, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l1_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 9.630249490060926, "median_motion_ratio": 0.9927570153885945, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l1_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.7506145411444938, "median_motion_ratio": 0.9971762734427299, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l1_only_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 20.933896049802357, "median_motion_ratio": 0.9963963172683623, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l1_only_swap", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 15.476841932361234, "median_motion_ratio": 1.0002725591018038, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l2_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.9974120240110684, "median_motion_ratio": 0.9996758088083707, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l2l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.1903306773850035, "median_motion_ratio": 0.9986669477570222, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l3_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.315351783171269, "median_motion_ratio": 0.9992147515795999, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l3_only_edit", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 17.62814842753441, "median_motion_ratio": 1.0049785728234366, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l3_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 5.216119283645907, "median_motion_ratio": 1.0001754993866308, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l3l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.3848843214663624, "median_motion_ratio": 0.9996589945596489, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "l3l2_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.092086539254524, "median_motion_ratio": 1.0040497071171042, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "repeat", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 20000, "split": "train", "variant": "robot_only", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 11.989240336012752, "median_motion_ratio": 1.0009494634001341, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "full_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "full_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 22.739138020209367, "median_motion_ratio": 0.9638832156900046, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "full_swap_l1", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 15.678778625870644, "median_motion_ratio": 0.9982614558212688, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l1_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 10.724929984409005, "median_motion_ratio": 0.986468678463513, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l1_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 4.846795088002075, "median_motion_ratio": 1.002873189754784, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l1_only_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 19.895521634573335, "median_motion_ratio": 0.9716467701053406, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l1_only_swap", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 19.16700390617788, "median_motion_ratio": 0.9966477829666154, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l2_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 8.983249426819384, "median_motion_ratio": 0.9989165092234081, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l2l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.9682474590064136, "median_motion_ratio": 1.0008384600293034, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l3_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.3835280542809736, "median_motion_ratio": 1.0002040298500976, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l3_only_edit", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 20.384085777229355, "median_motion_ratio": 0.9933926871772438, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l3_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 15.739795265664512, "median_motion_ratio": 0.9959234001089081, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l3l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.7509546048617373, "median_motion_ratio": 0.9978473491636544, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "l3l2_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 7.0210697025793385, "median_motion_ratio": 1.0001167271561386, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "repeat", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": 20000, "split": "val", "variant": "robot_only", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 20.098579552723095, "median_motion_ratio": 0.9991159601176431, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "full_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "full_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 9.666107536759228, "median_motion_ratio": 1.0015329918604516, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "full_swap_l1", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.3401961433102021, "median_motion_ratio": 1.0000621652872512, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l1_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.0113196570754928, "median_motion_ratio": 0.9997287040138738, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l1_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.1241362671340434, "median_motion_ratio": 1.0103766246580967, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l1_only_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 8.012865652905448, "median_motion_ratio": 0.9968880768126646, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l1_only_swap", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.3185832317874027, "median_motion_ratio": 0.9997620234406602, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l2_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.083387501058443, "median_motion_ratio": 1.0114122506669316, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l2l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.9840993704479735, "median_motion_ratio": 1.0048524770654677, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l3_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.6605901255388744, "median_motion_ratio": 0.997717731057236, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l3_only_edit", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 10.428471929010223, "median_motion_ratio": 1.0269941109727438, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l3_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.7475166673813125, "median_motion_ratio": 1.0155405874124055, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l3l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.6266416855882306, "median_motion_ratio": 1.007991717046071, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "l3l2_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.6705266830018337, "median_motion_ratio": 1.0043260075145795, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "repeat", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": "official", "split": "train", "variant": "robot_only", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 13.75673059031696, "median_motion_ratio": 1.0168264727421366, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "full_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "full_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 9.669794502448948, "median_motion_ratio": 0.9975153501490577, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "full_swap_l1", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.4075892215683673, "median_motion_ratio": 0.9998447085684382, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l1_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.3336278391173384, "median_motion_ratio": 1.0016724172755667, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l1_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.933210684301551, "median_motion_ratio": 1.0053947013965783, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l1_only_stop", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 8.03073080468868, "median_motion_ratio": 0.9969565552885629, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l1_only_swap", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.5207545856298085, "median_motion_ratio": 1.0004262820033216, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l2_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 3.6058526671053293, "median_motion_ratio": 1.0076704445666294, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l2l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.2147546491503496, "median_motion_ratio": 1.0036079341670379, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l3_conflict", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.8561667118988492, "median_motion_ratio": 1.0022251431569353, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l3_only_edit", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 10.147668981089202, "median_motion_ratio": 0.9923736513333563, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l3_only_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 2.987600994803121, "median_motion_ratio": 1.0105862147457323, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l3l1_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.8513445915339295, "median_motion_ratio": 1.0064569509132606, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "l3l2_original", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 