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

In recent years,mobile robots have been widely used in people's production and life.Path planning is one of the key technologies in the field of mobile robots.It is of great significance for the realization of autonomous movement of robots.Its research direction is developing towards intelligence.Deep reinforcement learning is a branch of the field of machine learning.It has excellent performance in decision-making problems and can be well used to solve the path planning problem of mobile robots.This paper combines deep reinforcement learning with the path planning of mobile robots,and focuses on the core problem of path planning,and carries out certain research on the path tracking and map construction of robots.The main contents are as follows:(1)A four-wheeled mobile robot model is built under the ROS system,and the automatic tracking of the predetermined path of the robot is completed using a pure tracking algorithm.First,the pure tracking algorithm is simulated,and the effect of the algorithm on path fitting under different conditions is analyzed and compared.Later,under the ROS system,the 3D model of the mobile robot was written through XACRO files,and the motion simulation was performed in the Gazebo 3D physical simulation environment to analyze the feasibility of the pure tracking algorithm for the control of the four-wheel mobile robot in this paper.Finally,the robot is controlled in a simulation environment so that the robot can move autonomously according to the movement path planned by the deep reinforcement learning algorithm.(2)Use the laser SLAM method to perceive the surrounding environment of the robot to meet the positioning requirements in the path planning process.In Gazebo,a threedimensional environment for path planning of a mobile robot is built.Three laser SLAM algorithms,Gmapping,Hector?SLAM,and Catrographer are used to create a map of the environment,and the obstacle coordinates in the environment where the robot is located are obtained from the map.In order to further compare the accuracy of different SLAM algorithms,the Turtle Bot3 mobile robot is used for experiments in an indoor environment to analyze and compare the accuracy of the three SLAM algorithms.(3)A two-dimensional environment for path planning of mobile robots is built,and Deep Deterministic Policy Gradient(DDPG)deep reinforcement learning algorithm is used for path planning of mobile robots.Use the Tensor Flow to complete the construction of the neural network of the DDPG deep reinforcement learning algorithm.Use the Pyglet in Python to write a dynamic experimental environment for path planning,and train the neural network in this environment to achieve deep reinforcement learning Visualization of training process and training results.(4)The use of artificial potential field method to set the reward function of deep reinforcement learning is proposed.In the deep reinforcement learning path planning environment built in this paper,the reward function of the artificial potential field method is compared with the reward function only referring to the information of the mobile robot from the target point,and the reward function proposed in this paper is analyzed in terms of neural network training speed and accuracy.