Mixture of experts method improves humanoid robot motion tracking accuracy
MAPO: Mixture of Experts with Alternating Policy Optimization for Universal Humanoid Motion Tracking
Robotics
Summary
Tracking human movements with humanoid robots is difficult because people move in many different ways. The authors found that common methods to train motion tracking models can make mistakes when updating parts that choose how to blend experts, leading to poor results. They created a new method called MAPO that improves the training process by separately updating the routing and expert parts. This method helps robots better mimic a wide range of human motions more accurately and reliably.
What this means in practice
- •For robotics engineers: Improve humanoid robot controllers to more accurately mimic diverse human motions in real-world environments.
- •For animation developers: Create more precise humanoid animations by leveraging improved motion tracking policies from robot control methods.
Authors
Seonghyeon Lim, Hyeonwoo Lee, Mincheol Kim, Kimin Lee, Hyun Myung
Abstract
Universal humanoid motion tracking is essential for enabling humanoid robots to perform diverse tasks in human environments. To capture the wide diversity of human motions, mixture-of-experts (MoE) policies have been widely adopted, leveraging dynamic routing to enhance policy expressiveness. However, we identify that divergence between old and updated policies can be misestimated during policy-gradient updates of MoE policies, due to their dynamic routing. Because most policy-gradient algorithms regulate policy updates based on the divergence, this misestimation can lead to either overly conservative or excessively large updates, which can yield a suboptimal policy. In this paper, we propose Mixture of Experts with Alternating Policy Optimization (MAPO), which alternately optimizes the router and the experts to avoid the divergence misestimation. MAPO consistently tracks 182 motions that PPO with MoE fails to track, whereas PPO with MoE tracks only one motion that MAPO fails to track. MAPO also attains the highest tracking success rate among the compared methods and the lowest tracking error across most error metrics, demonstrating its superior performance. Videos can be found on the project website: https://mapo-corl2026.github.io/MAPO/