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

As one of the important branches of intelligent robots,mobile robots have been researched as hot spots by scholars at home and abroad regarding their path planning.In recent years,with the rapid development of artificial intelligence and intelligent manufacturing technology,mobile robots have been combined with various intelligent algorithms,and the application fields have been expanding,but also facing some challenges.At present,most scholars are researching on path planning algorithms in known environments.Mobile robots lack selflearning ability.When facing unknown environments,it is difficult to find a path to the destination without collision,and most of them are output discrete action.It does not conform to the scene application.Therefore,this thesis applies advanced deep reinforcement learning algorithms to the path planning problem of mobile robots,enabling mobile robots to explore and learn in unknown environments by reinforcement learning algorithms,train their decisionmaking capabilities,and achieved the path planning and obstacle avoidance under continuous action space.The main research of this article is as follows:First,this thesis introduced the principle of strategy-based reinforcement learning algorithms and explained the related theories of twin delayed deep deterministic policy gradient algorithm(TD3).The I-TD3 algorithm is proposed to improve the deficiencies in TD3,which solves the problems of low sample sampling efficiency and weak continuity of action space exploration in the TD3 algorithm by prioritized experience replay and OU exploration noise.In order to test the effectiveness of the algorithm,use the reinforcement learning development platform Open AI Gym to conduct an Inverted Pendulum-v2 experiment on the improved I-TD3 algorithm.Experiments results show that under the same conditions,the I-TD3 algorithm reduces the time required for training and improves the stability of the algorithm.Secondly,in order to apply reinforcement learning algorithm to robot path planning and achieve good effect of path planning in the continuous motion space,this thesis established a motion model of a two-wheel differential mobile robot,and built an autonomous path planning algorithm framework based on I-TD3.The state space,action space,reward function and overall process of the algorithm are designed in the framework.Also,the deep neural network structure of the algorithm is designed to make the algorithm framework more in line with actual demand by select the appropriate activation function.Finally,the experimental design and result analysis of the proposed autonomous path planning algorithm are carried out to verify its feasibility and effectiveness.The experiment is based on the ROS software platform,choose Gazebo as the simulation platform and Turtlebot3 as the experimental object,then setting up static and dynamic obstacle experimental scenes.After setting the experimental training process and experimental parameters,the I-TD3 algorithm and the TD3 algorithm are respectively carried out in two scenarios for path planning experiments,then analyze and compare experimental results.The experimental results show that in an unknown environment,the improved I-TD3 algorithm improves the learning efficiency of the agent on experience samples,accelerates the convergence speed of training,and the path for successful planning is also shorter,which has better path planning than TD3 performance.