Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
video
video
9
15.8
label
class label
3 classes
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
0observation.images.cam_high
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
1observation.images.cam_left_wrist
End of preview. Expand in Data Studio

E098 Train50 V4: compacted, smooth simulation demonstrations

50 unique source seeds, one qualified demonstration per seed. All three camera streams are 320×240 H.264 at50/3 FPS; physics250Hz. Repeated arm knots are capped at3; drawer readiness uses14cm and0.3s stability. Close-to-motion boundaries use C2 quintic transitions with analytic acceleration at most3rad/s² and jerk at most100rad/s³, matched outgoing velocity and exact source endpoints. These bounds apply to the repaired transition intervals.

Sampled normalized scheduling coordinates occupy all50 equal strata, one each. Reference durations include the repaired interpolation. Feedback may advance actual left-arm startup when the right arm waits; actual offsets are recorded separately and have their own distribution. Original50 remains a separate baseline.

Qualified50/50; frames10891; videos150. Command holds in all14 dimensions: 113.10s/711.18s before, 42.06s/650.46s after (adjacent target change <=1e-4). observation.state follows the original RoboTwin drive-target convention. Measured12D joint positions are retained separately in meta/actual_joint_state/.

meta/repair-comparison.json records per-episode timing, hold rates and gate events. All videos are fully decoded with timestamps checked. Task success and scene penetration at most10mm are checked in reference simulation and fresh video replay.

Result lifecycle: reported → audit pending. User review pending.

Pinned OpenPI normalization covers all dataset frames, including the final remainder. See meta/norm_stats.json and its source/configuration receipt.

Review and provenance

Official LeRobot visualizer

Source baseline: Shiki42/PutCab-AB-RandomDelta-Train50-V3 @ d62cf7b1f52630f4eb897976f9807f6c1b7d00c1. Raw source75: Shiki42/robotwin_putcab_fixed_role_source75 @ b5ffd87a13cde9a376caca08245b6f202319cb00.

Uniformity applies to the sampled and quantized scheduled u coordinates; feedback-adjusted effective start offsets have a separate measured histogram in meta/distribution.json.

Root SHA256SUMS identifies the publication tree. .ctr-revision.json and meta/source-SHA256SUMS preserve the sealed source identity; the source README entry maps to meta/source-README.md. Executed code/runtime identities and episode membership are in meta/run-provenance.json; the frozen sampling plan also preserves inherited preparation metadata.

Downloads last month
4