LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning

ICLR 2026 Conference Submission18465 Authors

Keywords: Multi-Agent Reinforcement Learning, Large Language Models, Multi-Agent Communication

TL;DR: We propose LMAC, an LLM-driven framework that designs interpretable communication and reliable latent representations, consistently improving MARL under partial observability.

Abstract: Communication can be essential in cooperative multi-agent reinforcement learning (MARL), where agents may need to overcome partial observability by exchanging information to accomplish tasks. However, prior methods often rely on messages that are uninterpretable or contain redundant information. To overcome this issue, we propose LLM-driven Multi-Agent Communication (LMAC), a novel MARL framework that combines LLM-based communication protocol design with a meta-cognitive latent representation module. LMAC employs iterative refinement with phase-specific feedback to produce interpretable protocols that enhance state recovery and shared understanding, while its latent module incorporates reliability signals with cycle consistency to ensure compact and trustworthy representations. Experiments across diverse MARL benchmarks demonstrate that LMAC consistently improves performance over other communication baselines.

Primary Area: reinforcement learning

Submission Number: 18465