Abstract: Information interaction methods have become a research hotspot in multi-agent reinforcement learning in recent years. With the aid of information interaction mechanisms, agents can acquire comprehensive information about other agents and the environment, thus alleviating the partial observability and non-stationary environment problems faced by multi-agent reinforcement learning. Starting from several existing challenges of multi-agent information interaction, this paper classifies and summarizes prevailing information interaction methods according to core solutions. Open issues and prospective research directions in this field are also discussed, providing references for further relevant research.