MaskAdapt: Learning Flexible Motion Adaptation via Mask-Invariant Prior for Physics-Based Characters

2026-03-31

AI Summary

This work addresses the challenge of balancing flexibility and stability in physics-based humanoid characters when faced with partial observation loss or the need for localized motion adjustments. The authors propose a two-stage residual learning framework: first, a mask-invariant base policy is trained using random body-part masking and a regularization term enforcing consistency in action distributions; second, a residual policy is learned on top of the frozen base policy to adjust only the motions of target body parts. By incorporating a mask-invariant prior and a local residual adaptation mechanism, the method enables independent multi-part adjustments within a single motion sequence and supports text-driven partial motion tracking. Experiments demonstrate that the approach achieves strong robustness and motion diversity under masked observations, significantly outperforms existing methods in local adaptation, and successfully enables complex motion composition and text-guided control in challenging scenarios.

Technology Category

Computer Vision: Motion & TrackingHumans and AI: Human-Aware Planning and Behavior PredictionIntelligent Robots: Motion and Path Planning

Application Category

Responsible Web: Machine-in-the-loop, human agency and autonomySearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingUser Modeling, Personalization and Recommendation: User privacy protection in personalized systems

Abstract

We present MaskAdapt, a framework for flexible motion adaptation in physics-based humanoid control. The framework follows a two-stage residual learning paradigm. In the first stage, we train a mask-invariant base policy using stochastic body-part masking and a regularization term that enforces consistent action distributions across masking conditions. This yields a robust motion prior that remains stable under missing observations, anticipating later adaptation in those regions. In the second stage, a residual policy is trained atop the frozen base controller to modify only the targeted body parts while preserving the original behaviors elsewhere. We demonstrate the versatility of this design through two applications: (i) motion composition, where varying masks enable multi-part adaptation within a single sequence, and (ii) text-driven partial goal tracking, where designated body parts follow kinematic targets provided by a pre-trained text-conditioned autoregressive motion generator. Through experiments, MaskAdapt demonstrates strong robustness and adaptability, producing diverse behaviors under masked observations and delivering superior targeted motion adaptation compared to prior work.

Problem

• motion adaptation
• physics-based characters
• partial observation
• targeted control
• humanoid animation

Innovation

• mask-invariant prior
• residual learning
• physics-based animation
• motion adaptation
• text-driven control