Event

OmniInteract proposes a dual-humanoid whole-body interactive control framework

1 reports · 1 sources

Yinghao Li and other researchers propose OmniInteract, a whole-body interactive control framework for dual-humanoid robots, achieving this goal through surface-aware retargeting and scalable residual policy learning. Methodologically, Surface Contact-Aware Retargeting uses dense human-robot surface correspondences to jointly retarget two humanoids, preserving the interactive contact interface and its near-field geometric context; partner-aware experts learn corrections on top of a frozen whole-body motion prior. The paper states that experiments show OmniInteract improves retargeting fidelity and dual-humanoid coordination, and better maintains interaction performance as skill diversity grows; in real-robot deployment, it achieved contact establishment, stable whole-body coordination, and recovery from external perturbations.