ZUO66/organize_table_regular_rgb_head_chest

Organize Table Regular

352 episodes / 116,279 robot frames with three native LeRobot v3 video cameras: robot head (observation.images.cam_head), robot chest (observation.images.cam_chest), and upright human first-person RGB (observation.images.aria_rgb, 1408 x 1408). Human RGB is sampled from seven Aria recordings using the existing robot-frame mapping. The original robot state/action data and head/chest videos are preserved.

See human video, gaze and hand documentation, annotated human previews, and episode mapping. Gaze and hands remain NPZ sidecars; upright-video pixel coordinates and masks are included alongside the unchanged original angles and device-frame 3D coordinates. Do not rotate the already upright robot chest video again.

Device-clock alignment is approximate; absolute synchronization and human-to-robot spatial calibration are unverified. Native MPS hand estimates can miss visible finger positions even at high confidence. Use the validity masks, including RGB-relative sample-gap masks. Human video segments cover all matched robot frames; the source VRS containers and their other sensor streams are not uploaded.