MotionRL: Align Text-to-Motion Generation to Human Preferences with Multi-Reward Reinforcement Learning

TL;DR

MotionRL employs multi-reward reinforcement learning to optimize text-to-motion generation, aligning outputs with human preferences and balancing multiple objectives like motion quality and text adherence.

Contribution

It introduces a novel multi-reward RL framework for text-to-motion tasks, incorporating human preferences and multi-objective optimization to improve alignment and control.

Findings

Enhanced alignment with human preferences

Improved performance on multiple metrics

Allows control over generated motion results

Abstract

We introduce MotionRL, the first approach to utilize Multi-Reward Reinforcement Learning (RL) for optimizing text-to-motion generation tasks and aligning them with human preferences. Previous works focused on improving numerical performance metrics on the given datasets, often neglecting the variability and subjectivity of human feedback. In contrast, our novel approach uses reinforcement learning to fine-tune the motion generator based on human preferences prior knowledge of the human perception model, allowing it to generate motions that better align human preferences. In addition, MotionRL introduces a novel multi-objective optimization strategy to approximate Pareto optimality between text adherence, motion quality, and human preferences. Extensive experiments and user studies demonstrate that MotionRL not only allows control over the generated results across different objectives but also significantly enhances performance across these metrics compared to other algorithms.