DexWild: Dexterous Human Interactions for In-the-Wild Robot Policies

In three linesDexWild introduces a low-cost system for collecting dexterous manipulation data through natural human hand gestures. The framework co-trains on human and robot demonstrations, achieving 68.5% success rate in unseen environments (4× better than robot-only) and 5.8× improved cross-embodiment generalization.

Summary generated by Claude — human-verified