Multi-agent pathfinding algorithm based on deep reinforcement learning
Multi-agent pathfinding algorithm has been widely applied to practical problems.Reinforcement learning based ap-proaches typically model multi-agent pathfinding as a partially observable Markov decision process in which agents make decisions independently based on local observations.The introduction of communication solves the problem of insufficient local observation information,but the existing communication schemes often ignore excessive information exchange between agents,resulting in addi-tional system overhead and latency.In order to improve the efficiency of information exchange,a communication based multi-agent path planning algorithm is proposed.Random communication discarding technology is applied to solve the problem of system insta-bility caused by excessive dependence on communication.Random communication dropout technique is applied to solve the prob-lem of system instability caused by excessive dependence on communication.Experimental results show that introducing random communication dropout can coordinate the behavior of agents more effectively.
万方数据
