Research On Navigation Technology Of The Mobile Robot Based On Deep Reinforcement Learning

With the continuous development of artificial intelligence technology,the functions of mobile robots are becoming more and more powerful,and they are gradually emerging in all walks of life.In practical applications,the working scenarios faced by mobile robots are often unknown and dynamic,and maintaining good navigation performance has become a challenge to be solved.To solve this problem,this paper designs an improved navigation algorithm based on deep reinforcement learning.First,the algorithm introduces Point of Interest(POI)as a local navigation point to realize the dismantling of the global navigation task;second,the improved POI selection rules reduce the collision between the robot and the glass wall during the navigation process;finally,the design improvement The reward function optimizes the navigation strategy and realizes obstacle avoidance navigation in real-world environments.The main work of this paper is summarized as follows:In the navigation task,the actions output by the deep reinforcement learning algorithm will show the disadvantage of local optimum,which will cause the robot to easily fall into the local optimum environment and cannot escape,resulting in the failure of navigation.In response to this problem,this paper first splits the navigation framework into two parts: global navigation based on the current environment and POI selection of local navigation points,and local obstacle avoidance navigation based on deep reinforcement learning.The best waypoints are selected from POI as local navigation points to improve the problem that the robot cannot escape after falling into the local optimal environment.At the same time,by improving the POI setting mechanism,the collision between the robot and the glass wall during the navigation process is reduced.Second,an improved reward function is designed for the deep reinforcement learning algorithm to optimize the navigation path.Furthermore,under the Robot Operating System(ROS)framework,the Gazebo simulation software is used to train and evaluate the navigation strategy based on deep reinforcement learning.Finally,in the real environment,Turtlebot2 is used as the experimental platform to verify the performance of the navigation strategy in the real environment.This method obtains the features strongly associated with the action value function from the data of the lidar and the odometer,and realizes the direct mapping from the sensor data to the control strategy.The experimental results show that this method can effectively avoid obstacles,reduce the collision between the robot and the glass wall,plan the motion path reasonably,and complete the robot navigation task.