1.8916791142306892, "median_motion_ratio": 1.0062688289622257, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "repeat", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 0.0, "median_motion_ratio": 1.0, "ratio_eligible": 68}, {"step": "official", "split": "val", "variant": "robot_only", "paired_seed_results": 68, "cases": 34, "xyz_delta_mm": 12.470725708065883, "median_motion_ratio": 1.0131743623804836, "ratio_eligible": 68}], "histories": [{"step": 10000, "split": "train", "history": "current_repeat", "cases": 34, "xyz_delta_mm": 18.87255798347349}, {"step": 10000, "split": "train", "history": "reverse_past", "cases": 34, "xyz_delta_mm": 16.840559598195423}, {"step": 10000, "split": "train", "history": "shuffle_past", "cases": 34, "xyz_delta_mm": 12.436817710883641}, {"step": 10000, "split": "val", "history": "current_repeat", "cases": 34, "xyz_delta_mm": 20.290774764041142}, {"step": 10000, "split": "val", "history": "reverse_past", "cases": 34, "xyz_delta_mm": 31.127197883667094}, {"step": 10000, "split": "val", "history": "shuffle_past", "cases": 34, "xyz_delta_mm": 13.035747018875554}, {"step": 20000, "split": "train", "history": "current_repeat", "cases": 34, "xyz_delta_mm": 18.328693866565384}, {"step": 20000, "split": "train", "history": "reverse_past", "cases": 34, "xyz_delta_mm": 17.0816768274081}, {"step": 20000, "split": "train", "history": "shuffle_past", "cases": 34, "xyz_delta_mm": 11.965817035696304}, {"step": 20000, "split": "val", "history": "current_repeat", "cases": 34, "xyz_delta_mm": 20.340384002980393}, {"step": 20000, "split": "val", "history": "reverse_past", "cases": 34, "xyz_delta_mm": 30.078892174060456}, {"step": 20000, "split": "val", "history": "shuffle_past", "cases": 34, "xyz_delta_mm": 14.045228699722108}, {"step": "official", "split": "train", "history": "current_repeat", "cases": 34, "xyz_delta_mm": 28.22182356955155}, {"step": "official", "split": "train", "history": "reverse_past", "cases": 34, "xyz_delta_mm": 34.050694108266136}, {"step": "official", "split": "train", "history": "shuffle_past", "cases": 34, "xyz_delta_mm": 28.49338698604911}, {"step": "official", "split": "val", "history": "current_repeat", "cases": 34, "xyz_delta_mm": 28.497968807213883}, {"step": "official", "split": "val", "history": "reverse_past", "cases": 34, "xyz_delta_mm": 40.37954963004107}, {"step": "official", "split": "val", "history": "shuffle_past", "cases": 34, "xyz_delta_mm": 25.587931884836188}], "seed_noise": [{"case": "case03", "step": 10000, "split": "train", "xyz_delta_mm": 4.859290551394224, "video_mae_255": 6.754718780517578, "action_max_abs": 1.7930141687393188, "video_max_abs": 208, "reference_motion_mm": 61.368875205516815, "prediction_motion_mm": 63.10272216796875, "motion_ratio": 1.0282528717797987}, {"case": "case02", "step": 10000, "split": "val", "xyz_delta_mm": 5.006591789424419, "video_mae_255": 6.817587375640869, "action_max_abs": 1.7896437644958496, "video_max_abs": 205, "reference_motion_mm": 61.6983063519001, "prediction_motion_mm": 62.567077577114105, "motion_ratio": 1.0140809574295104}, {"case": "case00", "step": 10000, "split": "val", "xyz_delta_mm": 8.85472260415554, "video_mae_255": 5.488814830780029, "action_max_abs": 0.25964006781578064, "video_max_abs": 209, "reference_motion_mm": 88.31264078617096, "prediction_motion_mm": 86.95779740810394, "motion_ratio": 0.9846585566233099}, {"case": "case01", "step": 10000, "split": "train", "xyz_delta_mm": 8.660980500280857, "video_mae_255": 5.299277305603027, "action_max_abs": 0.12782281637191772, "video_max_abs": 202, "reference_motion_mm": 72.79584556818008, "prediction_motion_mm": 69.39432770013809, "motion_ratio": 0.9532731869313043}, {"case": "case03", "step": 20000, "split": "train", "xyz_delta_mm": 4.770033527165651, "video_mae_255": 6.098296642303467, "action_max_abs": 1.804248571395874, "video_max_abs": 211, "reference_motion_mm": 60.93297526240349, "prediction_motion_mm": 60.57199835777283, "motion_ratio": 0.9940758365552291}, {"case": "case02", "step": 20000, "split": "val", "xyz_delta_mm": 4.784680437296629, "video_mae_255": 6.330921173095703, "action_max_abs": 1.802001714706421, "video_max_abs": 219, "reference_motion_mm": 60.57267263531685, "prediction_motion_mm": 59.89069864153862, "motion_ratio": 0.9887412266273256}, {"case": "case00", "step": 20000, "split": "val", "xyz_delta_mm": 6.158262956887484, "video_mae_255": 5.415081977844238, "action_max_abs": 0.18640828132629395, "video_max_abs": 221, "reference_motion_mm": 86.26893162727356, "prediction_motion_mm": 88.16435188055038, "motion_ratio": 1.0219710644089812}, {"case": "case01", "step": 20000, "split": "train", "xyz_delta_mm": 6.5837460570037365, "video_mae_255": 5.122698783874512, "action_max_abs": 0.1651044487953186, "video_max_abs": 203, "reference_motion_mm": 71.62628322839737, "prediction_motion_mm": 69.62849199771881, "motion_ratio": 0.9721081265056274}, {"case": "case05", "step": 10000, "split": "train", "xyz_delta_mm": 15.07353875786066, "video_mae_255": 10.575150489807129, "action_max_abs": 0.7141414880752563, "video_max_abs": 239, "reference_motion_mm": 63.07900696992874, "prediction_motion_mm": 65.80308079719543, "motion_ratio": 1.0431851095653}, {"case": "case04", "step": 10000, "split": "val", "xyz_delta_mm": 12.527675367891788, "video_mae_255": 5.142563343048096, "action_max_abs": 0.3142310380935669, "video_max_abs": 235, "reference_motion_mm": 65.04750996828079, "prediction_motion_mm": 72.78396934270859, "motion_ratio": 1.1189355192566224}, {"case": "case06", "step": 10000, "split": "val", "xyz_delta_mm": 9.043396450579166, "video_mae_255": 5.45402717590332, "action_max_abs": 0.08441409468650818, "video_max_abs": 223, "reference_motion_mm": 105.32421618700027, "prediction_motion_mm": 102.2670567035675, "motion_ratio": 0.970973821651757}, {"case": "case07", "step": 10000, "split": "train", "xyz_delta_mm": 5.161365028470755, "video_mae_255": 2.977247714996338, "action_max_abs": 0.1770879030227661, "video_max_abs": 242, "reference_motion_mm": 42.869582772254944, "prediction_motion_mm": 41.0386361181736, "motion_ratio": 0.957290308519953}, {"case": "case05", "step": 20000, "split": "train", "xyz_delta_mm": 13.145960867404938, "video_mae_255": 9.989330291748047, "action_max_abs": 0.7342581152915955, "video_max_abs": 239, "reference_motion_mm": 63.782766461372375, "prediction_motion_mm": 65.38401544094086, "motion_ratio": 1.0251047276310634}, {"case": "case04", "step": 20000, "split": "val", "xyz_delta_mm": 11.618522927165031, "video_mae_255": 4.580613136291504, "action_max_abs": 0.2856041193008423, "video_max_abs": 234, "reference_motion_mm": 67.96912103891373, "prediction_motion_mm": 75.71803033351898, "motion_ratio": 1.114006318989602}, {"case": "case06", "step": 20000, "split": "val", "xyz_delta_mm": 6.531308405101299, "video_mae_255": 5.080315589904785, "action_max_abs": 0.07883498072624207, "video_max_abs": 223, "reference_motion_mm": 104.49819266796112, "prediction_motion_mm": 103.34552824497223, "motion_ratio": 0.9889695276677996}, {"case": "case07", "step": 20000, "split": "train", "xyz_delta_mm": 5.841850768774748, "video_mae_255": 3.34954833984375, "action_max_abs": 0.14846092462539673, "video_max_abs": 233, "reference_motion_mm": 45.24983465671539, "prediction_motion_mm": 41.73379763960838, "motion_ratio": 0.9222972405583099}, {"case": "case08", "step": 10000, "split": "val", "xyz_delta_mm": 6.909861229360104, "video_mae_255": 4.403946399688721, "action_max_abs": 1.3699874877929688, "video_max_abs": 238, "reference_motion_mm": 75.23546367883682, "prediction_motion_mm": 70.79532742500305, "motion_ratio": 0.9409834666163858}, {"case": "case09", "step": 10000, "split": "train", "xyz_delta_mm": 11.985908262431622, "video_mae_255": 5.933545112609863, "action_max_abs": 0.09655994176864624, "video_max_abs": 231, "reference_motion_mm": 81.8466767668724, "prediction_motion_mm": 86.12328767776489, "motion_ratio": 1.052251491200721}, {"case": "case10", "step": 10000, "split": "val", "xyz_delta_mm": 1.4004220720380545, "video_mae_255": 2.1754720211029053, "action_max_abs": 0.16227439045906067, "video_max_abs": 206, "reference_motion_mm": 2.2337986156344414, "prediction_motion_mm": 1.7958201933652163, "motion_ratio": 0.8039311067686239}, {"case": "case11", "step": 10000, "split": "train", "xyz_delta_mm": 2.84767197445035, "video_mae_255": 2.6584413051605225, "action_max_abs": 0.18133699893951416, "video_max_abs": 208, "reference_motion_mm": 3.352660918608308, "prediction_motion_mm": 2.8260936960577965, "motion_ratio": 0.8429405074554662}, {"case": "case08", "step": 20000, "split": "val", "xyz_delta_mm": 8.842774666845798, "video_mae_255": 4.375576496124268, "action_max_abs": 1.4155374765396118, "video_max_abs": 243, "reference_motion_mm": 81.1711996793747, "prediction_motion_mm": 73.53457063436508, "motion_ratio": 0.9059194754398825}, {"case": "case09", "step": 20000, "split": "train", "xyz_delta_mm": 10.985714383423328, "video_mae_255": 5.708379745483398, "action_max_abs": 0.054878294467926025, "video_max_abs": 231, "reference_motion_mm": 84.79466289281845, "prediction_motion_mm": 88.09243142604828, "motion_ratio": 1.0388912275929234}, {"case": "case10", "step": 20000, "split": "val", "xyz_delta_mm": 1.0366709902882576, "video_mae_255": 2.046793222427368, "action_max_abs": 0.1381344199180603, "video_max_abs": 206, "reference_motion_mm": 2.208596793934703, "prediction_motion_mm": 2.075938507914543, "motion_ratio": 0.9399354891827839}, {"case": "case11", "step": 20000, "split": "train", "xyz_delta_mm": 1.8253334565088153, "video_mae_255": 2.5103912353515625, "action_max_abs": 0.1590675711631775, "video_max_abs": 