LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning

In three linesLMAC leverages LLM reasoning to design communication protocols in MARL, enabling agents to reconstruct the underlying state uniformly and accurately. The approach iteratively refines protocols using an explicit state-awareness criterion. Experiments on MARL benchmarks demonstrate substantial performance gains over prior baselines.

Summary generated by Claude — human-verified