Cross embodiment model improves humanoid robot whole body control
X-WBC: A Cross-Embodiment Foundation Model for Humanoid Whole-Body Control
Robotics
Summary
Controlling humanoid robots to perform human-like whole-body movements is tricky because each robot's body is different, so control methods usually work for only one robot at a time. The authors developed a new approach called X-WBC that learns shared human motion patterns usable across different robots, while customizing some parts to each robot’s unique body. This allows training one big model that works for various robots, improving how well they follow commands. Their tests show this approach works well on many simulated and real humanoid robots.
What this means in practice
- •For robotics engineers: Develop robot control policies that share learning across different humanoid robot platforms to improve movement accuracy and reduce training time.
- •For vr hardware developers: Integrate sparse VR sensor data to enhance humanoid robot motion control consistency across various command input types.
Authors
Juntong Zhang, Chun Gu, Li Zhang
Abstract
Scaling humanoid whole-body control toward general-purpose deployment requires large human motion corpora and training experience shared across robot bodies. Existing methods usually train one policy per robot, leaving motion experience isolated across embodiments. We introduce X-WBC, a cross-embodiment foundation framework that separates relatively shared human motion semantics from embodiment-specific physical execution. Human-centered command tokens align full human motion, robot reference motion, and sparse VR observations. A causal Transformer learns reusable temporal structure from mixed multi-robot rollouts, while lightweight robot-specific modules map the shared representation to each robot's proprioception and action space. Across nine simulated embodiments, external motions, and four real robots, experiments show that joint training improves tracking, the aligned representation supports consistent control across command sources, and the learned policy remains competitive beyond the training corpus. These results support heterogeneous humanoids as joint data sources and establish cross-embodiment joint training as a practical route toward whole-body control foundation models.