Research Analyzer
← Back ICRA 2026

ExBody2: Advanced Expressive Humanoid Whole-Body Control

Mazeyu Ji, Xuanbin Peng, Fangchen Liu, Jialong LI, Ge Yang, Xuxin Cheng, Xiaolong Wang

PDF

AI summary

Key figure (auto-extracted from paper)
ExBody2 enables real-world humanoid robots to perform stable, expressive whole-body motions by decoupling tracking into velocity control and local keypoint imitation, guided by an automated dataset curation method that balances feasibility and diversity.
Humanoid control motion imitation reinforcement learning sim-to-real dataset curation whole-body tracking

Problem

Humanoid robots struggle to perform expressive, dynamic whole-body motions while maintaining stability due to the kinematic gap with humans and the trade-off between motion diversity and physical feasibility.

Approach

The framework trains a generalist policy using automated data curation to filter infeasible motions while preserving diversity, then fine-tunes specialist policies for specific tasks, decoupling global tracking into velocity control and local keypoint imitation via a teacher-student reinforcement learning pipeline.

Key results

  • Automated dataset curation balances motion feasibility and diversity
  • Decoupled tracking architecture improves stability and motion fidelity
  • Generalist policy outperforms baselines across diverse motions
  • Specialist fine-tuning boosts accuracy for targeted motion groups

Why it matters

Provides a scalable, data-driven framework for achieving human-level expressiveness and stability in real-world humanoid robots, advancing practical whole-body control.

Abstract

This paper tackles the challenge of enabling real- world humanoid robots to perform expressive and dynamic whole-body motions while maintaining stability. We propose Advanced Expressive Whole-Body Control (ExBody2), a whole- body tracking framework trained in simulation with Reinforce- ment Learning and then transferred to the real world. The framework decouples keypoint tracking from velocity control and leverages a privileged teacher policy to distill precise mimic skills into the student policy, enabling robust, high-fidelity reproduction of complex motions such as walking, crouching, and dancing. A significant contribution is the identification of an empirical trade-off between feasibility and diversity in motion datasets, which guides the development of an automatic dataset curation method. This principle facilitates pretraining a versatile model generalizing well across diverse motions and can be fine-tuned for specific tasks to achieve superior tracking accuracy. Extensive experiments show that Exbody2 achieves consistently better performance than strong baselines and provides insights that may inform future work on whole- body humanoid control. 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 613

Index terms

Humanoid and Bipedal Locomotion Humanoid Robot Systems Reinforcement Learning

Related papers