Datasets:
Modalities:
Video
Size:
100K - 1M
Tags:
robotics
lerobot
imitation-learning
visuomotor-learning
tactile-sensing
dexterous-manipulation
License:
Commit History
3 days ago
-
Add camera calibration and camera-frame TCP supplement
Verified signaturePublish the reviewed English README, fixed camera calibration JSON and four explicitly named six-dimensional camera-frame TCP arrays per observation. Add separate download packages and checksums; preserve the original recordings and root-frame TCP values.549f933 -
Revert premature camera calibration supplement publication
Verified signatureRestore the prior dataset card and withdraw the four supplement files. Dataset recordings and existing metadata are unchanged.98eba73 -
Add camera intrinsics and extrinsics supplement
Verified signatureAdd separately downloadable fixed camera parameters and recorded head-camera transforms for 495 seasons. Preserve the existing recordings and TCP values; document coordinate conventions, frame correspondence, validity and calibration limits.c1f9f3e
18 days ago
-
Simplify coordinate and sensor definitions in README
Verified signatureRemove implementation commentary, historical metadata notes, transform notation, naming-alias prose and update-scope language. Retain the TCP definitions, units and axes, complete hand-joint order, tactile frame and units, and one front-view coordinate illustration.8e04aab -
Keep one HOME front-view coordinate illustration
Verified signatureRetains the single HOME front-view illustration and removes seven coordinate close-up PNGs and their README references. TCP definitions, axis directions and units, the complete hand-joint mapping, and elastomer-frame tactile units remain unchanged. Recorded data, videos and annotations are untouched.fcf40ab -
Document TCP frames, hand joint mapping and tactile units
Verified signatureClarifies observed TCP reference frames, axis directions, metre/radian units and rotation-vector representation; documents the 22-joint hand mapping and elastomer-frame tactile wrench units (N and N*m). Adds HOME-pose Isaac Sim coordinate illustrations and a machine-readable joint mapping. Camera diagrams show nominal URDF frames only. Recorded data, videos, annotations and calibration parameter packages are unchanged.45afd55
25 days ago
-
Unify three task names
Verified signatureTask-string changes only; no data, video, segment or subtask_index changes. - DealingtheCards -> Card Dealing (653 episodes), "Stack the Two Blocks." -> "Stack the Two Blocks" (383), brush_cup -> Brush Cup (3). The renamed task strings are updated in annotations, tasks.parquet, subtasks.parquet, the episode tasks column and the v2.1 jsonl tables; no task_index changes. Previous revision: ac26a631f84db37d5ef8c0a8aa512d298eb3bbfd5c3715a -
Unify fold_and_close_box task name; carton subtask timing from corrected goal track; README counts
Verified signatureLanguage and subtask-index changes only. - fold_and_close_box: 415 episodes named "fold_box" renamed "Fold Flat Box into Closed Box" (task name and task strings only; segments unchanged). - reorient_and_relocate_carton/season_POC22050_2026_06_02_15_42_17_train (94 episodes): subtask intervals taken from the annotators' corrected goal track (2026-08-19); frame subtask_index and its stats recomputed. - README: language segments 224,087 and covered frames updated to the current annotations. Previous revision: c8032fd891e3f6f19b0d4886793579c1dce5eea7ac26a63
26 days ago
-
Move three fold-from-flat seasons from close_kraft_paper_box to fold_and_close_box
Verified signatureThe three POC22027 2026-05-20 seasons (160 episodes) start from a flat sheet and fold the whole box, the fold_and_close_box task; their language already said so. Files are moved unchanged. close_kraft_paper_box now holds only the press-to-close task. - close_kraft_paper_box: 5 seasons / 645 episodes -> 2 / 485 - fold_and_close_box: 7 seasons / 537 episodes -> 10 / 697 - README task table updated. Totals unchanged. Previous revision: 034e3ec49f6ada72717e6fa12ec8775cde64a11ec8032fd -
Add step subtasks to per-hand-only episodes; align frame subtask_index in tic-tac-toe
Verified signatureLanguage and subtask-index changes only; no state, action, tactile or video data changes. - 527 episodes in 7 tasks had only per-hand action descriptions (no instruction, success criteria or skill labels): deal_playing_cards (283), hammer_nails (154), reposition_and_stack_blocks (63), fit_trash_bag_into_bin (23), rotate_bottle_cap (2), cover_ball_with_cup (1), insert_knife_into_cutlery_bin (1). Each now has step subtasks with skill labels and the task fields of fully annotated episodes of the same task, with step intervals aligned by joint motion and fingertip contact. The per-hand descriptions are kept with label hand_actions. - 11 of those episodes visibly fail (checked on video). They keep only the steps that happen, and an `error` segment with a short note covers the failure frames. - play_tic_tac_toe (521) and sweep_waste_into_dustpan/season_POC22054_2026_05_30_11_16_03_train episode 0: frame-level subtask_index now follows their step subtasks. - transfer_salt_between_cups_with_s…034e3ec
