Comm-CoT: Standardized Chain-of-Thought Communication Framework for Efficient LLM based Multi-Agent Decision-Making in Real-Time Strategy Games

In recent years, Large Language Models (LLMs) have become a hotspot in AI research due to their remarkable success in natural language processing. LLM-based Multi-Agent Systems have also achieved significant progress in tasks such as distributed decision-making. However, due to inherent limitations of LLMs, such as hallucination and lack of contextual information, their communication efficiency within multi-agent frameworks remains suboptimal, thereby constraining the overall decision-making capabilities of the system. To address this issue, we propose a standardized communication framework based on Chain-of-Thought (CoT) method, named Comm-CoT (Communication-Chain of Thought), and conduct experiments in a real-time strategy game environment using the LLM-PySC2 platform. On a specially designed multi-unit cooperative combat scenario, the experimental results demonstrate that Comm-CoT significantly improves the win rate of LLMs while enhancing communication efficiency and the ability of agents to utilize communication information in multi-agent systems. The code is open-sourced and available at https://github.com/NUDT-Decision-Team/LLM-PySC2-Comm-CoT.