SMP: Reusable Score-Matching Motion Priors for Physics-Based Character Control

2025-12-02

AI Summary

Motion priors in virtual character control suffer from poor cross-task transferability, necessitating repeated retraining or reliance on reference motion data. Method: This paper proposes Score-Matching Motion Priors (SMMP), a general-purpose motion prior framework based on score matching. SMMP pretrains a motion diffusion model on large-scale motion capture data and distills it into a frozen, task-agnostic implicit reward function via Score Distillation Sampling (SDS). Contribution/Results: Without fine-tuning, SMMP seamlessly integrates with arbitrary downstream controllers, enabling zero-shot style transfer and motion composition. Evaluated on multiple physics-based humanoid control tasks, SMMP achieves motion quality comparable to state-of-the-art adversarial imitation learning methods, while significantly improving prior generalizability and deployment efficiency.

Technology Category

• Computer Vision: Motion & Tracking
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Data-driven motion priors that can guide agents toward producing naturalistic behaviors play a pivotal role in creating life-like virtual characters. Adversarial imitation learning has been a highly effective method for learning motion priors from reference motion data. However, adversarial priors, with few exceptions, need to be retrained for each new controller, thereby limiting their reusability and necessitating the retention of the reference motion data when training on downstream tasks. In this work, we present Score-Matching Motion Priors (SMP), which leverages pre-trained motion diffusion models and score distillation sampling (SDS) to create reusable task-agnostic motion priors. SMPs can be pre-trained on a motion dataset, independent of any control policy or task. Once trained, SMPs can be kept frozen and reused as general-purpose reward functions to train policies to produce naturalistic behaviors for downstream tasks. We show that a general motion prior trained on large-scale datasets can be repurposed into a variety of style-specific priors. Furthermore SMP can compose different styles to synthesize new styles not present in the original dataset. Our method produces high-quality motion comparable to state-of-the-art adversarial imitation learning methods through reusable and modular motion priors. We demonstrate the effectiveness of SMP across a diverse suite of control tasks with physically simulated humanoid characters. Video demo available at https://youtu.be/ravlZJteS20

Problem

• Creating reusable motion priors for character control
• Eliminating retraining for new controllers in imitation learning
• Generating naturalistic behaviors across diverse control tasks

Innovation

• Reusable motion priors using score distillation sampling
• Task-agnostic pre-trained diffusion models for naturalistic behaviors
• Composable styles synthesis from large-scale motion datasets