29 days ago
-
Re-time placeholder subtask intervals in 112 episodes
Verified signature112 episodes in six tasks had subtask intervals on a fixed 100-frame grid (0-99, 100-199, ...) instead of following the motion. The intervals now come from aligning joint motion and fingertip contact with hand-labelled episodes of the same task, spot-checked on the head camera. Subtask text is unchanged. - annotations.jsonl / annotation_cache.jsonl (v3.0 and v2.1): new start_step / end_step, and a timing_correction note on each re-timed episode. - Data parquet (v3.0 files and v2.1 episode files): per-frame subtask_index follows the new intervals. No other column changes. - Five B-scored episodes that fail get an `error` segment over the failure frames (with a short note). - subtask_index stats in meta/episodes, meta/stats.json and v2.1 episodes_stats.jsonl recomputed. - README: frames covered by language segments 28,897,909 -> 28,920,254. Tasks: sweep_waste_into_dustpan (81), fold_and_close_box (18), clean_garment_with_lint_roller (8), empty_dustpan_into_bin (2), nest_cups (2), reloc…344e85b
30 days ago
-
Correct language annotations (place_coffee_filter_in_dripper, one sweep episode), add tic-tac-toe step subtasks; update dataset card
Verified signatureLanguage-only fix; no data or video files change. - place_coffee_filter_in_dripper (21 seasons, 784 episodes): text described placing pens in a holder; now describes the coffee-filter task shown in the video (from the raw annotation files). Segment intervals unchanged. - sweep_waste_into_dustpan/season_POC22054_2026_05_30_11_16_03_train episode 0: sweep text replaced with the dustpan-emptying text its video shows. - play_tic_tac_toe (521 episodes): step subtasks with skill labels, instruction and success criteria added; per-hand descriptions kept (label hand_actions). - README: segment counts updated; field descriptions note that the torso joint-torque dims and the TCP force/torque slots are 0. Previous revision: 6b48181a41d8836692fb1204cc60ade64e107d2073dcc29
-
Correct dataset license to CC BY-NC 4.0 and prohibit commercial use
Verified signature6b48181
-
Use SharpaDex naming consistently in README links and examples
Verified signature6d033f4
-
Remove 15 specified coffee-filter episodes and rebuild consistent LeRobot exports
Verified signature8386477 -
Upload DexSharpa task data pipeline batch 101/101
Verified signaturecf14120 -
Upload DexSharpa task data pipeline batch 100/101
Verified signaturec10b3e4 -
Upload DexSharpa task data pipeline batch 99/101
Verified signature808b531 -
Upload DexSharpa task data pipeline batch 98/101
Verified signature7347454 -
Upload DexSharpa task data pipeline batch 97/101
Verified signaturecdb8b30 -
Upload DexSharpa task data pipeline batch 96/101
Verified signaturec195819 -
Upload DexSharpa task data pipeline batch 95/101
Verified signaturec95803a -
Upload DexSharpa task data pipeline batch 94/101
Verified signatureaf88d54 -
Upload DexSharpa task data pipeline batch 93/101
Verified signatureb7a3d99 -
Upload DexSharpa task data pipeline batch 92/101
Verified signature9f6c3b1 -
Upload DexSharpa task data pipeline batch 91/101
Verified signaturec61c28e -
Upload DexSharpa task data pipeline batch 90/101
Verified signaturee5a3ead -
Upload DexSharpa task data pipeline batch 89/101
Verified signatureda6ba93 -
Upload DexSharpa task data pipeline batch 88/101
Verified signaturebf7db77 -
Upload DexSharpa task data pipeline batch 87/101
Verified signatured47c0d3 -
Upload DexSharpa task data pipeline batch 86/101
Verified signature9c6cae5 -
Upload DexSharpa task data pipeline batch 85/101
Verified signature439647d -
Upload DexSharpa task data pipeline batch 84/101
Verified signature7e27997 -
Upload DexSharpa task data pipeline batch 83/101
Verified signaturef43f466 -
Upload DexSharpa task data pipeline batch 82/101
Verified signatured44ff1e -
Upload DexSharpa task data pipeline batch 81/101
Verified signature84e3c69 -
Upload DexSharpa task data pipeline batch 80/101
Verified signature11167be -
Upload DexSharpa task data pipeline batch 79/101
Verified signaturef62210f -
Upload DexSharpa task data pipeline batch 78/101
Verified signaturebde6b3f -
Upload DexSharpa task data pipeline batch 77/101
Verified signature8311e76 -
Upload DexSharpa task data pipeline batch 76/101
Verified signature55308ff -
Upload DexSharpa task data pipeline batch 75/101
Verified signaturea1dde34 -
Upload DexSharpa task data pipeline batch 74/101
Verified signatureba02464 -
Upload DexSharpa task data pipeline batch 73/101
Verified signatureeba749b -
Upload DexSharpa task data pipeline batch 72/101
Verified signatureab90eff -
Upload DexSharpa task data pipeline batch 71/101
Verified signature20d00ba -
Upload DexSharpa task data pipeline batch 70/101
Verified signature712365e -
Upload DexSharpa task data pipeline batch 69/101
Verified signature8753387 -
Upload DexSharpa task data pipeline batch 68/101
Verified signaturea2e2840 -
Upload DexSharpa task data pipeline batch 67/101
Verified signaturec6bb8d0