Research On Collision Avoidance Planning For Mobile Robot Based On Deep Reinforcement Learning

Mobile robots have broad prospects for development,and the collision avoidance planning algorithm is a key technology in the field of the mobile robot research.Traditional collision avoidance planning algorithms do not have the ability to respond quickly in the face of unknown and complex environments,and they rely too much on artificial rules.The deep reinforcement learning algorithm,as a branch of machine learning,can form a good strategy based on its own experience,which has strong self-learning ability.Using the deep reinforcement learning algorithm to complete the collision avoidance planning task for the mobile robot is one of the hot research directions in the future.Therefore,this thesis has carried out the research of collision avoidance planning for the mobile robot based on deep reinforcement learning.The main research contents of this thesis are as follows:Firstly,the research status of mobile robots and collision avoidance planning algorithms is summarized;and the inverse kinematics model of the Mecanum wheel robot and the perception model of the laser rangefinder are constructed.Secondly,in order to solve the problem of collision avoidance planning of the mobile robot,a collision avoidance planning algorithm for the mobile robot based on Proximal Policy Optimization(PPO)is designed,and the simulation environment for the collision avoidance planning task is built,including state space,reward function,and action space.In order to accurately control the mobile robot,the continuous action space is designed and the heading angle of the mobile robot is restricted.Then the collision avoidance planning algorithm for the mobile robot is simulated and verified in different obstacle environments.Thirdly,in order to enable the mobile robot to dynamically change its linear velocity during the obstacle avoidance process,the Dynamic Window Approach(DWA)method is designed to adjust the linear velocity of the mobile robot in real time,and the PPO collision avoidance planning method combined with speed control is proposed.In order to enable the mobile robot to adjust its linear velocity based on the distance from the obstacles,the objective function of the DWA is modified.The simulation results show that the PPO collision avoidance planning algorithm combined with speed control can make the mobile robot reach the target point in a shorter time.Finally,aiming at solving the problem that the Probabilistic Roadmap(PRM)algorithm is difficult to generate a feasible path in a complex environment due to the randomness of sampling,this thesis uses Shi-Tomasi corner detection algorithm to expand the initial set of landmark points.Then combined with the local collision avoidance algorithm,a double-layer collision avoidance planning method based on the hybrid algorithm is proposed.The improved PRM algorithm can ensure the global optimality of the path.And the PPO collision avoidance planning algorithm combined with speed control can ensure that the mobile robot will not collide with obstacles when following the global path,and can avoid unknown obstacles at the same time.The experimental results show that the double-layer collision avoidance planning algorithm designed in this thesis can satisfy the global optimality of the path and realize realtime planning.