Control Of Four-Wheeled Mobile Robot Based On Gaussian Process Regression And Reinforcement Learning

In robot control tasks,the uncertainty in the model cannot be ignored and needs to be addressed in the control.Sliding mode control is a nonlinear control method,which can deal with uncertainty effectively and has strong robustness.However,sliding mode controller often has chattering phenomenon,which makes it unable to be applied to actual control systems.As an emerging method of machine learning,Gaussian process regression can be used to handle uncertainties in systems.Gauss process regression is applied to Sliding mode control,which can improve chattering while maintaining good control effect.Reinforcement learning is a learning method of machine learning that is widely used in robot control systems.For controllers,various control parameters can be adaptively adjusted using reinforcement learning to better complete the set tasks.In this paper,a sliding mode controller is designed to achieve the path tracking task of a four-wheel mobile robot.The Gaussian process regression algorithm is used to estimate the uncertainty,and reinforcement learning is used to adaptively adjust the controller parameters.The specific research work is as follows:Aiming at the path tracking task of the four-wheel mobile robot,two kinds of vehicle robot structures of the four-wheel mobile robot are designed,the kinematics model is deduced,and the model is simplified according to the actual control requirements.The error is defined according to the path tracking task,and the error model is deduced based on the kinematics model.Considering the uncertainty in the model,the uncertainty is introduced into the model,and the kinematics model and error model with uncertainty are obtained respectively.The principle of Gaussian process regression was derived,and data-driven modeling and uncertainty estimation based on Gaussian process regression were implemented,respectively.The estimation of uncertainty in the error model was also completed.Four sliding mode controllers for four-wheel mobile robots are designed,which are nested sliding mode controller,quasi continuous sliding mode controller,linear sliding mode controller and terminal sliding mode controller.Gauss process regression is used in linear sliding mode controller and terminal sliding mode controller to estimate the uncertainty,and control chattering is suppressed.The effectiveness of the four designed controllers was verified through numerical simulation,and the controllers were compared through experimental results.Based on the terminal sliding mode controller,an adaptive terminal sliding mode controller based on reinforcement learning is designed.Based on the path tracking task,the design and construction of a reinforcement learning environment were completed,and the model was trained.A simulation environment and physical prototype of a four wheeled mobile robot were built,and the effectiveness of the designed adaptive controller was verified through experiments.This paper explores the application of Gaussian process regression and reinforcement learning in the field of robotics,and mainly studies the uncertainty processing in robot control system.