Application Research Of Reinforcement Learning In Robot Indoor Navigation

With the advancement of intelligent technology,the tasks performed by mobile robots have become more and more diversified in the areas like industrial production and life.The mobile robot must avoid obstacles in a timely manner and carry out autonomous navigation based on this in order to do the mission more effectively.The development of mobile robots currently offers a wide range of application possibilities for reinforcement learning technology.It can be used to enable mobile robots to navigate themselves without the need for prior knowledge when used for robot indoor navigation.The following are the primary research topics covered in this essay:(1)To solve these problems of poor sample utilization rate and low generalization capability in the reinforcement learning process of robots.On the basis of D3 QN algorithm,an empirical replay algorithm is proposed for the replay of experience samples.The track point with the highest advantage function value is chosen as the target point after computing the value of the advantage function for the track points in the track samples,and then the track samples are re-marked,and the old and new track samples are put into the experience pool together to increase the diversity of experience samples,so that the agent can learn to use the failed experience samples to learn,and realize the navigation to the target point more efficiently.The improved algorithm can effectively improve the utilization rate of navigation samples,reduce the difficulty of learning navigation strategies,and enhance the autonomous navigation ability and transfer generalization capability of mobile robots in different scenarios.(2)In order to deal with the issue of poor navigation ability of mobile robots in complicated environment,hierarchical reinforcement learning is added on the basis of advantage hindsight experience replay.Hierarchical reinforcement learning can decompose complex problems into several simple sub-problems through two-layer agent learning planning,so as to enhance the success ratio of navigation in complicated surroundings.To enhance the navigation success rate of the hierarchical reinforcement learning algorithm,the hierarchical reinforcement learning with the intermediate target points is introduced,and the navigation task is divided into two parts.The navigation of the overall point is carried out after realizing the navigation of the intermediate subitems,which enhances the navigation success ratio of the mobile robot.(3)In a simulation environment,the robot’s navigation mission is accomplished,and the effectiveness of the modified algorithm is confirmed.Add Turtlebot3 robots to the Gazebo simulation environment to execute navigation tasks in according to the various degrees of indoor environment complexity.The enhanced method was used to generate comparison experiments,ablation experiments,and comparison parameters after choosing the D3 QN algorithm as the baseline algorithm.A variety of comparison factors are utilized to demonstrate the algorithm’s efficacy.