The Research And Implementation Of Wheeled Robot Navigation And Motion Control Base On Reinforcement Learning

With the widening of mobile robots application area,mobile robots are required to work in an unknown environment.The autonomous navigation and trajectory tracking capabilities of mobile robots are key technologies to improve the adaptability of mobile robots.This article takes differential wheeled mobile robots as the research object.The autonomous navigation and tracking optimization method of mobile robot in unknown environments are studied based on reinforcement learning.By using the proposed navigation method and monocular vision technology,the designed mobile robot's reactive navigation is realized.A reactive navigation strategy based on reinforcement learning was proposed.Firstly,the Markov model of mobile robot navigation established through analyzing the characteristics of mobile robot navigation environment.In the Markov model,a novel design method of state space and reward function are proposed to increase algorithm's performance.In order to solve the problem of "dimensional disaster" in continuous state space,BP neural network is used to approximate state value and strategy space.Mobile robot learning method adopts Actor-Critic strategy gradient algorithm to avoid the slow convergence rate problem of Q-learning algorithm.Finally,the Matlab simulation experiment is designed to validate the performance of this algorithm.The conducted experimental simulation scenarios indicate that the proposed navigation method make mobile robot plan path and avoid obstacle automaticlly.Compared with traditional methods,it has higher learning efficiency,stronger generalization ability and environmental adaptability.The mobile robot trajectory tracking control method based on reinforcement learning algorithm is proposed to solve the problem of wheel-type mobile robot trajectory tracking.Firstly,the mobile robot kinematic model and dynamics model are established.Then the heuristic dynamic programming iterative control method is used in the trajectory tracking controller to realized online learning optimization.The mobile robot control experiment is implemented using Matlab software.The result shows that the designed controller can effectively improve the trajectory tracking accuracy,and has good stability and robustness.A mobile robot visual navigation system is designed.A monocular vision processing method is proposed to extract obstacle depth information.The method mainly uses image segmentation Kmeans++ algorithm and edge detection technology to extract obstacle edge features,and then calculates the distance from obstacles to camera based on the camera imaging principle.A small differential steering mobile robot platform was designed and its hardware and software system is built.The visual navigation experiment was carried out on this platform.It was proved that the proposed navigation algorithm can avoid obstacles in real time according to the obstacle information acquired by mobile robot monocular vision.