PhysiFlow: Physics-Aware Humanoid Whole-Body VLA via Multi-Brain Latent Flow Matching and Robust Tracking

TL;DR

This paper introduces a physics-aware, multi-brain VLA framework that enhances humanoid robot control by integrating semantic guidance with efficient vision-language-action inference for stable, coordinated whole-body movements.

Contribution

The paper proposes a novel multi-brain VLA framework that combines semantic motion intent with physics-awareness to improve humanoid robot control stability and efficiency.

Findings

Enabled reliable vision-language-guided full-body coordination

Improved stability in dynamic limb-coordinated tasks

Demonstrated effectiveness through experimental evaluation

Abstract

In the domain of humanoid robot control, the fusion of Vision-Language-Action (VLA) with whole-body control is essential for semantically guided execution of real-world tasks. However, existing methods encounter challenges in terms of low VLA inference efficiency or an absence of effective semantic guidance for whole-body control, resulting in instability in dynamic limb-coordinated tasks. To bridge this gap, we present a semantic-motion intent guided, physics-aware multi-brain VLA framework for humanoid whole-body control. A series of experiments was conducted to evaluate the performance of the proposed framework. The experimental results demonstrated that the framework enabled reliable vision-language-guided full-body coordination for humanoid robots.