REvolve: Reward Evolution with Large Language Models using Human Feedback
This addresses the challenge of reward design for reinforcement learning in subjective tasks, offering a method to translate implicit human knowledge into explicit rewards, though it is incremental as it builds on existing LLM-based reward generation approaches.
The paper tackles the problem of designing effective reward functions for reinforcement learning in tasks where good behavior is hard to quantify, by introducing REvolve, a framework that uses large language models guided by human feedback to generate and refine rewards. The result shows that agents trained with REvolve-designed rewards outperform state-of-the-art baselines in autonomous driving, humanoid locomotion, and dexterous manipulation tasks.
Designing effective reward functions is crucial to training reinforcement learning (RL) algorithms. However, this design is non-trivial, even for domain experts, due to the subjective nature of certain tasks that are hard to quantify explicitly. In recent works, large language models (LLMs) have been used for reward generation from natural language task descriptions, leveraging their extensive instruction tuning and commonsense understanding of human behavior. In this work, we hypothesize that LLMs, guided by human feedback, can be used to formulate reward functions that reflect human implicit knowledge. We study this in three challenging settings -- autonomous driving, humanoid locomotion, and dexterous manipulation -- wherein notions of ``good" behavior are tacit and hard to quantify. To this end, we introduce REvolve, a truly evolutionary framework that uses LLMs for reward design in RL. REvolve generates and refines reward functions by utilizing human feedback to guide the evolution process, effectively translating implicit human knowledge into explicit reward functions for training (deep) RL agents. Experimentally, we demonstrate that agents trained on REvolve-designed rewards outperform other state-of-the-art baselines.