Event

X-WBC: A Cross-Embodiment Humanoid Whole-Body Control Foundation Model

1 reports · 1 sources

Juntong Zhang, Chun Gu, Li Zhang, and others propose X-WBC, a cross-embodiment humanoid whole-body control foundation framework. Its approach separates relatively shared human motion semantics from the physical execution of each robot embodiment: human-centric command tokens align full-body human motion, robot reference motion, and sparse VR observations; a causal Transformer learns reusable temporal structure from mixed multi-robot rollouts; and lightweight robot-specific modules then map this to each embodiment's perception and action spaces. The authors report that experiments on nine simulated embodiments, external motion, and four real robots show that joint training improves tracking performance, that the aligned representation supports consistent control across different command sources, and that the learned policies remain competitive beyond the training corpus. These are all results reported by the paper's authors.