Research On Intelligent Robust Control Of Nonholonomic Mobile Robots

Wheeled mobile robot is not only a typical multiple-input multiple-output dynamic system with nonholonomic constraints but also an underactuated nonlinear system. The common smooth feedback control law can not apply to this kind of systems, especially when the system is affected by load variation, friction, external disturbances and other uncertainties, it is very difficult to find a common and effective control approach to achieve motion control. Therefore, the control problem of nonholonomic mobile robot system with uncertainties has attracted great attention and concern of domestic and foreign experts in control field.In this dissertation, domestic and overseas researches on motion control of nonholonomic wheeled mobile robot are analyzed thoroughly, and the common methods and existing problems are summarized. Based on the former studies, tracking control of nonholonomic wheeled mobile robot is implemented, and formation control of multiple robots based on trajectory tracking is studied. Main work of this paper includes the following issues:Firstly, the concepts of the nonholonomic constraints and dynamic nonholonomic systems are introduced comprehensively, and a typical nonholonomic wheeled mobile robot model which describes the basic properties and forms of motion control is established. As for the trajectory tracking control problem of nonholonomic wheeled mobile robot with load changes and external interference, the sliding mode controller which overcomes the robot parameters and non-parametric uncertainty is designed; the adaptive fuzzy sliding mode dynamics control algorithm of nonholonomic mobile robot is proposed based on that. Adaptive shunting kinematics is applied to solve the problem of controller speed jump caused by large changes in the initial position error; at the same time, the gain of sliding mode control is adjusted by using adaptive fuzzy control algorithm, which not only enhances the ability of random uncertainty adaption, but also eliminates the input chattering of sliding mode control.Secondly, further research is conducted on trajectory tracking of nonholonomic mobile robot without an accurate model. Based on the analyses of universal approximation properties of the neural network radial basis neural network, a double-adaptive neural sliding mode hybrid control law is designed through combining the neural network with sliding mode control. An adaptive neural network model is applied to approximate the unmodeled part of the robot system, and another adaptive neural network is used to adjust the gain of sliding mode switch control, which eliminate the chattering of sliding mode control and achieve a precise tracking control of the mobile robot without an accurate model.Thirdly, an adaptive dynamic recurrent fuzzy neural network sliding control algorithm based on genetic optimization is proposed. Segmented adaptive mutation operator is applied to improve the mutation operation, and the optimal parameters of the kinematics controller are selected by using genetic algorithm including the above mutation operation. The dynamic multi-input multi-output recurrent fuzzy neural network is designed to achieve online estimation of dynamic nonlinear uncertain part of mobile robot system, which makes the estimation error of the uncertainty reduce greatly; the uncertainty interference of the mobile robot are decreased excellently and the input chattering of sliding mode control is eliminated through combining the designed neural network with the adaptive robust controller, hereby ensuring the fine accuracy of trajectory tracking.Finally, the studies are extended from trajectory tracking of a single robot to formation control of multiple nonholonomic mobile robots. Based on the basic principles of leader-follower, taking account of the impact on the formation control caused by the uncertainty of single robot's dynamics and the formation dynamics, the controller is designed separately based on the conditions that actuator mode included in the dynamics model and not included. As for dynamic model, an adaptive sliding mode formation control which combines neural network with sliding mode control is proposed to remove the chattering of sliding mode control and overcome the disturbances of formation caused by the uncertainties of a single robot and formation dynamics. The other case is adaptive formation control problem of mobile robot including actuator dynamics. By using backstepping methodology, the dynamics of the robot including actuator model is introduced to the controller. RBF neural network is used to model uncertainties of formation control including changes in the load, friction (random change), external disturbances and unmodeled part of the formation dynamics. The formation control of multi-nonholonomic mobile robot including actuator dynamics is implemented.