Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
video
video
93.2
174
label
class label
3 classes
0observation.images.left_wrist
0observation.images.left_wrist
0observation.images.left_wrist
0observation.images.left_wrist
0observation.images.left_wrist
0observation.images.left_wrist
0observation.images.left_wrist
1observation.images.right_wrist
1observation.images.right_wrist
1observation.images.right_wrist
1observation.images.right_wrist
1observation.images.right_wrist
1observation.images.right_wrist
1observation.images.right_wrist
2observation.images.top
2observation.images.top
2observation.images.top
2observation.images.top
2observation.images.top
2observation.images.top
2observation.images.top

ABC-130k (256px re-conversion) — abc130k_val_t106_insert_the_plug

One task of the ABC-130k dual-arm (YAM) teleoperation corpus, in LeRobot v2.1 layout (per-episode AV1 mp4s + parquet). Re-converted from the raw XDOF/ABC-130k mcap recordings (Apache-2.0): full-frame squash to 256x256 (the public 224px port letterboxes — 37.5% of every stereo frame is black bars), SVT-AV1 crf 30 / preset 12 / keyframe every 2 frames, tick-aligned to the same 30 Hz grid as the original port (frame counts match parquet rows exactly; joint streams verified against source to <=1.2e-7 rad).

Cameras: observation.images.top, observation.images.left_wrist, observation.images.right_wrist (stereo stations: one eye per episode, sha1 rule matching the upstream exporter). Part of a 386-repo family (abc130k_t* train / abc130k_val_t* val) published 2026-08-20.

Downloads last month
17