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

Yufei Xue^* Yunfeng Lin^* Wentao Dong Yang Tang Jingbo Wang
Jiangmiao Pang Ming Zhou Minghuan Liu^† Weinan Zhang^†

^* Equal Contribution Shanghai AI Lab Shanghai Jiao Tong University

Cross-Embodiment Locomotion

Single Policy Controls All Humanoids. Generalist zero-shot generalization across seven humanoids
with diverse DoFs, dynamic characteristics, and morphological structures.

Cross-Embodiment Teleoperation

Whole-Body Teleoperation. Real-time teleoperation of diverse humanoid robots using a single policy
driven by human teleoperator.

Cross-Embodiment Loco-Manipulation

H1-2 Clean up the Toys

G1 Clean up the Toys

G1 Open Door

Long-Horizon Loco-Manipulation. The robot first walks toward the box on the right and bends to grasp the plush toy. Next, it opens the door with the other hand, walks through, stops in front of the basket, squats, and places the toy inside, then neatly arranges the toys outside the basket.

Real-World Robots Introduction

Simulation Experiment

Cross-Embodiment Parallel Training

Zero-shot Specific Embodiment Development

Robot Introduction in the Simulation Experiment

Experiment Results

Comparison of the Training Curves Shows that, at Convergence, the Generalist Policy Achieves Approximately 85% of the Return Obtained by the Specialist Policy. After Fine-Tuning, the Generalist Policy Exhibits an Additional Improvement of about 10% in Return Compared with the Specialist Policies.

Generalist Average Command Tracking Errors and Survival Rates, Aggregated Across
All Robots, Compared with Specialist.

Abstract

Learning-based whole-body controllers have become a key driver for humanoid robots, yet most existing approaches require robot-specific training. In this paper, we study the problem of cross-embodiment humanoid control and show that a single policy can robustly generalize across a wide range of humanoid robot designs with one-time training. We introduce XHugWBC, a novel cross-embodiment training framework that enables generalist humanoid control through: (1) physics-consistent morphological randomization, (2) semantically aligned observation and action spaces across diverse humanoid robots, and (3) effective policy architectures modeling morphological and dynamical properties. XHugWBC is not tied to any specific robot. Instead, it internalizes a broad distribution of morphological and dynamical characteristics during training. By learning motion priors from diverse randomized embodiments, the policy acquires a strong structural bias that supports zero-shot transfer to previously unseen robots. Experiments on twelve simulated humanoids and seven real-world robots demonstrate the strong generalization and robustness of the resulting universal controller.

Method

XHugWBC Framework: 1). physics-consistent morphological randomization yields diverse humanoid embodiments, 2) unified state–action representation with semantic alignment across different robots, 3) graph-based policy for cross-humanoid control.

BibTeX

@article{xue2026xhugwbc, title={Scalable and General Whole-Body Control for Cross-Humanoid Locomotion}, author={Xue, Yufei and Lin, Yunfeng and Dong, Wentao and Tang, Yang and Wang, Jingbo and Pang, Jiangmiao and Zhou, Ming and Liu, Minghuan and Zhang, Weinan}, journal={arXiv preprint}, year={2026} }

XHugWBC

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