Related Research
Jaeger BumbleBee
PDF Paper arXiv arXiv github Project

F A S T

General Humanoid Whole-Body Control via Pretraining and Fast Adaptation


WHU    BeingBeyond    PKU

PDF Paper arXiv arXiv github Code

General Humanoid Whole-Body Control via Pretraining and Fast Adaptation

Zepeng Wang1,2, Jiangxing Wang2, Shiqing Yao2, Yu Zhang2, Ziluo Ding2, Ming Yang2,3, Yuxuan Wang2,3, Haobin Jiang2,3, Chao Ma1, Xiaochuan Shi1†, Zongqing Lu2,3§


1Wuhan University     2BeingBeyond     3Peking University


†Corresponding Author   §Project Leader

Paper arXiv Project

Various motion tracking demos

High dynamic motion

FAST tracks a high-dynamic kick motion.

Text-based motion

Text: “person hops on one leg”

Text-based motion

Text: “a person does a spinning kick”

Video-based motion

FAST tracks a long-horizon dance motion generated from a video.

Teleoperation demos

Walk and kick football

High dynamic motion

Long-horizon motion "Taichi"

Play with human

Push chair

Grab toys and hold basket

Abstract

    Learning a general whole-body controller for humanoid robots remains challenging due to the diversity of motion distributions, the difficulty of fast adaptation, and the need for robust balance in high-dynamic scenarios. Existing approaches often require task-specific training or suffer from performance degradation when adapting to new motions. In this paper, we present FAST, a general humanoid whole-body control framework that enables Fast Adaptation and Stable Motion Tracking. FAST introduces Parseval-Guided Residual Policy Adaptation, which learns a lightweight delta action policy under orthogonality and KL constraints, enabling efficient adaptation to out-of-distribution motions while mitigating catastrophic forgetting. To further improve physical robustness, we propose Center-of-Mass-Aware Control, which incorporates CoM-related observations and objectives to enhance balance when tracking challenging reference motions. Extensive experiments in simulation and real-world deployment demonstrate that FAST consistently outperforms state-of-the-art baselines in robustness, adaptation efficiency, and generalization.

Overview of F A S T

FAST

    Our framework consists of three stages. (1) We construct a curated humanoid motion dataset via human-to-humanoid retargeting with auxiliary physical signals. (2) We train a general whole-body controller with a Mixture-of-Experts architecture and center-of-mass-aware control. (3) We perform fast adaptation via Parseval-guided residual policy learning.

BibTeX

@article{wang2026FAST,
title={General Humanoid Whole-Body Control via Pretraining and Fast Adaptation},
author={Zepeng Wang and Jiangxing Wang and Shiqing Yao and Yu Zhang and Ziluo Ding and Ming Yang and Yuxuan Wang and Haobin Jiang and Chao Ma and Xiaochuan Shi and Zongqing Lu},
journal={arXiv preprint arXiv:2602.11929},
year={2026}
}