FARM: Frame-Accelerated Augmentation and Residual Mixture-of-Experts for Physics-Based High-Dynamic Humanoid Control

2025-08-27

AI Summary

Existing physical humanoid controllers perform well on everyday low-dynamic motions but struggle to stably execute explosive high-dynamic maneuvers, hindering real-world deployment. To address this, we propose a unified control framework featuring (i) the first publicly available dataset for high-dynamic humanoid motion—HDHM; (ii) a frame-wise acceleration augmentation strategy; and (iii) a residual Mixture-of-Experts (MoE) network integrated with a robust base controller, enabling adaptive modeling and joint optimization of both low- and high-dynamic behaviors. Experiments on HDHM demonstrate a 42.8% reduction in trajectory tracking failure rate and a 14.6% decrease in global mean joint position error, while preserving near-perfect accuracy on low-dynamic tasks. This work establishes a new paradigm, introduces a novel benchmark dataset, and proposes an innovative architecture for high-dynamic humanoid control.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Manipulation

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Unified physics-based humanoid controllers are pivotal for robotics and character animation, yet models that excel on gentle, everyday motions still stumble on explosive actions, hampering real-world deployment. We bridge this gap with FARM (Frame-Accelerated Augmentation and Residual Mixture-of-Experts), an end-to-end framework composed of frame-accelerated augmentation, a robust base controller, and a residual mixture-of-experts (MoE). Frame-accelerated augmentation exposes the model to high-velocity pose changes by widening inter-frame gaps. The base controller reliably tracks everyday low-dynamic motions, while the residual MoE adaptively allocates additional network capacity to handle challenging high-dynamic actions, significantly enhancing tracking accuracy. In the absence of a public benchmark, we curate the High-Dynamic Humanoid Motion (HDHM) dataset, comprising 3593 physically plausible clips. On HDHM, FARM reduces the tracking failure rate by 42.8% and lowers global mean per-joint position error by 14.6% relative to the baseline, while preserving near-perfect accuracy on low-dynamic motions. These results establish FARM as a new baseline for high-dynamic humanoid control and introduce the first open benchmark dedicated to this challenge. The code and dataset will be released at https://github.com/Colin-Jing/FARM.

Problem

• Bridging performance gap between gentle and explosive humanoid motions
• Enhancing tracking accuracy for high-dynamic humanoid control tasks
• Addressing lack of public benchmark for high-dynamic humanoid motions

Innovation

• Frame-accelerated augmentation for high-velocity training
• Residual mixture-of-experts for adaptive capacity allocation
• Robust base controller for reliable low-dynamic motion tracking