Bimanual FastUMI Multi-View Manipulation Demonstrations — LeRobot v2.1

This dataset contains synchronized 4-camera bimanual hand-held FastUMI demonstrations covering a variety of egocentric robotic manipulation tasks — Jenga, Candles, Contact Lens Case, Bolts, and Bucket. Each demonstration is captured with 4 cameras: head-mounted (egocentric), left wrist, right wrist, and third-person side view. Videos are aligned across cameras to identical wall-clock duration, and packaged as HuggingFace LeRobot v2.1 datasets (single episode per demonstration) spec (640×480 VGA at 30 fps). Using this dataset helps teams pretrain vision-language-action models, train imitation-learning / behavior-cloning policies, benchmark multi-view manipulation perception, and develop egocentric visual representations for robotic manipulation.