208, "reference_motion_mm": 1.9457601010799408, "prediction_motion_mm": 2.0244556944817305, "motion_ratio": 1.0404446536641963}, {"case": "case12", "step": 10000, "split": "val", "xyz_delta_mm": 22.7058045566082, "video_mae_255": 8.148382186889648, "action_max_abs": 0.25400134921073914, "video_max_abs": 226, "reference_motion_mm": 90.75386822223663, "prediction_motion_mm": 109.2914268374443, "motion_ratio": 1.2042619116775628}, {"case": "case13", "step": 10000, "split": "train", "xyz_delta_mm": 13.67220003157854, "video_mae_255": 11.336880683898926, "action_max_abs": 0.17309340834617615, "video_max_abs": 237, "reference_motion_mm": 53.88927832245827, "prediction_motion_mm": 54.584696888923645, "motion_ratio": 1.0129045811729782}, {"case": "case13", "step": 20000, "split": "train", "xyz_delta_mm": 24.6354378759861, "video_mae_255": 13.31935977935791, "action_max_abs": 0.32221999764442444, "video_max_abs": 242, "reference_motion_mm": 52.46538296341896, "prediction_motion_mm": 72.83881306648254, "motion_ratio": 1.3883213835924686}, {"case": "case12", "step": 20000, "split": "val", "xyz_delta_mm": 23.350805044174194, "video_mae_255": 8.32028865814209, "action_max_abs": 0.41306209564208984, "video_max_abs": 229, "reference_motion_mm": 94.01366114616394, "prediction_motion_mm": 110.90128868818283, "motion_ratio": 1.1796295063518858}, {"case": "case14", "step": 10000, "split": "val", "xyz_delta_mm": 24.01084639132023, "video_mae_255": 16.42254638671875, "action_max_abs": 0.35153928399086, "video_max_abs": 249, "reference_motion_mm": 40.53713008761406, "prediction_motion_mm": 29.432402923703194, "motion_ratio": 0.7260603515860669}, {"case": "case15", "step": 10000, "split": "train", "xyz_delta_mm": 25.029856711626053, "video_mae_255": 10.786911010742188, "action_max_abs": 0.16705358028411865, "video_max_abs": 229, "reference_motion_mm": 76.04879140853882, "prediction_motion_mm": 59.68943238258362, "motion_ratio": 0.7848833791707259}, {"case": "case16", "step": 10000, "split": "val", "xyz_delta_mm": 8.519397117197514, "video_mae_255": 12.738804817199707, "action_max_abs": 1.6761760711669922, "video_max_abs": 212, "reference_motion_mm": 109.80608314275742, "prediction_motion_mm": 113.60842734575272, "motion_ratio": 1.0346278101737945}, {"case": "case17", "step": 10000, "split": "train", "xyz_delta_mm": 7.457987405359745, "video_mae_255": 9.13404655456543, "action_max_abs": 1.6705589294433594, "video_max_abs": 204, "reference_motion_mm": 99.90929067134857, "prediction_motion_mm": 101.74142569303513, "motion_ratio": 1.018337984479475}, {"case": "case14", "step": 20000, "split": "val", "xyz_delta_mm": 33.81600230932236, "video_mae_255": 17.439359664916992, "action_max_abs": 0.2944738268852234, "video_max_abs": 240, "reference_motion_mm": 49.662403762340546, "prediction_motion_mm": 27.88584865629673, "motion_ratio": 0.561508234473395}, {"case": "case15", "step": 20000, "split": "train", "xyz_delta_mm": 25.97220242023468, "video_mae_255": 10.633186340332031, "action_max_abs": 0.16927438974380493, "video_max_abs": 241, "reference_motion_mm": 79.51131463050842, "prediction_motion_mm": 61.789821833372116, "motion_ratio": 0.7771198617518948}, {"case": "case16", "step": 20000, "split": "val", "xyz_delta_mm": 7.984507828950882, "video_mae_255": 12.317995071411133, "action_max_abs": 0.14483997225761414, "video_max_abs": 220, "reference_motion_mm": 105.77497631311417, "prediction_motion_mm": 110.81548780202866, "motion_ratio": 1.0476531563948888}, {"case": "case17", "step": 20000, "split": "train", "xyz_delta_mm": 7.082600146532059, "video_mae_255": 8.92894458770752, "action_max_abs": 0.08238917589187622, "video_max_abs": 206, "reference_motion_mm": 98.3300507068634, "prediction_motion_mm": 100.58581084012985, "motion_ratio": 1.022940699379798}, {"case": "case18", "step": 10000, "split": "val", "xyz_delta_mm": 10.740941390395164, "video_mae_255": 5.205928325653076, "action_max_abs": 0.08238917589187622, "video_max_abs": 229, "reference_motion_mm": 44.90366578102112, "prediction_motion_mm": 44.79418322443962, "motion_ratio": 0.997561834770564}, {"case": "case19", "step": 10000, "split": "train", "xyz_delta_mm": 25.84696002304554, "video_mae_255": 5.438304901123047, "action_max_abs": 0.15159228444099426, "video_max_abs": 193, "reference_motion_mm": 57.44068697094917, "prediction_motion_mm": 43.77755522727966, "motion_ratio": 0.7621349523451959}, {"case": "case20", "step": 10000, "split": "val", "xyz_delta_mm": 27.223043143749237, "video_mae_255": 9.22502613067627, "action_max_abs": 0.10151070356369019, "video_max_abs": 240, "reference_motion_mm": 116.41017347574234, "prediction_motion_mm": 99.72706437110901, "motion_ratio": 0.8566868461190827}, {"case": "case21", "step": 10000, "split": "train", "xyz_delta_mm": 16.47402159869671, "video_mae_255": 10.692731857299805, "action_max_abs": 0.1300877332687378, "video_max_abs": 237, "reference_motion_mm": 174.72882568836212, "prediction_motion_mm": 166.4527952671051, "motion_ratio": 0.9526350023320265}, {"case": "case18", "step": 20000, "split": "val", "xyz_delta_mm": 10.449454188346863, "video_mae_255": 4.763973236083984, "action_max_abs": 0.06957307457923889, "video_max_abs": 222, "reference_motion_mm": 43.10205951333046, "prediction_motion_mm": 43.97868365049362, "motion_ratio": 1.0203383352689224}, {"case": "case19", "step": 20000, "split": "train", "xyz_delta_mm": 13.513152487576008, "video_mae_255": 4.682894706726074, "action_max_abs": 0.15743142366409302, "video_max_abs": 199, "reference_motion_mm": 63.282981514930725, "prediction_motion_mm": 60.01363694667816, "motion_ratio": 0.9483376969607349}, {"case": "case20", "step": 20000, "split": "val", "xyz_delta_mm": 22.342631593346596, "video_mae_255": 8.827874183654785, "action_max_abs": 0.09791743755340576, "video_max_abs": 240, "reference_motion_mm": 115.0735691189766, "prediction_motion_mm": 100.80088675022125, "motion_ratio": 0.8759690650248402}, {"case": "case21", "step": 20000, "split": "train", "xyz_delta_mm": 13.9161953702569, "video_mae_255": 9.507036209106445, "action_max_abs": 0.14819270372390747, "video_max_abs": 239, "reference_motion_mm": 176.3867735862732, "prediction_motion_mm": 172.39578068256378, "motion_ratio": 0.9773736271571555}, {"case": "case22", "step": 10000, "split": "val", "xyz_delta_mm": 7.8741321340203285, "video_mae_255": 5.890917778015137, "action_max_abs": 0.18272560834884644, "video_max_abs": 218, "reference_motion_mm": 24.0292027592659, "prediction_motion_mm": 25.7205069065094, "motion_ratio": 1.0703853625185844}, {"case": "case23", "step": 10000, "split": "train", "xyz_delta_mm": 5.315748974680901, "video_mae_255": 5.748421669006348, "action_max_abs": 1.5013138055801392, "video_max_abs": 209, "reference_motion_mm": 23.011012002825737, "prediction_motion_mm": 24.25680123269558, "motion_ratio": 1.0541388283886368}, {"case": "case24", "step": 10000, "split": "val", "xyz_delta_mm": 10.25972981005907, "video_mae_255": 3.9497148990631104, "action_max_abs": 0.03938581049442291, "video_max_abs": 207, "reference_motion_mm": 48.292871564626694, "prediction_motion_mm": 45.23910954594612, "motion_ratio": 0.936765780958087}, {"case": "case25", "step": 10000, "split": "train", "xyz_delta_mm": 5.750259850174189, "video_mae_255": 3.646623373031616, "action_max_abs": 0.04827128350734711, "video_max_abs": 226, "reference_motion_mm": 48.86201769113541, "prediction_motion_mm": 51.393166184425354, "motion_ratio": 1.0518019642432643}, {"case": "case22", "step": 20000, "split": "val", "xyz_delta_mm": 7.040530908852816, "video_mae_255": 5.636440753936768, "action_max_abs": 0.23103240132331848, "video_max_abs": 220, "reference_motion_mm": 27.076801285147667, "prediction_motion_mm": 28.079260140657425, "motion_ratio": 1.0370227947146635}, {"case": "case23", "step": 20000, "split": "train", "xyz_delta_mm": 4.7296518459916115, "video_mae_255": 5.518006324768066, "action_max_abs": 1.4830049276351929, "video_max_abs": 204, "reference_motion_mm": 26.553843170404434, "prediction_motion_mm": 25.126894935965538, "motion_ratio": 0.9462620824683751}, {"case": "case24", "step": 20000, "split": "val", "xyz_delta_mm": 8.294573053717613, "video_mae_255": 3.7658331394195557, "action_max_abs": 0.03978133201599121, "video_max_abs": 209, "reference_motion_mm": 49.27403852343559, "prediction_motion_mm": 47.09241911768913, "motion_ratio": 0.9557247696531137}, {"case": "case25", "step": 20000, "split": "train", "xyz_delta_mm": 5.0424933433532715, "video_mae_255": 3.514774799346924, "action_max_abs": 0.035415105521678925, "video_max_abs": 224, "reference_motion_mm": 50.60531944036484, "prediction_motion_mm": 52.84758284687996, "motion_ratio": 1.0443088479889449}, {"case": "case26", "step": 10000, "split": "val", "xyz_delta_mm": 12.473706156015396, "video_mae_255": 10.692045211791992, "action_max_abs": 1.8514331579208374, "video_max_abs": 245, "reference_motion_mm": 111.54628545045853, "prediction_motion_mm": 111.41840368509293, "motion_ratio": 0.9988535542456732}, {"case": "case27", "step": 10000, "split": "train", "xyz_delta_mm": 5.8629922568798065, "video_mae_255": 10.947822570800781, "action_max_abs": 1.343642234802246, "video_max_abs": 244, "reference_motion_mm": 49.012042582035065, "prediction_motion_mm": 47.288134694099426, "motion_ratio": 0.9648268507673352}, {"case": "case28", "step": 10000, "split": "val", "xyz_delta_mm": 31.914379447698593, "video_mae_255": 21.163131713867188, "action_max_abs": 0.24681472778320312, "video_max_abs": 242, "reference_motion_mm": 100.95492005348206, "prediction_motion_mm": 91.71055257320404, "motion_ratio": 0.9084307384386942}, {"case": "case29", "step": 10000, "split": "train", "xyz_delta_mm": 