LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning

2026-05-18

AI Summary

This work addresses the challenges of inefficient communication and insufficient state information sharing in multi-agent reinforcement learning (MARL) by introducing, for the first time, large language models (LLMs) to guide communication protocol design. Leveraging the reasoning capabilities of LLMs, agents generate concise and informative messages that enable accurate and consistent reconstruction of the environment state. The authors further propose an explicit state-awareness criterion to iteratively refine the communication protocol. This approach significantly enhances both state reconstruction fidelity and downstream task performance, outperforming existing communication mechanisms across multiple MARL benchmarks. The results demonstrate that LLM-informed protocols facilitate more efficient state sharing and improved inter-agent knowledge alignment.

Abstract

Communication is a key component in multi-agent reinforcement learning (MARL) for mitigating partial observability, yet prior approaches often rely on inefficient information exchange or fail to transmit sufficient state information. To address this, we propose LLM-driven Multi-Agent Communication (LMAC), which leverages an LLM's reasoning capability to design a communication protocol that enables all agents to reconstruct the underlying state as accurately and uniformly as possible. LMAC iteratively refines the protocol using an explicit state-awareness criterion, improving state recovery while narrowing differences in agents' knowledge. Experiments on diverse MARL benchmarks show that LMAC improves state reconstruction across agents and yields substantial performance gains over prior communication baselines.

Problem

• multi-agent reinforcement learning
• communication
• partial observability
• state information
• information exchange

Innovation

• LLM-guided communication
• multi-agent reinforcement learning
• state reconstruction
• communication protocol
• partial observability