Learning Communication With Limited Range in Multi-Agent Cooperative Tasks
Chengyu Ning, Guoming Lu, Xuewan He, Aiguo Chen, Guangchun Luo · IEEE Transactions on Consumer Electronics · 2024
In multi-agent systems, instability and partial observability will bring various problems. Communication is an effective way to solve these problems, but when the number of agents in the environment is large, it becomes a key problem to weigh the cost and effect of communication. Therefore, we put forward a model, and designed a communication mechanism in the model, so that agents can learn effective and efficient communication in the partially observable distributed environment of MARL with the lowest communication cost as much as possible. Finally, we show the advantages of our model in multi-agent writing navigation scenarios, in which agents can formulate more coordinated and complex policies than existing methods.