RLDISCOVER: LLM-driven co-evolution of reinforcement learning algorithms

2026-10-06

AI Summary

This study addresses the poor scalability and high evaluation costs associated with the joint search of reinforcement learning algorithm components by proposing a large language model (LLM)-driven self-evolutionary framework. Through LLM-guided program evolution, the framework advances from single-component editing to multi-component co-evolution. By integrating progressive training, probabilistic evaluation, and repeated verification mechanisms, it effectively balances search breadth with evaluation fidelity. Experimental results demonstrate that the proposed method improves returns by 32%–84% on benchmarks such as SAC while significantly reducing computational overhead. Furthermore, the framework discovers novel algorithms with enhanced interpretability.

Abstract

LLM-guided program evolution has enabled discoveries in mathematics and computational optimization, raising the prospect of reinforcement learning (RL) algorithms that self-evolve to improve how agents learn. However, realizing this prospect faces two obstacles. Joint search over coupled algorithmic components is difficult to scale: simultaneous changes can disrupt learning, while isolated changes overlook their dependencies. Evaluating candidate algorithms also requires costly training, with fitness remaining uncertain across random seeds. We introduce RLDiscover, a framework for the self-evolution of model-free deep RL algorithms. Progressive Co-Evolution advances from targeted component edits to joint evolution, while Progressive Probabilistic Evaluation balances search breadth and evaluation fidelity through staged training and repeated evaluation. Experiments across SAC, PPO, and DQN on four benchmark suites show substantial improvements in mean return, with per-family median gains of 32%-84% and a peak return ratio of approximately 363x over a near-zero baseline. These gains include transitions from failed learning to successful task completion, and improvements persist when evolution starts from stronger open-source implementations. On measured SAC locomotion runs, evaluation uses approximately one-fifteenth the estimated compute required to fully evaluate the same candidate pool. Remarkably, independent searches repeatedly discover interpretable combinations of adaptive robust losses, progress-dependent value targets, and running statistics, with selected programs transferring to unseen tasks. These findings point toward a broader role for self-evolution in AI: discovering interpretable algorithms that improve how agents learn.

Problem

• Reinforcement Learning
• Algorithm Self-Evolution
• Large Language Models
• Joint Search
• Evaluation Cost

Innovation

• Reinforcement Learning
• LLM-driven Evolution
• Progressive Co-Evolution
• Algorithm Discovery
• Probabilistic Evaluation