22.172708064317703, "video_mae_255": 10.912333488464355, "action_max_abs": 0.14097702503204346, "video_max_abs": 224, "reference_motion_mm": 50.77080428600311, "prediction_motion_mm": 60.6192909181118, "motion_ratio": 1.1939793306529083}, {"case": "case26", "step": 20000, "split": "val", "xyz_delta_mm": 6.235072389245033, "video_mae_255": 9.878087043762207, "action_max_abs": 1.835705041885376, "video_max_abs": 244, "reference_motion_mm": 110.12531816959381, "prediction_motion_mm": 109.81687903404236, "motion_ratio": 0.9971991986885663}, {"case": "case27", "step": 20000, "split": "train", "xyz_delta_mm": 3.9408234879374504, "video_mae_255": 10.19118881225586, "action_max_abs": 1.4102981090545654, "video_max_abs": 245, "reference_motion_mm": 50.300393253564835, "prediction_motion_mm": 48.721250146627426, "motion_ratio": 0.9686057502775987}, {"case": "case28", "step": 20000, "split": "val", "xyz_delta_mm": 40.526196360588074, "video_mae_255": 22.38998794555664, "action_max_abs": 0.2533276081085205, "video_max_abs": 243, "reference_motion_mm": 102.39552706480026, "prediction_motion_mm": 94.02796626091003, "motion_ratio": 0.918281969498581}, {"case": "case29", "step": 20000, "split": "train", "xyz_delta_mm": 23.12076836824417, "video_mae_255": 10.81105899810791, "action_max_abs": 0.13874956965446472, "video_max_abs": 228, "reference_motion_mm": 53.03381010890007, "prediction_motion_mm": 60.4938305914402, "motion_ratio": 1.1406653692657884}, {"case": "case30", "step": 10000, "split": "val", "xyz_delta_mm": 17.153039574623108, "video_mae_255": 10.599491119384766, "action_max_abs": 0.15027418732643127, "video_max_abs": 245, "reference_motion_mm": 46.05508595705032, "prediction_motion_mm": 54.37678471207619, "motion_ratio": 1.1806901145030202}, {"case": "case31", "step": 10000, "split": "train", "xyz_delta_mm": 17.263252288103104, "video_mae_255": 7.231408596038818, "action_max_abs": 0.5605563521385193, "video_max_abs": 243, "reference_motion_mm": 63.259296119213104, "prediction_motion_mm": 54.774463176727295, "motion_ratio": 0.8658721569317493}, {"case": "case32", "step": 10000, "split": "val", "xyz_delta_mm": 22.04960398375988, "video_mae_255": 19.92990493774414, "action_max_abs": 0.16510450839996338, "video_max_abs": 255, "reference_motion_mm": 51.590997725725174, "prediction_motion_mm": 49.11154881119728, "motion_ratio": 0.9519402798195634}, {"case": "case33", "step": 10000, "split": "train", "xyz_delta_mm": 15.69405384361744, "video_mae_255": 12.288717269897461, "action_max_abs": 0.10939870774745941, "video_max_abs": 228, "reference_motion_mm": 19.574541598558426, "prediction_motion_mm": 15.959739685058594, "motion_ratio": 0.8153314653475182}, {"case": "case30", "step": 20000, "split": "val", "xyz_delta_mm": 15.442360192537308, "video_mae_255": 10.926141738891602, "action_max_abs": 0.28227537870407104, "video_max_abs": 246, "reference_motion_mm": 50.712522119283676, "prediction_motion_mm": 56.30890280008316, "motion_ratio": 1.1103550059615637}, {"case": "case31", "step": 20000, "split": "train", "xyz_delta_mm": 15.788722783327103, "video_mae_255": 6.809448719024658, "action_max_abs": 0.5938435196876526, "video_max_abs": 247, "reference_motion_mm": 62.727876007556915, "prediction_motion_mm": 59.9198155105114, "motion_ratio": 0.9552342486981829}, {"case": "case32", "step": 20000, "split": "val", "xyz_delta_mm": 22.747298702597618, "video_mae_255": 20.312854766845703, "action_max_abs": 0.10119307041168213, "video_max_abs": 255, "reference_motion_mm": 52.040550857782364, "prediction_motion_mm": 51.16782337427139, "motion_ratio": 0.9832298569264576}, {"case": "case33", "step": 20000, "split": "train", "xyz_delta_mm": 14.170972630381584, "video_mae_255": 11.667192459106445, "action_max_abs": 0.06753292679786682, "video_max_abs": 225, "reference_motion_mm": 18.764305859804153, "prediction_motion_mm": 11.03737112134695, "motion_ratio": 0.5882110003861422}, {"case": "case34", "step": 10000, "split": "val", "xyz_delta_mm": 30.9965331107378, "video_mae_255": 7.580082416534424, "action_max_abs": 1.5795600414276123, "video_max_abs": 225, "reference_motion_mm": 108.46061259508133, "prediction_motion_mm": 96.75648808479309, "motion_ratio": 0.8920887110053164}, {"case": "case35", "step": 10000, "split": "train", "xyz_delta_mm": 19.33613233268261, "video_mae_255": 10.024009704589844, "action_max_abs": 0.1529885083436966, "video_max_abs": 211, "reference_motion_mm": 75.82493126392365, "prediction_motion_mm": 76.15761458873749, "motion_ratio": 1.0043875189765208}, {"case": "case36", "step": 10000, "split": "val", "xyz_delta_mm": 15.017474070191383, "video_mae_255": 12.974401473999023, "action_max_abs": 0.11452865600585938, "video_max_abs": 247, "reference_motion_mm": 57.62761831283569, "prediction_motion_mm": 60.584791004657745, "motion_ratio": 1.051315198829992}, {"case": "case34", "step": 20000, "split": "val", "xyz_delta_mm": 29.604926705360413, "video_mae_255": 7.0571722984313965, "action_max_abs": 1.648090124130249, "video_max_abs": 218, "reference_motion_mm": 111.28665506839752, "prediction_motion_mm": 110.38258671760559, "motion_ratio": 0.9918762195679591}, {"case": "case37", "step": 10000, "split": "train", "xyz_delta_mm": 13.462206348776817, "video_mae_255": 13.103846549987793, "action_max_abs": 0.05617213249206543, "video_max_abs": 254, "reference_motion_mm": 60.55048853158951, "prediction_motion_mm": 62.93655186891556, "motion_ratio": 1.0394061781364692}, {"case": "case35", "step": 20000, "split": "train", "xyz_delta_mm": 15.807097777724266, "video_mae_255": 9.534019470214844, "action_max_abs": 0.16402259469032288, "video_max_abs": 246, "reference_motion_mm": 84.69758927822113, "prediction_motion_mm": 86.34728938341141, "motion_ratio": 1.0194775331771393}, {"case": "case36", "step": 20000, "split": "val", "xyz_delta_mm": 9.824343025684357, "video_mae_255": 11.861042022705078, "action_max_abs": 1.9547898769378662, "video_max_abs": 247, "reference_motion_mm": 61.831310391426086, "prediction_motion_mm": 62.63956427574158, "motion_ratio": 1.0130719190519948}, {"case": "case37", "step": 20000, "split": "train", "xyz_delta_mm": 8.919290266931057, "video_mae_255": 11.4197998046875, "action_max_abs": 1.9952337741851807, "video_max_abs": 255, "reference_motion_mm": 65.73782116174698, "prediction_motion_mm": 65.33642113208771, "motion_ratio": 0.9938939255581405}, {"case": "case38", "step": 10000, "split": "val", "xyz_delta_mm": 3.9908927865326405, "video_mae_255": 4.488595485687256, "action_max_abs": 0.08717221021652222, "video_max_abs": 229, "reference_motion_mm": 51.11810937523842, "prediction_motion_mm": 51.40717327594757, "motion_ratio": 1.0056548237844096}, {"case": "case39", "step": 10000, "split": "train", "xyz_delta_mm": 10.224399156868458, "video_mae_255": 5.774147033691406, "action_max_abs": 0.030624210834503174, "video_max_abs": 237, "reference_motion_mm": 145.2651470899582, "prediction_motion_mm": 141.83464646339417, "motion_ratio": 0.9763845582007388}, {"case": "case40", "step": 10000, "split": "val", "xyz_delta_mm": 9.9800331518054, "video_mae_255": 6.334898948669434, "action_max_abs": 0.05567365884780884, "video_max_abs": 245, "reference_motion_mm": 68.32008808851242, "prediction_motion_mm": 69.91571187973022, "motion_ratio": 1.023355119055915}, {"case": "case38", "step": 20000, "split": "val", "xyz_delta_mm": 4.569541662931442, "video_mae_255": 4.7358503341674805, "action_max_abs": 0.05900883674621582, "video_max_abs": 233, "reference_motion_mm": 51.826100796461105, "prediction_motion_mm": 52.406154572963715, "motion_ratio": 1.0111923098127849}, {"case": "case41", "step": 10000, "split": "train", "xyz_delta_mm": 5.715950392186642, "video_mae_255": 3.5198066234588623, "action_max_abs": 0.1148408055305481, "video_max_abs": 235, "reference_motion_mm": 51.715925335884094, "prediction_motion_mm": 49.49517175555229, "motion_ratio": 0.9570586126824866}, {"case": "case39", "step": 20000, "split": "train", "xyz_delta_mm": 7.58680235594511, "video_mae_255": 5.297281265258789, "action_max_abs": 0.0719003677368164, "video_max_abs": 230, "reference_motion_mm": 145.08196711540222, "prediction_motion_mm": 143.45046877861023, "motion_ratio": 0.9887546442247075}, {"case": "case40", "step": 20000, "split": "val", "xyz_delta_mm": 11.342337355017662, "video_mae_255": 6.273787021636963, "action_max_abs": 0.04793351888656616, "video_max_abs": 242, "reference_motion_mm": 67.74280965328217, "prediction_motion_mm": 71.79796695709229, "motion_ratio": 1.0598610734418166}, {"case": "case41", "step": 20000, "split": "train", "xyz_delta_mm": 4.997202195227146, "video_mae_255": 3.4560530185699463, "action_max_abs": 0.10785049200057983, "video_max_abs": 231, "reference_motion_mm": 48.06036129593849, "prediction_motion_mm": 48.57064411044121, "motion_ratio": 1.010617540125439}, {"case": "case42", "step": 10000, "split": "val", "xyz_delta_mm": 8.0103212967515, "video_mae_255": 4.566621780395508, "action_max_abs": 0.10785046219825745, "video_max_abs": 238, "reference_motion_mm": 65.69024920463562, "prediction_motion_mm": 65.33186137676239, "motion_ratio": 0.9945442766283806}, {"case": "case43", "step": 10000, "split": "train", "xyz_delta_mm": 8.920122869312763, "video_mae_255": 4.263001441955566, "action_max_abs": 0.09119552373886108, "video_max_abs": 204, "reference_motion_mm": 55.51913380622864, "prediction_motion_mm": 50.79815536737442, "motion_ratio": 0.9149666409542475}, {"case": "case44", "step": 10000, "split": "val", "xyz_delta_mm": 50.7056899368763, "video_mae_255": 8.169696807861328, "action_max_abs": 0.21633189916610718, "video_max_abs": 245, "reference_motion_mm": 45.67787051200867, "prediction_motion_mm": 39.170894771814346, "motion_ratio": 