Reinforcement Learning Control Of A Variable Structure Wheeled Robot

At present,robots are widely used in every corner of our lives,and the demand for variable structure wheeled robots is becoming more and more urgent.In this paper,the dynamic model of the robot in bicycle mode is comprehensively analyzed based on the analysis of chaplekin dynamic model of the robot.Then the linear motion control algorithm of robot bicycle mode is designed,and the simulation experiment is designed to verify the effectiveness of the control algorithm.Finally,due to the modern demand of intelligent robot,this paper carries out reinforcement learning balance control for variable structure wheeled robot,so as to improve the adaptability of variable structure wheeled robot to the environment.The main content is divided into four parts,as follows:The thesis first conducts an academic investigation on the research status of variable structure wheeled robots,including bicycle robots,Segway robots and AC reinforcement learning control algorithms.Then the comprehensive analysis of the dynamic model of the robot in bicycle mode is completed.Firstly,the chaplekin dynamic model of the robot with the front fork not fixed is analyzed.The affine nonlinear system of the robot can be obtained by the identical deformation of the dynamic model in bicycle mode,and the feasibility of linearization of the system is analyzed.After that,the dynamic model of chaplekin with fixed front fork is analyzed.Then the linear motion control algorithm of the robot in bicycle mode is designed.Firstly,the state variables of the front fork angle in the dynamic model are approximately linearized,and the analysis method of the front fork unfixed part in Chapter 2 is used to decouple the system after approximately linearization.The cascade sliding mode control method is used to design the linear motion controller of the robot,and the simulation results verify the effectiveness of the algorithm.Finally,based on the modern requirements of robot intelligence,the variable structure wheeled robot is controlled by reinforcement learning.First,according to the requirements of the intensive learning control simulation environment of the variable structure wheeled robot,the experimental environment based on the MuJoCo simulation engine is built and designed.Then design the physical model of the variable structure wheeled robot based on the MuJoCo environment.Immediately use the AC reinforcement learning algorithm to complete the reinforcement learning control simulation of the robot.Finally,the establishment of the robot data inventory and the design of the WiFi information transmission scheme are completed.