Forging Agentic AI: A Comprehensive Survey on the Symbiotic Convergence of Large Language Models and Reinforcement Learning

Abstract

Reinforcement learning (RL) has shown remarkable success in solving complex decision-making problems; however, it still suffers from fundamental limitations such as low sample efficiency, sparse rewards, and poor generalization. Recent advances in large language models (LLMs) have introduced new possibilities for overcoming these challenges by enhancing multiple components of the RL pipeline. This survey provides a comprehensive overview of LLM-augmented reinforcement learning (LLM-RL), highlighting the ways in which LLMs contribute to reward design, exploration, planning, state and action representation, policy learning, and generalization. We categorize integration strategies, analyze representative frameworks, and discuss key application areas including robotics, gaming, and virtual environments. Finally, we discuss current limitations, open research challenges, and future directions to highlight the potential synergy between LLMs and RL in building more general and adaptable autonomous agents.

Ähnliche Arbeiten