0.8575464296549542}, {"case": "case42", "step": 20000, "split": "val", "xyz_delta_mm": 7.2044250555336475, "video_mae_255": 4.340359210968018, "action_max_abs": 0.10252454876899719, "video_max_abs": 234, "reference_motion_mm": 68.24090331792831, "prediction_motion_mm": 65.09941816329956, "motion_ratio": 0.9539647776936239}, {"case": "case45", "step": 10000, "split": "train", "xyz_delta_mm": 24.19254742562771, "video_mae_255": 6.5817365646362305, "action_max_abs": 0.11008107662200928, "video_max_abs": 229, "reference_motion_mm": 52.215371280908585, "prediction_motion_mm": 71.55230641365051, "motion_ratio": 1.3703303195664924}, {"case": "case43", "step": 20000, "split": "train", "xyz_delta_mm": 3.533650888130069, "video_mae_255": 3.9110450744628906, "action_max_abs": 0.10326555371284485, "video_max_abs": 206, "reference_motion_mm": 52.600570023059845, "prediction_motion_mm": 53.32555249333382, "motion_ratio": 1.0137827873339802}, {"case": "case44", "step": 20000, "split": "val", "xyz_delta_mm": 42.450595647096634, "video_mae_255": 7.734443664550781, "action_max_abs": 0.14733630418777466, "video_max_abs": 237, "reference_motion_mm": 47.09170386195183, "prediction_motion_mm": 40.9141406416893, "motion_ratio": 0.8688184390530463}, {"case": "case45", "step": 20000, "split": "train", "xyz_delta_mm": 38.63703832030296, "video_mae_255": 7.279902935028076, "action_max_abs": 0.31388452649116516, "video_max_abs": 225, "reference_motion_mm": 48.51716756820679, "prediction_motion_mm": 76.34203881025314, "motion_ratio": 1.573505681322583}, {"case": "case46", "step": 10000, "split": "val", "xyz_delta_mm": 5.333492532372475, "video_mae_255": 4.779666423797607, "action_max_abs": 0.05749279260635376, "video_max_abs": 191, "reference_motion_mm": 12.309684418141842, "prediction_motion_mm": 13.739705085754395, "motion_ratio": 1.1161703760256443}, {"case": "case47", "step": 10000, "split": "train", "xyz_delta_mm": 11.654643341898918, "video_mae_255": 4.89067268371582, "action_max_abs": 0.04200589656829834, "video_max_abs": 198, "reference_motion_mm": 16.68434776365757, "prediction_motion_mm": 11.061806231737137, "motion_ratio": 0.6630050145461694}, {"case": "case48", "step": 10000, "split": "val", "xyz_delta_mm": 3.4637716598808765, "video_mae_255": 2.618743419647217, "action_max_abs": 0.021579742431640625, "video_max_abs": 206, "reference_motion_mm": 11.764817871153355, "prediction_motion_mm": 11.957269161939621, "motion_ratio": 1.016358204002303}, {"case": "case46", "step": 20000, "split": "val", "xyz_delta_mm": 4.182498902082443, "video_mae_255": 4.550407886505127, "action_max_abs": 0.05030620098114014, "video_max_abs": 196, "reference_motion_mm": 9.569013491272926, "prediction_motion_mm": 11.382454074919224, "motion_ratio": 1.1895117595246554}, {"case": "case49", "step": 10000, "split": "train", "xyz_delta_mm": 4.705072846263647, "video_mae_255": 2.3251492977142334, "action_max_abs": 0.020427465438842773, "video_max_abs": 201, "reference_motion_mm": 7.926419377326965, "prediction_motion_mm": 7.553516421467066, "motion_ratio": 0.9529544251813669}, {"case": "case47", "step": 20000, "split": "train", "xyz_delta_mm": 16.22096076607704, "video_mae_255": 6.062294006347656, "action_max_abs": 0.046930789947509766, "video_max_abs": 213, "reference_motion_mm": 13.18307127803564, "prediction_motion_mm": 11.19556836783886, "motion_ratio": 0.8492382489421744}, {"case": "case48", "step": 20000, "split": "val", "xyz_delta_mm": 4.002254921942949, "video_mae_255": 2.7211713790893555, "action_max_abs": 0.02063697576522827, "video_max_abs": 208, "reference_motion_mm": 12.737891636788845, "prediction_motion_mm": 12.989785522222519, "motion_ratio": 1.0197751631601393}, {"case": "case49", "step": 20000, "split": "train", "xyz_delta_mm": 5.115297622978687, "video_mae_255": 2.429959535598755, "action_max_abs": 0.021108388900756836, "video_max_abs": 203, "reference_motion_mm": 8.422354236245155, "prediction_motion_mm": 8.105733431875706, "motion_ratio": 0.962407090050085}, {"case": "case50", "step": 10000, "split": "val", "xyz_delta_mm": 14.158107340335846, "video_mae_255": 5.489195823669434, "action_max_abs": 0.05336350202560425, "video_max_abs": 223, "reference_motion_mm": 8.592493832111359, "prediction_motion_mm": 15.245051123201847, "motion_ratio": 1.7742289283035555}, {"case": "case51", "step": 10000, "split": "train", "xyz_delta_mm": 25.101354345679283, "video_mae_255": 3.932450771331787, "action_max_abs": 0.07732772827148438, "video_max_abs": 236, "reference_motion_mm": 25.973206385970116, "prediction_motion_mm": 10.395056568086147, "motion_ratio": 0.4002223065420687}, {"case": "case52", "step": 10000, "split": "val", "xyz_delta_mm": 17.806757241487503, "video_mae_255": 11.60663890838623, "action_max_abs": 0.4975520670413971, "video_max_abs": 235, "reference_motion_mm": 133.20784270763397, "prediction_motion_mm": 135.9059065580368, "motion_ratio": 1.0202545420416766}, {"case": "case53", "step": 10000, "split": "train", "xyz_delta_mm": 13.527734205126762, "video_mae_255": 10.672208786010742, "action_max_abs": 0.34332430362701416, "video_max_abs": 239, "reference_motion_mm": 32.555270940065384, "prediction_motion_mm": 31.370125710964203, "motion_ratio": 0.9635959033705157}, {"case": "case50", "step": 20000, "split": "val", "xyz_delta_mm": 13.216651044785976, "video_mae_255": 5.233017921447754, "action_max_abs": 0.06768739223480225, "video_max_abs": 233, "reference_motion_mm": 8.608851581811905, "prediction_motion_mm": 11.591242626309395, "motion_ratio": 1.346433088798795}, {"case": "case51", "step": 20000, "split": "train", "xyz_delta_mm": 26.096777990460396, "video_mae_255": 3.9619224071502686, "action_max_abs": 0.07567363977432251, "video_max_abs": 225, "reference_motion_mm": 27.068085968494415, "prediction_motion_mm": 10.417155921459198, "motion_ratio": 0.3848501121794916}, {"case": "case52", "step": 20000, "split": "val", "xyz_delta_mm": 27.21668966114521, "video_mae_255": 12.819482803344727, "action_max_abs": 0.5176687240600586, "video_max_abs": 232, "reference_motion_mm": 135.7959359884262, "prediction_motion_mm": 139.9296671152115, "motion_ratio": 1.0304407572781678}, {"case": "case53", "step": 20000, "split": "train", "xyz_delta_mm": 12.530714273452759, "video_mae_255": 9.86771297454834, "action_max_abs": 0.40233319997787476, "video_max_abs": 222, "reference_motion_mm": 31.684301793575287, "prediction_motion_mm": 31.095346435904503, "motion_ratio": 0.9814117615244339}, {"case": "case54", "step": 10000, "split": "val", "xyz_delta_mm": 9.253970347344875, "video_mae_255": 4.2018723487854, "action_max_abs": 0.15041393041610718, "video_max_abs": 224, "reference_motion_mm": 145.89619636535645, "prediction_motion_mm": 140.67068696022034, "motion_ratio": 0.9641833746504928}, {"case": "case55", "step": 10000, "split": "train", "xyz_delta_mm": 11.795425787568092, "video_mae_255": 7.73665714263916, "action_max_abs": 0.0682726800441742, "video_max_abs": 226, "reference_motion_mm": 105.68448156118393, "prediction_motion_mm": 106.38011246919632, "motion_ratio": 1.0065821480858537}, {"case": "case56", "step": 10000, "split": "val", "xyz_delta_mm": 6.174944341182709, "video_mae_255": 3.2322206497192383, "action_max_abs": 0.09054115414619446, "video_max_abs": 215, "reference_motion_mm": 73.29229265451431, "prediction_motion_mm": 70.09408622980118, "motion_ratio": 0.9563636733293518}, {"case": "case57", "step": 10000, "split": "train", "xyz_delta_mm": 6.5627144649624825, "video_mae_255": 4.337466239929199, "action_max_abs": 0.1502043902873993, "video_max_abs": 208, "reference_motion_mm": 124.16840344667435, "prediction_motion_mm": 124.48923289775848, "motion_ratio": 1.0025838252097838}, {"case": "case54", "step": 20000, "split": "val", "xyz_delta_mm": 7.2256713174283504, "video_mae_255": 4.247467041015625, "action_max_abs": 0.1712430715560913, "video_max_abs": 213, "reference_motion_mm": 147.09903299808502, "prediction_motion_mm": 143.69288086891174, "motion_ratio": 0.9768444968008891}, {"case": "case55", "step": 20000, "split": "train", "xyz_delta_mm": 9.241235442459583, "video_mae_255": 7.61778450012207, "action_max_abs": 0.05659446120262146, "video_max_abs": 212, "reference_motion_mm": 107.67634212970734, "prediction_motion_mm": 105.80186545848846, "motion_ratio": 0.9825915643664709}, {"case": "case56", "step": 20000, "split": "val", "xyz_delta_mm": 5.476026795804501, "video_mae_255": 3.336080312728882, "action_max_abs": 0.039944589138031006, "video_max_abs": 213, "reference_motion_mm": 74.5394304394722, "prediction_motion_mm": 70.9589347243309, "motion_ratio": 0.9519650781602262}, {"case": "case57", "step": 20000, "split": "train", "xyz_delta_mm": 10.397601872682571, "video_mae_255": 4.47537088394165, "action_max_abs": 0.1797088384628296, "video_max_abs": 208, "reference_motion_mm": 122.26887047290802, "prediction_motion_mm": 124.9370276927948, "motion_ratio": 1.021822048486806}, {"case": "case58", "step": 10000, "split": "val", "xyz_delta_mm": 5.488485097885132, "video_mae_255": 4.836658477783203, "action_max_abs": 0.058216601610183716, "video_max_abs": 206, "reference_motion_mm": 55.64777925610542, "prediction_motion_mm": 57.83575400710106, "motion_ratio": 1.0393182761332849}, {"case": "case59", "step": 10000, "split": "train", "xyz_delta_mm": 17.064563930034637, "video_mae_255": 8.466052055358887, "action_max_abs": 0.4841409921646118, "video_max_abs": 238, "reference_motion_mm": 59.18610095977783, "prediction_motion_mm": 50.31557008624077, "motion_ratio": 0.8501247635899285}, {"case": "case60", "step": 10000, "split": "val", "xyz_delta_mm": 23.758964613080025, "video_mae_255": 21.32270050048828, "action_max_abs": 