Enhancing Multi-Agent Communication Collaboration through GPT-Based Semantic Information Extraction and Prediction
Xinfeng Deng, Li Zhou, Dezun Dong, Jibo Wei · 2024
In recent years, research has shown that effective communication among multiple agents can enhance collaboration. However, the volume of observation information in multi-agent systems is massive and redundant, posing significant challenges to direct transmission in communication systems. This paper proposes a multi-agent communication method based on GPT for semantic information extraction (GMAC), simultaneously utilizing GPT to generate predictions of actions in the next time step, aiding agents in making wiser decisions. This method is simple yet effective, enabling efficient and concise communication in multi-agent systems. GMAC leverages GPT for semantic information extraction, significantly reducing the amount of information exchanged between agents. Experimental results demonstrate that GMAC substantially reduces communication overhead while improving convergence speed and accuracy.