Research On Path Planning Method Of Mobile Robot Based On Deep Reinforcement Learning

With the popularization of service robot,sweeping robot and automated guided vehicle,path planning,as the core of mobile robot technology,has become a research hotspot.In the face of complex environment,mobile robots need autonomous learning to complete the task of path planning.With the development of deep reinforcement learning in recent years and its outstanding performance in some fields,it is considered to apply deep reinforcement learning to the path planning of mobile robots.In this paper,a mobile robot equipped with RPLIDAR A2 lidar sensor is used to study the path planning of mobile robot based on deep reinforcement learning.Firstly,a path planning simulation system platform based on ROS and gazebo is established.Specifically,it includes: building a "back" scene through the gazebo editor to train the static obstacle avoidance algorithm;building three closed environments with the gazebo editor to train the trend target and the dynamic and static path planning algorithm by adding the target points and static or dynamic obstacles;using the URDF to design the bobac mobile robot simulation model,and then add the simulation robot model to the simulation scene,so as to complete the path planning simulation system experimental platform.Secondly,for the "back" scene built in the simulation system platform,a deep reinforcement learning-based algorithm is designed to study the static obstacle avoidance of mobile robots.The lidar's ranging data is used as the state input of the deep reinforcement learning algorithm.After processing by the deep reinforcement learning method,the motion of the mobile robot is directly output.It overcomes the problems of difficult to reproduce,difficult to adjust parameters,and difficult to converge under the condition of discrete action of deep reinforcement learning algorithms.The feasibility of the static obstacle avoidance algorithm based on DQN algorithm is verified by simulation experiments.Then,DQN algorithm and migration learning are combined to study the path planning of mobile robot.The specific steps are as follows: firstly,in the simulation scene with no obstacles and only path planning target points,the robot is trained to reach any target point based on DQN algorithm;then,by placing four static obstacles,the weight parameters trained before are used as the initial parameters of the simulation scene by migration learning,and the mobile robot is trained to learn the ability of static obstacle avoidance and trend target Secondly,by making the four obstacles move in a circle,the weight parameters trained before are used as the initial parameters of the simulation scene to train the mobile robot's ability of dynamic obstacle avoidance and target orientation;Finally,a real bobac mobile robot platform is built.SLAM method and deep reinforcement learning method are applied to the path planning experiments.The experiments show that the method based on deep reinforcement learning designed in this paper can solve the path planning problem of mobile robot.