1.6626948118209839, "video_max_abs": 235, "reference_motion_mm": 63.015006482601166, "prediction_motion_mm": 64.60673362016678, "motion_ratio": 1.025259493355842}, {"case": "case61", "step": 10000, "split": "train", "xyz_delta_mm": 20.549656823277473, "video_mae_255": 12.544095993041992, "action_max_abs": 0.1811186671257019, "video_max_abs": 247, "reference_motion_mm": 91.11180901527405, "prediction_motion_mm": 95.11897712945938, "motion_ratio": 1.0439807765589812}, {"case": "case58", "step": 20000, "split": "val", "xyz_delta_mm": 5.234279204159975, "video_mae_255": 4.637275218963623, "action_max_abs": 0.04948410391807556, "video_max_abs": 219, "reference_motion_mm": 55.2908219397068, "prediction_motion_mm": 57.79671669006348, "motion_ratio": 1.0453220744862373}, {"case": "case59", "step": 20000, "split": "train", "xyz_delta_mm": 16.849661245942116, "video_mae_255": 8.283096313476562, "action_max_abs": 0.4559776186943054, "video_max_abs": 238, "reference_motion_mm": 59.0459369122982, "prediction_motion_mm": 51.62464827299118, "motion_ratio": 0.8743133054128691}, {"case": "case60", "step": 20000, "split": "val", "xyz_delta_mm": 20.08817531168461, "video_mae_255": 20.47930335998535, "action_max_abs": 1.5818068981170654, "video_max_abs": 242, "reference_motion_mm": 68.16570460796356, "prediction_motion_mm": 64.99320268630981, "motion_ratio": 0.9534589726622863}, {"case": "case61", "step": 20000, "split": "train", "xyz_delta_mm": 19.002538174390793, "video_mae_255": 12.073037147521973, "action_max_abs": 0.1396637111902237, "video_max_abs": 246, "reference_motion_mm": 91.71884506940842, "prediction_motion_mm": 98.78849983215332, "motion_ratio": 1.077079631316715}, {"case": "case62", "step": 10000, "split": "val", "xyz_delta_mm": 21.62899076938629, "video_mae_255": 7.94291353225708, "action_max_abs": 0.11717090010643005, "video_max_abs": 220, "reference_motion_mm": 123.15793335437775, "prediction_motion_mm": 122.24636226892471, "motion_ratio": 0.9925983567552238}, {"case": "case63", "step": 10000, "split": "train", "xyz_delta_mm": 19.331131130456924, "video_mae_255": 7.548034191131592, "action_max_abs": 0.11717090010643005, "video_max_abs": 216, "reference_motion_mm": 119.87701058387756, "prediction_motion_mm": 117.83957481384277, "motion_ratio": 0.9830039491299359}, {"case": "case64", "step": 10000, "split": "val", "xyz_delta_mm": 14.424479566514492, "video_mae_255": 11.779873847961426, "action_max_abs": 0.08271515369415283, "video_max_abs": 231, "reference_motion_mm": 135.00402867794037, "prediction_motion_mm": 127.72983312606812, "motion_ratio": 0.9461186779157142}, {"case": "case65", "step": 10000, "split": "train", "xyz_delta_mm": 8.08767694979906, "video_mae_255": 6.264397144317627, "action_max_abs": 0.12338215112686157, "video_max_abs": 219, "reference_motion_mm": 132.77819752693176, "prediction_motion_mm": 129.65306639671326, "motion_ratio": 0.9764635219604888}, {"case": "case62", "step": 20000, "split": "val", "xyz_delta_mm": 21.491682156920433, "video_mae_255": 8.467308044433594, "action_max_abs": 0.12782281637191772, "video_max_abs": 225, "reference_motion_mm": 135.22657752037048, "prediction_motion_mm": 126.12535059452057, "motion_ratio": 0.9326964632786117}, {"case": "case63", "step": 20000, "split": "train", "xyz_delta_mm": 19.286777824163437, "video_mae_255": 8.722930908203125, "action_max_abs": 1.486314296722412, "video_max_abs": 234, "reference_motion_mm": 134.78682935237885, "prediction_motion_mm": 122.78296798467636, "motion_ratio": 0.9109418818932206}, {"case": "case64", "step": 20000, "split": "val", "xyz_delta_mm": 17.375420778989792, "video_mae_255": 7.163748741149902, "action_max_abs": 0.11384224891662598, "video_max_abs": 235, "reference_motion_mm": 119.22196298837662, "prediction_motion_mm": 130.45506179332733, "motion_ratio": 1.0942200457313882}, {"case": "case65", "step": 20000, "split": "train", "xyz_delta_mm": 7.244421169161797, "video_mae_255": 6.294121742248535, "action_max_abs": 0.0958893895149231, "video_max_abs": 213, "reference_motion_mm": 131.20315968990326, "prediction_motion_mm": 128.74610722064972, "motion_ratio": 0.9812729169399521}, {"case": "case66", "step": 10000, "split": "val", "xyz_delta_mm": 20.904703065752983, "video_mae_255": 6.414676666259766, "action_max_abs": 0.14793282747268677, "video_max_abs": 236, "reference_motion_mm": 45.42567580938339, "prediction_motion_mm": 33.413369208574295, "motion_ratio": 0.7355613012514001}, {"case": "case67", "step": 10000, "split": "train", "xyz_delta_mm": 21.851522848010063, "video_mae_255": 6.111066818237305, "action_max_abs": 0.08290988206863403, "video_max_abs": 230, "reference_motion_mm": 19.099554046988487, "prediction_motion_mm": 17.94983074069023, "motion_ratio": 0.939803657013681}, {"case": "case66", "step": 20000, "split": "val", "xyz_delta_mm": 16.01909101009369, "video_mae_255": 6.016247749328613, "action_max_abs": 0.12685638666152954, "video_max_abs": 237, "reference_motion_mm": 44.8831170797348, "prediction_motion_mm": 39.99066352844238, "motion_ratio": 0.8909956823497579}, {"case": "case67", "step": 20000, "split": "train", "xyz_delta_mm": 24.190420284867287, "video_mae_255": 8.063840866088867, "action_max_abs": 0.08505493402481079, "video_max_abs": 230, "reference_motion_mm": 16.80212840437889, "prediction_motion_mm": 18.106553703546524, "motion_ratio": 1.077634527470203}, {"case": "case01", "step": "official", "split": "train", "xyz_delta_mm": 2.0431214943528175, "video_mae_255": 6.520833969116211, "action_max_abs": 0.055897265672683716, "video_max_abs": 111, "reference_motion_mm": 71.08477503061295, "prediction_motion_mm": 71.76460325717926, "motion_ratio": 1.0095636263359284}, {"case": "case02", "step": "official", "split": "val", "xyz_delta_mm": 0.4124034894630313, "video_mae_255": 6.907436847686768, "action_max_abs": 0.006367862224578857, "video_max_abs": 126, "reference_motion_mm": 60.28071418404579, "prediction_motion_mm": 60.16377732157707, "motion_ratio": 0.9980601281180629}, {"case": "case00", "step": "official", "split": "val", "xyz_delta_mm": 3.590976819396019, "video_mae_255": 6.272903919219971, "action_max_abs": 0.045250147581100464, "video_max_abs": 111, "reference_motion_mm": 86.27630770206451, "prediction_motion_mm": 84.73500609397888, "motion_ratio": 0.9821352854666873}, {"case": "case03", "step": "official", "split": "train", "xyz_delta_mm": 0.3584741207305342, "video_mae_255": 6.923488140106201, "action_max_abs": 0.007373332977294922, "video_max_abs": 128, "reference_motion_mm": 60.676563531160355, "prediction_motion_mm": 60.58961525559425, "motion_ratio": 0.9985670204357989}, {"case": "case05", "step": "official", "split": "train", "xyz_delta_mm": 1.0015586158260703, "video_mae_255": 6.9730730056762695, "action_max_abs": 0.0077613890171051025, "video_max_abs": 133, "reference_motion_mm": 50.5710169672966, "prediction_motion_mm": 50.39304867386818, "motion_ratio": 0.9964808243120065}, {"case": "case04", "step": "official", "split": "val", "xyz_delta_mm": 2.202320145443082, "video_mae_255": 6.852756500244141, "action_max_abs": 0.0615534782409668, "video_max_abs": 138, "reference_motion_mm": 70.47375291585922, "prediction_motion_mm": 71.45785540342331, "motion_ratio": 1.0139640993539685}, {"case": "case06", "step": "official", "split": "val", "xyz_delta_mm": 1.2041761074215174, "video_mae_255": 6.658789157867432, "action_max_abs": 0.026950478553771973, "video_max_abs": 125, "reference_motion_mm": 103.25699299573898, "prediction_motion_mm": 103.58352959156036, "motion_ratio": 1.003162367858561}, {"case": "case07", "step": "official", "split": "train", "xyz_delta_mm": 0.4144686972722411, "video_mae_255": 6.167463779449463, "action_max_abs": 0.007985293865203857, "video_max_abs": 112, "reference_motion_mm": 43.578583747148514, "prediction_motion_mm": 43.60909014940262, "motion_ratio": 1.0007000319797246}, {"case": "case08", "step": "official", "split": "val", "xyz_delta_mm": 0.4266915493644774, "video_mae_255": 6.224708080291748, "action_max_abs": 0.005323588848114014, "video_max_abs": 140, "reference_motion_mm": 78.84060591459274, "prediction_motion_mm": 78.84981483221054, "motion_ratio": 1.0001168042471384}, {"case": "case09", "step": "official", "split": "train", "xyz_delta_mm": 0.43098567402921617, "video_mae_255": 6.445977687835693, "action_max_abs": 0.006703078746795654, "video_max_abs": 138, "reference_motion_mm": 86.0132947564125, "prediction_motion_mm": 85.91917157173157, "motion_ratio": 0.9989057135301294}, {"case": "case10", "step": "official", "split": "val", "xyz_delta_mm": 0.3222604573238641, "video_mae_255": 6.1078033447265625, "action_max_abs": 0.005362451076507568, "video_max_abs": 112, "reference_motion_mm": 1.9144451944157481, "prediction_motion_mm": 1.8671524012461305, "motion_ratio": 0.9752968675689614}, {"case": "case11", "step": "official", "split": "train", "xyz_delta_mm": 0.3044823242817074, "video_mae_255": 5.970761775970459, "action_max_abs": 0.006549954414367676, "video_max_abs": 111, "reference_motion_mm": 1.5149177052080631, "prediction_motion_mm": 1.4623792376369238, "motion_ratio": 0.9653192596597691}, {"case": "case12", "step": "official", "split": "val", "xyz_delta_mm": 1.0095974430441856, "video_mae_255": 6.747316360473633, "action_max_abs": 0.008741557598114014, "video_max_abs": 122, "reference_motion_mm": 95.81474214792252, "prediction_motion_mm": 95.85867822170258, "motion_ratio": 1.0004585523354248}, {"case": "case13", "step": "official", "split": "train", "xyz_delta_mm": 0.7761824526824057, "video_mae_255": 6.52078914642334, "action_max_abs": 0.0173015296459198, "video_max_abs": 122, "reference_motion_mm": 60.9629712998867, "prediction_motion_mm": 