PhysMoDPO: Physically-Plausible Humanoid Motion with Preference Optimization

2026-03-13

AI Summary

This work addresses the discrepancy between semantic intent and physical feasibility in human motion generation, where motions produced by existing diffusion models often deviate from original action semantics or text instructions after being converted by whole-body controllers (WBCs) to satisfy physical constraints. To resolve this, we propose PhysMoDPO, a novel framework that integrates WBC directly into the diffusion model training pipeline. By leveraging physics-based and task-oriented rewards to construct preference signals, PhysMoDPO employs Direct Preference Optimization (DPO) to guide end-to-end learning of motions that are both physically plausible and semantically aligned with input instructions. Our approach eliminates handcrafted heuristics and demonstrates significant improvements in physical realism and task success across text-to-motion and spatial control tasks, achieving zero-shot transfer to both simulation environments and real-world deployment on the G1 humanoid robot.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Motion and Path Planning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Recent progress in text-conditioned human motion generation has been largely driven by diffusion models trained on large-scale human motion data. Building on this progress, recent methods attempt to transfer such models for character animation and real robot control by applying a Whole-Body Controller (WBC) that converts diffusion-generated motions into executable trajectories. While WBC trajectories become compliant with physics, they may expose substantial deviations from original motion. To address this issue, we here propose PhysMoDPO, a Direct Preference Optimization framework. Unlike prior work that relies on hand-crafted physics-aware heuristics such as foot-sliding penalties, we integrate WBC into our training pipeline and optimize diffusion model such that the output of WBC becomes compliant both with physics and original text instructions. To train PhysMoDPO we deploy physics-based and task-specific rewards and use them to assign preference to synthesized trajectories. Our extensive experiments on text-to-motion and spatial control tasks demonstrate consistent improvements of PhysMoDPO in both physical realism and task-related metrics on simulated robots. Moreover, we demonstrate that PhysMoDPO results in significant improvements when applied to zero-shot motion transfer in simulation and for real-world deployment on a G1 humanoid robot.

Problem

• humanoid motion
• physics compliance
• motion generation
• text-to-motion
• trajectory deviation

Innovation

• Direct Preference Optimization
• Whole-Body Control
• Physically-Plausible Motion
• Diffusion Models
• Humanoid Robotics