A 10-hour, lossless, per-episode subset of pick-and-place manipulation from griffinlabs/EgoDex-LeRobot-v3.0, sampled equally across 9 object categories (~1.11 h each).
Total: 8,844 episodes / 1,080,723 frames / 10.01 hours @ 30 fps
Format: LeRobot v3.0 (one folder per object category, each a standalone loadable dataset)
Modality: egocentric RGB video (1920x1080, AV1) + full-body/hand pose state (no action head; EgoDex is state-only)… See the full description on the dataset page:
https://huggingface.co/datasets/Kavin60606/EgoDex-PickPlace-10hr.
