Research On Agent Autonomous Navigation Technology Based On Reinforcement Learning

The autonomous navigation of the agent aims to enable the agent to move safely from the starting point to the target point like a human,without colliding with other obstacles.This technology is the basis for mobile robots to achieve more advanced functions and is one of the focuses of research in the field of artificial intelligence.With the continuous expansion of robot application scenarios,traditional intelligent navigation technology has gradually failed to meet actual needs.This article focuses on the autonomous navigation of the agent and uses deep reinforcement learning technology to conduct research.Firstly,this paper studies the autonomous navigation method of single agent based on deep reinforcement learning technology.Aiming at the disadvantage that DDPG can only use a small amount of status information when implementing autonomous navigation of a single agent,a short-and long-term memory network is introduced to encode the historical information passed by the agent to improve navigation effect.By building a simulation experiment environment on the Stage-ROS platform,the performance of the above algorithm in a single agent autonomous navigation scenario was tested.Experimental results show that the improved algorithm(LSTMDDPG)has better performance than the DDPG algorithm.Secondly,this paper studies the multi-agent autonomous navigation method based on deep reinforcement learning technology.In view of the feature that the MADDPG algorithm does not have a global view of the system when it is distributed,the long-term and short-term memory network is introduced on the basis of the MADDPG algorithm to encode the historical state information of the system.This coded information is passed to the strategy network to improve the performance of the MADDPG algorithm in multi-agent autonomous navigation scenarios.The method proposed in this paper can improve the "catastrophic forgetting" problem of the DDPG algorithm and the shortcoming of the MADDPG algorithm when it is executed,Improve the autonomous navigation performance of the agent,It has a certain role in promoting the research and practical application of autonomous navigation of agents.