Research On Control Strategy Of A Variable Structure Wheeled Robot

Wheeled mobile robot has been widely used in many fields in recent years because of its simple structure,light weight,large load,dexterous and flexible,fast speed and other characteristics.However,wheeled mobile robot is a typical nonholonomic constraint system,and has the characteristics of nonlinearity,strong coupling and parameter uncertainty.Bicycle robot is a special motion mode of variable structure wheeled robot.In this paper,the approximate linear system and affine nonlinear system of bicycle robot are studied.For the approximate linear system of bicycle robot,this paper uses the balanced truncation algorithm to reduce the order,and uses the low-order system to design the LQR controller to realize the stable control of the linear system of bicycle robot.In this paper,the expected poles of the closed-loop system and the parameters of LQR controller are optimized by using twice genetic algorithm,which improves the approximation accuracy of the low-order system to the original linear system and improves the control effect.Aiming at the affine nonlinear system of bicycle robot,a T-S fuzzy model is established in this paper.Taking the approximation error of the TS fuzzy model to the first derivative of the state variable of the affine nonlinear model as the quantitative index,the central value parameters of the membership function of the T-S fuzzy system are adjusted by the batch gradient descent method to improve the approximation accuracy of the original affine nonlinear system.In order to reduce the computational complexity of T-S fuzzy system without reducing the approximation accuracy,the T-S fuzzy system is processed in sections,and then the T-S fuzzy controller is designed by using the parallel distributed compensation(PDC)method to realize the stable control of T-S fuzzy system.Finally,EDMD algorithm and Deep neural network-based Koopman algorithm were used to linearize a class of front-wheel drive bicycle robot.The approximation ability of the two Koopman algorithms and the local linearization algorithm is compared.Simulation results show that the Koopman algorithm based on deep neural network has higher approximation accuracy.