Agility Meets Stability: Versatile Humanoid Control with Heterogeneous Data
Yixuan Pan, Ruoyi Qiao, Li Chen, Kashyap Chitta, Liang Pan, Haoguang Mai, Qingwen Bu, Cunyuan Zheng, Hao Zhao, Ping Luo, Hongyang Li
AI summary
Problem
Current humanoid controllers specialize in either agile motions or robust balance but fail to unify both due to reliance on human motion capture data and conflicting reinforcement learning objectives.
Approach
AMS trains a single whole-body tracking policy using a mix of human MoCap data and physically constrained synthetic balance motions, guided by a hybrid reward scheme and performance-driven adaptive sampling.
Key results
- First unified policy achieving simultaneous dynamic tracking and extreme balance maintenance
- Zero-shot execution of challenging balance poses and agile skills on a real Unitree G1 robot
- Synthetic balance motion generator that expands feasible motion space beyond human capabilities
- Adaptive sampling and reward shaping that improve training efficiency and cross-motion generalization
Why it matters
Enables humanoid robots to seamlessly blend dynamic agility with robust stability, advancing their readiness for real-world human-centric tasks.
Abstract
Humanoid robots are envisioned to perform a wide range of tasks in human-centered environments, requiring controllers that combine agility with robust balance. Recent advances in locomotion and whole-body tracking have enabled impressive progress in either agile dynamic skills or stability- critical behaviors, but existing methods remain specialized, focusing on one capability while compromising the other. In this work, we introduce AMS (Agility Meets Stability), the first framework that unifies both dynamic motion tracking and extreme balance maintenance in a single policy. Our key 2026 IEEE International Conference on Robotics and Automation (ICRA 2026) June 1-5, 2026. Vienna, Austria 979-8-3315-8160-2/26/$31.00 ©2026 IEEE 13406