61.4510141313076, "motion_ratio": 1.008005561753546}, {"case": "case14", "step": "official", "split": "val", "xyz_delta_mm": 1.6130342846736312, "video_mae_255": 6.939324378967285, "action_max_abs": 0.022924691438674927, "video_max_abs": 123, "reference_motion_mm": 21.364854648709297, "prediction_motion_mm": 20.778603851795197, "motion_ratio": 0.9725600381302141}, {"case": "case15", "step": "official", "split": "train", "xyz_delta_mm": 1.2608078541234136, "video_mae_255": 6.868097305297852, "action_max_abs": 0.01597064733505249, "video_max_abs": 117, "reference_motion_mm": 90.15687555074692, "prediction_motion_mm": 89.63774144649506, "motion_ratio": 0.994241880044305}, {"case": "case16", "step": "official", "split": "val", "xyz_delta_mm": 0.6812250940129161, "video_mae_255": 7.78455924987793, "action_max_abs": 0.024179279804229736, "video_max_abs": 112, "reference_motion_mm": 110.41446775197983, "prediction_motion_mm": 110.28885841369629, "motion_ratio": 0.9988623833376103}, {"case": "case17", "step": "official", "split": "train", "xyz_delta_mm": 0.7820180035196245, "video_mae_255": 7.371732234954834, "action_max_abs": 0.02510926127433777, "video_max_abs": 122, "reference_motion_mm": 99.86132383346558, "prediction_motion_mm": 99.84322637319565, "motion_ratio": 0.9998187740801423}, {"case": "case18", "step": "official", "split": "val", "xyz_delta_mm": 0.7791199022904038, "video_mae_255": 6.77703857421875, "action_max_abs": 0.00545153021812439, "video_max_abs": 119, "reference_motion_mm": 46.641744673252106, "prediction_motion_mm": 46.186063438653946, "motion_ratio": 0.9902301846170115}, {"case": "case19", "step": "official", "split": "train", "xyz_delta_mm": 0.548795738723129, "video_mae_255": 6.385767459869385, "action_max_abs": 0.0072687119245529175, "video_max_abs": 126, "reference_motion_mm": 65.782830119133, "prediction_motion_mm": 65.51661342382431, "motion_ratio": 0.9959530975662408}, {"case": "case20", "step": "official", "split": "val", "xyz_delta_mm": 8.565348573029041, "video_mae_255": 7.327448844909668, "action_max_abs": 0.06166797876358032, "video_max_abs": 142, "reference_motion_mm": 111.3542690873146, "prediction_motion_mm": 107.5989231467247, "motion_ratio": 0.9662756895503909}, {"case": "case21", "step": "official", "split": "train", "xyz_delta_mm": 0.7345532649196684, "video_mae_255": 7.649217128753662, "action_max_abs": 0.010724842548370361, "video_max_abs": 132, "reference_motion_mm": 175.12963712215424, "prediction_motion_mm": 174.81768131256104, "motion_ratio": 0.9982187149204471}, {"case": "case22", "step": "official", "split": "val", "xyz_delta_mm": 0.3909443039447069, "video_mae_255": 6.254572868347168, "action_max_abs": 0.005362451076507568, "video_max_abs": 126, "reference_motion_mm": 26.644401252269745, "prediction_motion_mm": 26.730598881840706, "motion_ratio": 1.0032351122757401}, {"case": "case23", "step": "official", "split": "train", "xyz_delta_mm": 0.3768705064430833, "video_mae_255": 6.199397563934326, "action_max_abs": 0.005362510681152344, "video_max_abs": 111, "reference_motion_mm": 25.272231549024582, "prediction_motion_mm": 25.275424122810364, "motion_ratio": 1.000126327339934}, {"case": "case24", "step": "official", "split": "val", "xyz_delta_mm": 0.8350833086296916, "video_mae_255": 6.313090801239014, "action_max_abs": 0.008003920316696167, "video_max_abs": 114, "reference_motion_mm": 46.91783711314201, "prediction_motion_mm": 47.02974110841751, "motion_ratio": 1.0023851055837387}, {"case": "case25", "step": "official", "split": "train", "xyz_delta_mm": 0.8109139744192362, "video_mae_255": 6.324921131134033, "action_max_abs": 0.005816221237182617, "video_max_abs": 117, "reference_motion_mm": 48.98219555616379, "prediction_motion_mm": 49.19878393411636, "motion_ratio": 1.004421777658052}, {"case": "case26", "step": "official", "split": "val", "xyz_delta_mm": 1.6133743338286877, "video_mae_255": 8.55902099609375, "action_max_abs": 0.12173590064048767, "video_max_abs": 168, "reference_motion_mm": 102.07764804363251, "prediction_motion_mm": 101.94084048271179, "motion_ratio": 0.9986597696602273}, {"case": "case27", "step": "official", "split": "train", "xyz_delta_mm": 0.4919205093756318, "video_mae_255": 7.664628505706787, "action_max_abs": 0.010647118091583252, "video_max_abs": 165, "reference_motion_mm": 45.95780000090599, "prediction_motion_mm": 45.93105986714363, "motion_ratio": 0.9994181589684051}, {"case": "case28", "step": "official", "split": "val", "xyz_delta_mm": 1.0637491941452026, "video_mae_255": 8.316859245300293, "action_max_abs": 0.012400627136230469, "video_max_abs": 146, "reference_motion_mm": 86.90039068460464, "prediction_motion_mm": 86.39218658208847, "motion_ratio": 0.9941518778165147}, {"case": "case29", "step": "official", "split": "train", "xyz_delta_mm": 1.5279788058251143, "video_mae_255": 6.959717750549316, "action_max_abs": 0.014067500829696655, "video_max_abs": 125, "reference_motion_mm": 55.51653355360031, "prediction_motion_mm": 55.5759035050869, "motion_ratio": 1.0010694102762967}, {"case": "case30", "step": "official", "split": "val", "xyz_delta_mm": 0.7462226785719395, "video_mae_255": 7.077837944030762, "action_max_abs": 0.021294236183166504, "video_max_abs": 125, "reference_motion_mm": 51.98238417506218, "prediction_motion_mm": 52.51000449061394, "motion_ratio": 1.0101499829976033}, {"case": "case31", "step": "official", "split": "train", "xyz_delta_mm": 0.6657427875325084, "video_mae_255": 6.895808219909668, "action_max_abs": 0.0044107139110565186, "video_max_abs": 138, "reference_motion_mm": 64.39077854156494, "prediction_motion_mm": 64.36291337013245, "motion_ratio": 0.9995672490368398}, {"case": "case32", "step": "official", "split": "val", "xyz_delta_mm": 2.3792360443621874, "video_mae_255": 7.915263652801514, "action_max_abs": 0.009085893630981445, "video_max_abs": 172, "reference_motion_mm": 27.00013294816017, "prediction_motion_mm": 26.096772402524948, "motion_ratio": 0.9665423667590948}, {"case": "case33", "step": "official", "split": "train", "xyz_delta_mm": 0.8952913340181112, "video_mae_255": 7.274055004119873, "action_max_abs": 0.010724902153015137, "video_max_abs": 137, "reference_motion_mm": 5.901174619793892, "prediction_motion_mm": 5.968917161226273, "motion_ratio": 1.0114795012513538}, {"case": "case34", "step": "official", "split": "val", "xyz_delta_mm": 4.516018554568291, "video_mae_255": 6.457488059997559, "action_max_abs": 0.050454914569854736, "video_max_abs": 112, "reference_motion_mm": 77.34815031290054, "prediction_motion_mm": 74.80285316705704, "motion_ratio": 0.9670929797862408}, {"case": "case35", "step": "official", "split": "train", "xyz_delta_mm": 3.3843100536614656, "video_mae_255": 6.398972034454346, "action_max_abs": 0.013019606471061707, "video_max_abs": 128, "reference_motion_mm": 42.87194088101387, "prediction_motion_mm": 39.96000811457634, "motion_ratio": 0.9320783545928265}, {"case": "case36", "step": "official", "split": "val", "xyz_delta_mm": 0.850763579364866, "video_mae_255": 8.193259239196777, "action_max_abs": 0.03994348645210266, "video_max_abs": 148, "reference_motion_mm": 64.25635516643524, "prediction_motion_mm": 64.2833411693573, "motion_ratio": 1.0004199740687463}, {"case": "case37", "step": "official", "split": "train", "xyz_delta_mm": 0.837853760458529, "video_mae_255": 8.146682739257812, "action_max_abs": 0.04309689998626709, "video_max_abs": 150, "reference_motion_mm": 64.77132439613342, "prediction_motion_mm": 64.80208784341812, "motion_ratio": 1.000474954736089}, {"case": "case38", "step": "official", "split": "val", "xyz_delta_mm": 2.482024021446705, "video_mae_255": 6.634893417358398, "action_max_abs": 0.17293834686279297, "video_max_abs": 114, "reference_motion_mm": 49.5547279715538, "prediction_motion_mm": 50.20212382078171, "motion_ratio": 1.0130642599753454}, {"case": "case39", "step": "official", "split": "train", "xyz_delta_mm": 0.6924166809767485, "video_mae_255": 6.495617389678955, "action_max_abs": 0.026617705821990967, "video_max_abs": 109, "reference_motion_mm": 142.64924824237823, "prediction_motion_mm": 142.62080192565918, "motion_ratio": 0.9998005855826824}, {"case": "case40", "step": "official", "split": "val", "xyz_delta_mm": 2.143802586942911, "video_mae_255": 7.064655303955078, "action_max_abs": 0.007655888795852661, "video_max_abs": 148, "reference_motion_mm": 75.24828612804413, "prediction_motion_mm": 74.006088078022, "motion_ratio": 0.9834920087361408}, {"case": "case41", "step": "official", "split": "train", "xyz_delta_mm": 0.8469467284157872, "video_mae_255": 6.622778415679932, "action_max_abs": 0.030277669429779053, "video_max_abs": 133, "reference_motion_mm": 51.96991562843323, "prediction_motion_mm": 52.156608551740646, "motion_ratio": 1.0035923268500608}, {"case": "case42", "step": "official", "split": "val", "xyz_delta_mm": 0.6035282858647406, "video_mae_255": 6.299911022186279, "action_max_abs": 0.005323559045791626, "video_max_abs": 139, "reference_motion_mm": 64.86459821462631, "prediction_motion_mm": 64.89831209182739, "motion_ratio": 1.0005197577435}, {"case": "case43", "step": "official", "split": "train", "xyz_delta_mm": 0.7930474821478128, "video_mae_255": 6.176351070404053, "action_max_abs": 0.04152017831802368, "video_max_abs": 125, "reference_motion_mm": 54.97750639915466, "prediction_motion_mm": 54.97245863080025, "motion_ratio": 0.9999081848434928}, {"case": "case44", "step": "official", "split": "val", "xyz_delta_mm": 0.7548587745986879, "video_mae_255": 6.117895126342773, "action_max_abs": 0.01673951745033264, "video_max_abs": 126, "reference_motion_mm": 36.174919456243515, "prediction_motion_mm": 36.135513335466385, "motion_ratio": 