Research On Mobile Robot Path Planning Based On Deep Reinforcement Learnin

Path planning technology is one of the hot research areas of mobile robot.When the environment is simple and known,mobile robot path planning has been fully studied.However,in the face of complex and unknown environment,most of the existing mobile robot path planning algorithms cannot effectively explore the environment or make correct decisions,resulting in low efficiency and security of path planning.Therefore,path planning of mobile robot in complex and unknown environment has important theoretical significance and application value.The deep reinforcement learning algorithm has strong perception and decision ability,so it is applied to the path planning problem of mobile robot in complex and unknown environment.Firstly,this paper improves the laser scanning Simultaneous Localization And Mapping(SLAM)algorithm,and proposes a Cartographer algorithm with adaptive delayed loop detection for environmental mapping before mobile robot path planning.On this basis,a heuristic deep Q network algorithm based on greedy sampling is proposed,which is combined with the mobile robot path planning technology.Finally,the mobile robot path planning in complex and unknown environment is completed.The main work of this paper is as follows:1)The Adaptive Delay Loopback Detection Cartographer Algorithm(ADL-Cartographer)is proposed.The ADL-Cartographer algorithm in the laser SLAM algorithm is proposed to solve the problem that the Cartographer algorithm in the laser SLAM algorithm works in an environment of equal width,long distance or geometric symmetry and is easy to detect error loops.First,the variance of the loopback transformation matrix is estimated by adding the unit complex number matrix as the threshold value,and then it is compared with the norm of the multiplication of the loopback point transformation matrix.When it is greater than the threshold value,no loopback detection is carried out.Otherwise,loopback detection is carried out.Finally,through simulation and comparison in the ROS system,the improved efficiency and superiority of the ADL-Cartographer algorithm are verified,so it is adopted as the environment mapping algorithm before path planning.2)The Greedy Sampling Heuristic Deep Q-Network Algorithm(GSHDQN)is proposed.Aiming at the problems of slow convergence speed,uneven path planning and low sample utilization rate,the deep Q network algorithm is improved in this paper.Firstly,the improved gravity function of artificial potential field and target guided action function are introduced into the action guidance strategy of the deep Q network algorithm.At the same time,a segmental reward function is designed,so as to propose a heuristic deep Q network algorithm,which effectively reduces the number of collisions in the training process of the algorithm,improves the convergence rate of the algorithm and makes the planned path better.Then,combining the heuristic deep Q network algorithm with the improved priority sampling strategy,a greedy sampling heuristic deep Q network algorithm is proposed,which effectively improves the sample utilization rate.Finally,the path planning simulation of the above three algorithms is carried out under the PyTorch machine learning framework.The simulation results show that compared with the deep Q network algorithm,the average total iteration time of the proposed algorithm can be reduced by 28.0%,the average path length can be reduced by 34.7%,and the number of collisions can be reduced by 43.2%.