Multi-agent Collaborative Path Planning Based On Deep Reinforcement Learnin

Agent is an important research direction in the field of artificial intelligence.As the application scenarios of agent systems become increasingly complex,it is necessary for agents to have stronger path planning ability.The traditional method based on agent path planning technology has low efficiency when the number of agents is large and the environment is complex,so it cannot meet the current application requirements.To solve this problem,this paper uses deep reinforcement learning algorithm to study single-agent and multi-agent path planning.The main research achievements are as follows:(1)Aiming at the problem that the exploration rate of the existing deep reinforcement learning algorithm is not enough in path planning,this paper introduces maximum entropy into the near-end strategy optimization algorithm,and realizes a PPO deep reinforcement learning method with maximum entropy double clipping.This method makes the strategies in learning more random by maximizing entropy,so as to realize more exploration in the environment,so as to encourage the agent to discover more new paths.By tailoring the value objective function,the algorithm convergence is more stable.Simulation results show that the proposed method converges faster in single-agent path planning and takes fewer turns to reach the target point and is more stable.(2)In order to solve the problem of strategy overfitting in the existing multi-agent algorithm,a noise-assisting MAPPO algorithm is proposed,which collects the noise of each agent to establish the noise dominance value and noise value function,and uses it to disturb the dominance value and Critic value network.This method can not only avoid the overfitting problem caused by the deviation of the sampling dominance value and the non-stationarity of the environment,but also make use of the different noises of each agent to expand the exploration range of the multi-agent.The simulation environment of multi-agent cooperative path planning is built,and the state space and action space of multi-agent are set based on the environment,and the appropriate return function is constructed according to the characteristics of multi-agent cooperative path planning.The feasibility and effectiveness of the noise-assisted MAPPO algorithm are verified by the multi-angle comparison between the path planning of fixed target points and the path planning of non-fixed multi-target points and the MARL algorithm of CDQN,IPPO and MADDPG.