Track Any Motions under Any Disturbances

2025-09-17

AI Summary

Humanoid robots struggle to stably track diverse, highly dynamic, multi-contact motions in real-world environments while robustly handling terrain variations, external disturbances, and intrinsic dynamical uncertainties. To address this, we propose Any2Track, a two-stage reinforcement learning framework. Our method integrates sim-to-real transfer, history-driven adaptive control, and contact-aware modeling. Key contributions include: (1) the AnyAdapter module—a decoupled, zero-shot online dynamics adaptation mechanism enabling cross-domain transfer without retraining; and (2) the unified AnyTracker policy, which enables end-to-end tracking of multiple motion types using a single model. We deploy Any2Track zero-shot on the Unitree G1 robot. Despite complex physical disturbances—including uneven terrain, pushes, and hardware variations—the system achieves high-fidelity motion tracking. Experiments demonstrate substantial improvements in generalization, robustness, and practical deployability compared to prior approaches.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Computer Vision: Motion & Tracking
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

A foundational humanoid motion tracker is expected to be able to track diverse, highly dynamic, and contact-rich motions. More importantly, it needs to operate stably in real-world scenarios against various dynamics disturbances, including terrains, external forces, and physical property changes for general practical use. To achieve this goal, we propose Any2Track (Track Any motions under Any disturbances), a two-stage RL framework to track various motions under multiple disturbances in the real world. Any2Track reformulates dynamics adaptability as an additional capability on top of basic action execution and consists of two key components: AnyTracker and AnyAdapter. AnyTracker is a general motion tracker with a series of careful designs to track various motions within a single policy. AnyAdapter is a history-informed adaptation module that endows the tracker with online dynamics adaptability to overcome the sim2real gap and multiple real-world disturbances. We deploy Any2Track on Unitree G1 hardware and achieve a successful sim2real transfer in a zero-shot manner. Any2Track performs exceptionally well in tracking various motions under multiple real-world disturbances.

Problem

• Tracking diverse humanoid motions under disturbances
• Overcoming sim2real gap and external force challenges
• Achieving real-time motion adaptation on physical hardware

Innovation

• Two-stage reinforcement learning framework
• History-informed adaptation module
• Zero-shot sim-to-real transfer capability