Neural Network-based Nonlinear Adaptive Trajectory Tracking Control For Two-wheeled-driving Planetary Rovers

With the in-depth development of planetary exploration,mission objectives of the planetary rover have become increasingly complex,and the working environment is becoming more and more demanding.The two-wheeled driving planetary rover is the most basic configuration of the planetary rovers.It can form a parent-child rover system together with a multi-wheeled parent rover.The child rover can then fulfill mobile exploration tasks in a complex environment in cooperation with the parent rover.Due to the complexity of the tasks and environment,the motion control system of the two-wheeled driving planetary rover is faced with many nonlinearities.Therefore,the design of the planetary rover control system faces nonlinear problems including state or input constraints,feedback signal time-delay,etc..They poses a challenge for the planetary rover motion control algorithm design,highlighting the necessity of research on adaptive control methods with strong nonlinear compensation or suppression.Neural network has strong self-learning ability.It can approximate the nonlinear links and uncertainties.Further,the influence of nonlinear problems can be eliminated through feedback compensation or feedforward suppression.Based on the dynamic modeling of a two-wheeled driving planetary rover considering wheel-ground interaction,this thesis seeks a method that can effectively deal with nonlinear problems such as constraints and time-delay.And also,the thesis seeks to solve the adaptive trajectory tracking control problem based on neural network.Based on the wheel-ground interaction,this subject established the dynamic model of two-wheeled driving planetary rover under slipping and skidding.Furthermore,the thesis considered the constraint nonlinearity caused by task requirements and input torque limitation,the planetary rover perception information solution and the communication delay,as well as the coupling nonlinear effects of constraints and time-delay.And the two-wheeled driving planetary rover model with constraint nonlinearity,the model with time-delay,and that in consideration of both inequality constraints and time-delay were established,respectively.The state and input constraint handling of the motion control system of the twowheeled driving planetary rover is a typical nonlinear problem,which increases the complexity of tracking control system design.Based on the approximation characteristics of the radial basis function(RBF)neural network,a neural network compensation method based on the Lyapunov stability theory was proposed.It overcomed the dead-zone constraint problem of the planetary rover.The adaptive tracking control algorithm design was completed for the input-constrained twowheeled driving planetary rover.A full-state constraint analysis method based on the Barrier Lyapunov function was proposed to solve the full-state constraint problem caused by the slipping and skidding of the two-wheeled driving planetary rover.The adaptive full-state constrained tracking control of two-wheeled driving planetary rovers was realized.The motion control system of the two-wheeled driving planetary rover suffers delay during the solution and transmission of perception information.It leads to a mismatch between the state and control signal of the controlled system,resulting in the deteriorated stability or even instability of the two-wheeled driving planetary rover motion control system.In order to solve the input-delay problem of the two-wheeled driving planetary rover,a neural network compensation method on basis of the input-delay separation lemma was proposed.And the input transition matrix and uncertain terms are approximated,so as to eliminate the input-delay effect.Based on the Lyapunov-Krasovskii functional,an adaptive optimal tracking control scheme for two-wheeled driving planetary rover(uncertain)time-delay system was proposed.and the scheme combines neural network-based pre-compensation methods for slipping and skidding terms and basic principles of reinforcement learning(RL),under the premise of boundedness guarantee of the time-delay deviation system.The coupling of constraints and time-delay further enhances the nonlinearity of the two-wheeled driving planetary rover motion control system.Considering the constraints and time-delay of the planetary rover control system,and according to the nonlinear constraint requirements of the system and input torque constraint equation of the planetary rover,the constraint information was converted into relaxation factor state information.An augmented system was constructed based on the original system equation,relaxation factor equation and torque precompensation equation.Furthermore,combined with the improved LyapunovKrasovskii functional method to deal with state uncertain time-delay,an online adaptive dynamic programming optimization scheme based on neural network was introduced to realize the adaptive optimization tracking control of the planetary rover time-delay system with inequality constraints.The experimental environment of two-wheeled driving planetary rover based on visual odometer and Opti Track visual capture system was established to carry out experimental verification of theoretical research results.The adaptive fullstate constrained tracking control algorithm for the two-wheeled driving planetary rover adopted the visual odometer to realize closed-loop speed feedback.While in terms of the adaptive optimal tracking control algorithm for the two-wheeled driving planetary rover with time-delay,and for those with both full-state constraints and time-delay,the closed-loop speed feedback through visual capture system was adopted.These three conditions were considered and verified by experiments,respectively.Through the analysis on the experimental results,the effectiveness of the designed control algorithm was verified.Also,the necessity of considering the single nonlinearity and the coupling problem was explained.In this thesis,based on the established two-wheeled driving planetary rover considering wheel-ground interaction system model,the effects of constraints and time-delay nonlinearity to the system model were explored deeply.The constraints or time-delay processing techniques such as neural network,Barrier Lyapunov function,Lyapunov-Krasovskii functional and augmented system method were combined together.And then,the adaptive tracking control system design of the two-wheeled driving planetary rover was accomplished to realize the adaptive tracking control of the two-wheeled driving planetary rover to the desired target trajectory under the nonlinear effect.The research in this thesis can be applied to the two-wheeled planetary exploration child rover,and meanwhile provides a reference for the motion control algorithm design of the four-wheeled or sixwheeled planetary exploration rover.