XHugWBC: Scalable and General Whole-Body Control for Cross-Humanoid Locomotion

ABSTRACT

We address the challenge of cross-embodiment humanoid control, where a single policy must generalize across diverse robot designs. We introduce XHugWBC, a framework that uses physics-consistent morphological randomization and semantically aligned observation spaces to train a 'universal' humanoid controller. For the first time, we demonstrate a single policy achieving robust zero-shot whole-body control across seven distinct real-world humanoid robots, including varied limb lengths and joint configurations. This suggests that humanoid locomotion is becoming a 'solved' foundation problem.