Robust and Generalized Humanoid Motion Tracking

TL;DR

This paper introduces a robust humanoid motion tracking framework that effectively handles noisy reference motions and disturbances, enabling zero-shot transfer and robust sim-to-real performance with minimal training data.

Contribution

The paper presents a dynamics-conditioned command aggregation method with a causal encoder and attention mechanism, along with a fall recovery curriculum, for improved robustness and generalization in humanoid control.

Findings

Supports zero-shot transfer to unseen motions

Achieves robust sim-to-real transfer on a physical robot

Requires only 3.5 hours of motion data for training

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.