Robust and Generalized Humanoid Motion Tracking

2026-01-30

AI Summary

This work addresses the challenge of dynamic contact failure in humanoid robots caused by closed-loop control amplifying local imperfections when executing noisy or inconsistent reference motions. To this end, the authors propose a dynamics-conditioned command aggregation framework that integrates a causal temporal encoder with a multi-head cross-attention mechanism to adaptively fuse contextual motion commands based on the current dynamical state. Robustness is further enhanced through a fall-recovery curriculum featuring randomized unstable initializations and annealed auxiliary lift forces. The model requires only approximately 3.5 hours of motion data for single-stage end-to-end training—without distillation—and demonstrates zero-shot generalization to unseen actions. Notably, it achieves high-fidelity sim-to-real transfer in highly dynamic contact scenarios.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Natural Language Processing: Safety and Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Learning a general humanoid whole-body controller is challenging because practical reference motions can exhibit noise and inconsistencies after being transferred to the robot domain, and local defects may be amplified by closed-loop execution, causing drift or failure in highly dynamic and contact-rich behaviors. We propose a dynamics-conditioned command aggregation framework that uses a causal temporal encoder to summarize recent proprioception and a multi-head cross-attention command encoder to selectively aggregate a context window based on the current dynamics. We further integrate a fall recovery curriculum with random unstable initialization and an annealed upward assistance force to improve robustness and disturbance rejection. The resulting policy requires only about 3.5 hours of motion data and supports single-stage end-to-end training without distillation. The proposed method is evaluated under diverse reference inputs and challenging motion regimes, demonstrating zero-shot transfer to unseen motions as well as robust sim-to-real transfer on a physical humanoid robot.

Problem

• humanoid motion tracking
• robust control
• generalization
• contact-rich dynamics
• reference motion inconsistency

Innovation

• dynamics-conditioned command aggregation
• causal temporal encoder
• multi-head cross-attention
• fall recovery curriculum
• sim-to-real transfer