0.9989106784100847}, {"case": "case45", "step": "official", "split": "train", "xyz_delta_mm": 1.0857073357328773, "video_mae_255": 6.286813259124756, "action_max_abs": 0.008731544017791748, "video_max_abs": 111, "reference_motion_mm": 49.48034882545471, "prediction_motion_mm": 49.009770154953, "motion_ratio": 0.9904895846195081}, {"case": "case46", "step": "official", "split": "val", "xyz_delta_mm": 0.47553679905831814, "video_mae_255": 5.607998847961426, "action_max_abs": 0.005362451076507568, "video_max_abs": 111, "reference_motion_mm": 15.706876292824745, "prediction_motion_mm": 15.634259209036827, "motion_ratio": 0.995376732939503}, {"case": "case47", "step": "official", "split": "train", "xyz_delta_mm": 0.5288936081342399, "video_mae_255": 5.663338661193848, "action_max_abs": 0.005244940519332886, "video_max_abs": 129, "reference_motion_mm": 8.017361164093018, "prediction_motion_mm": 7.989303208887577, "motion_ratio": 0.9965003503483039}, {"case": "case48", "step": "official", "split": "val", "xyz_delta_mm": 0.7749667856842279, "video_mae_255": 6.185427665710449, "action_max_abs": 0.005362451076507568, "video_max_abs": 129, "reference_motion_mm": 12.843003496527672, "prediction_motion_mm": 12.7712432295084, "motion_ratio": 0.9944125011693197}, {"case": "case49", "step": "official", "split": "train", "xyz_delta_mm": 0.7149951998144388, "video_mae_255": 5.97011137008667, "action_max_abs": 0.005362451076507568, "video_max_abs": 107, "reference_motion_mm": 8.741745725274086, "prediction_motion_mm": 8.785995654761791, "motion_ratio": 1.0050619099294744}, {"case": "case50", "step": "official", "split": "val", "xyz_delta_mm": 1.304950681515038, "video_mae_255": 5.922415733337402, "action_max_abs": 0.01597064733505249, "video_max_abs": 117, "reference_motion_mm": 29.539206996560097, "prediction_motion_mm": 29.45706620812416, "motion_ratio": 0.9972192622352556}, {"case": "case51", "step": "official", "split": "train", "xyz_delta_mm": 1.2466320767998695, "video_mae_255": 5.6838154792785645, "action_max_abs": 0.01146233081817627, "video_max_abs": 141, "reference_motion_mm": 14.899819158017635, "prediction_motion_mm": 14.260574243962765, "motion_ratio": 0.9570971360608165}, {"case": "case52", "step": "official", "split": "val", "xyz_delta_mm": 0.6924386252649128, "video_mae_255": 6.935956954956055, "action_max_abs": 0.00938427448272705, "video_max_abs": 127, "reference_motion_mm": 139.34971392154694, "prediction_motion_mm": 139.53663408756256, "motion_ratio": 1.0013413745945747}, {"case": "case53", "step": "official", "split": "train", "xyz_delta_mm": 0.6414704257622361, "video_mae_255": 6.643878936767578, "action_max_abs": 0.004542946815490723, "video_max_abs": 121, "reference_motion_mm": 27.085941284894943, "prediction_motion_mm": 26.774492114782333, "motion_ratio": 0.9885014455714598}, {"case": "case54", "step": "official", "split": "val", "xyz_delta_mm": 0.6743956473655999, "video_mae_255": 6.518054485321045, "action_max_abs": 0.0029483139514923096, "video_max_abs": 113, "reference_motion_mm": 151.38192474842072, "prediction_motion_mm": 151.2366682291031, "motion_ratio": 0.9990404632550482}, {"case": "case55", "step": "official", "split": "train", "xyz_delta_mm": 0.5296565359458327, "video_mae_255": 6.862786769866943, "action_max_abs": 0.003992676734924316, "video_max_abs": 129, "reference_motion_mm": 102.37450152635574, "prediction_motion_mm": 102.32768952846527, "motion_ratio": 0.9995427377208922}, {"case": "case56", "step": "official", "split": "val", "xyz_delta_mm": 0.5380018847063184, "video_mae_255": 6.2945075035095215, "action_max_abs": 0.005323559045791626, "video_max_abs": 124, "reference_motion_mm": 71.77922874689102, "prediction_motion_mm": 71.68492674827576, "motion_ratio": 0.9986862216234199}, {"case": "case57", "step": "official", "split": "train", "xyz_delta_mm": 3.909825347363949, "video_mae_255": 6.327184677124023, "action_max_abs": 0.03217458724975586, "video_max_abs": 137, "reference_motion_mm": 124.80412423610687, "prediction_motion_mm": 123.33344668149948, "motion_ratio": 0.9882161141419883}, {"case": "case58", "step": "official", "split": "val", "xyz_delta_mm": 0.7849712856113911, "video_mae_255": 6.210979461669922, "action_max_abs": 0.005362510681152344, "video_max_abs": 128, "reference_motion_mm": 48.94533008337021, "prediction_motion_mm": 48.74366894364357, "motion_ratio": 0.9958798696549161}, {"case": "case59", "step": "official", "split": "train", "xyz_delta_mm": 0.6797162350267172, "video_mae_255": 6.682684421539307, "action_max_abs": 0.005362391471862793, "video_max_abs": 137, "reference_motion_mm": 57.78397619724274, "prediction_motion_mm": 57.53174424171448, "motion_ratio": 0.9956349152113161}, {"case": "case60", "step": "official", "split": "val", "xyz_delta_mm": 1.1299485340714455, "video_mae_255": 8.030437469482422, "action_max_abs": 0.06880724430084229, "video_max_abs": 142, "reference_motion_mm": 73.58630746603012, "prediction_motion_mm": 73.14585149288177, "motion_ratio": 0.9940144302885197}, {"case": "case61", "step": "official", "split": "train", "xyz_delta_mm": 0.7633456261828542, "video_mae_255": 7.259426593780518, "action_max_abs": 0.005362451076507568, "video_max_abs": 167, "reference_motion_mm": 93.14870089292526, "prediction_motion_mm": 92.98796951770782, "motion_ratio": 0.9982744646605195}, {"case": "case62", "step": "official", "split": "val", "xyz_delta_mm": 0.4623126587830484, "video_mae_255": 6.376920223236084, "action_max_abs": 0.006987154483795166, "video_max_abs": 114, "reference_motion_mm": 132.09092617034912, "prediction_motion_mm": 131.9424957036972, "motion_ratio": 0.9988763008106969}, {"case": "case63", "step": "official", "split": "train", "xyz_delta_mm": 0.5328675033524632, "video_mae_255": 6.211803436279297, "action_max_abs": 0.0065498948097229, "video_max_abs": 111, "reference_motion_mm": 119.65824663639069, "prediction_motion_mm": 119.65885758399963, "motion_ratio": 1.0000051057710282}, {"case": "case64", "step": "official", "split": "val", "xyz_delta_mm": 0.3532420378178358, "video_mae_255": 6.552703380584717, "action_max_abs": 0.002725765109062195, "video_max_abs": 114, "reference_motion_mm": 112.8254160284996, "prediction_motion_mm": 112.79831826686859, "motion_ratio": 0.9997598257326685}, {"case": "case65", "step": "official", "split": "train", "xyz_delta_mm": 0.5205835332162678, "video_mae_255": 6.838939666748047, "action_max_abs": 0.006032764911651611, "video_max_abs": 124, "reference_motion_mm": 123.63800406455994, "prediction_motion_mm": 123.44866991043091, "motion_ratio": 0.9984686411305204}, {"case": "case66", "step": "official", "split": "val", "xyz_delta_mm": 0.837660685647279, "video_mae_255": 6.133299827575684, "action_max_abs": 0.005362451076507568, "video_max_abs": 130, "reference_motion_mm": 32.23424404859543, "prediction_motion_mm": 32.53881633281708, "motion_ratio": 1.0094487180702139}, {"case": "case67", "step": "official", "split": "train", "xyz_delta_mm": 0.8920464897528291, "video_mae_255": 5.975687026977539, "action_max_abs": 0.007761359214782715, "video_max_abs": 159, "reference_motion_mm": 13.658376410603523, "prediction_motion_mm": 13.837475329637527, "motion_ratio": 1.0131127532035917}], "verification": {"workers": 32, "rollouts": 384, "diagnostics": 7616, "identical_repeat_action_max": 0.0, "identical_repeat_video_max": 0, "no_future_GT_images_in_generated_feedback": true, "checkpoint_sha256": [{"node": 0, "step": 10000, "sha256": "70771c011be4f3dc13304be2173942ab5324d3b3304d3760ee33cd4ac116b3c9"}, {"node": 1, "step": 10000, "sha256": "70771c011be4f3dc13304be2173942ab5324d3b3304d3760ee33cd4ac116b3c9"}, {"node": 2, "step": 20000, "sha256": "3255bce6da6755a0849f5bbb7c83cd5dfec74a5d783cd4bbbf0e261a6be52017"}, {"node": 3, "step": 20000, "sha256": "3255bce6da6755a0849f5bbb7c83cd5dfec74a5d783cd4bbbf0e261a6be52017"}], "official": {"workers": 32, "matched_rollouts": 192, "native_rollouts": 192, "diagnostics": 3808, "repeat_action_and_video_exact": true, "full_video_does_not_change_action": true, "checkpoint_files": [{"name": "video.pt", "sha256": "89139f7d043e7c9ad9d24f394e442ce23291e06ef63d1aa44eea1594665e08c4", "bytes": 32790155854}, {"name": "dataset_stats.json", "sha256": "2649d4e3ff3d6d400c9953b7c20c79541474f9c38a8a9916b9caacecd8578f16", "bytes": 100256}, {"name": "action.pt", "sha256": "e3c8bb0e787c6f10e38afc154c75855a0f8b90d747fc5d8767c367d2050e8933", "bytes": 4057916513}]}}, "limitations": ["34 official validation episodes were monitored during training; not a new unseen benchmark.", "Only2 heldout task pairs have valid10.24s prompt chains.", "Recorded GT is not a counterfactual target; output sensitivity does not establish execution success.", "L2-only and robot-only are outside the six training dropout modes.", "History interventions preserve the current image, anchor, and proprioception.", "Stop motion ratios exclude original motion below1mm; unchanged arm-swap prompts excluded from aggregate.", "Official matched baseline uses the identical hierarchical prompts and shift3, which differ from its native task-only prompt distribution.", "Official native supplement uses its task-only template, shift1 and video seed offset1000003, with identical preprocessed RGB/current physical state/action seeds. This is a prompt/settings supplement, not an identical-hyperparameter comparison.", "Official action inference keeps the released noisy-video-cache path. Its displayed full10 video is generated after the action and does not condition that current action. Shared-FM generates video10 before clean-cache action10.", "Official checkpoint training exclusion of these34 heldout episodes has not been independently established."]}