Model Predictive Control Of Visual Servo For Mobile Robot

With the rapid development of robot kinematics,image processing technology and computer hardware technology,robot vision research has become an important research direction of robotics.Vision-based control of robots uses real-time feedback of visual information to achieve robot trajectory tracking or stabilization control tasks.Besides,since the mobile robots has a very complex characteristics of nonholonomic constraint,and thus the corresponding control laws should be re-designed in a completely different way.Therefore,the research of mobile robot vision controller design and analysis under the condition of nonholonomic constraint is a very challenging problem in the field of robot and automatic control for the various uncertainties brought about by visual sensors.This paper discusses the technology of visual servo control of wheeled mobile robot.Considering that the traditional visual control method often neglects or simplifies the kinematics and dynamics constraints of the robot,and these constraints have a significant impact for the control performance.Thus,this paper mainly introduces the model predictive control method(MPC)to solve the problem of nonholonomic constraints,and uses the primal-dual neural network(PDNN)to solve the convex optimization problem.In this paper,according to the visual control characteristics of the robot,the vision stabilization system of the mobile robot is established based on the kinematics model of the nonholonomic constrained.In order to stabilize the system,the model predictive control is used as the control method.The model predictive control can convert objective function into a quadratic programming(QP)problem with constraints,and iteratively solve the optimal problem by using the primal-dual neural network to obtain the optimal control input in real time.Secondly,since the actual system exist the unknown disturbance which can lead to the stability of the system will be affected,thus,a dynamic state feedback controller is designed based on model predictive control method,and this new control method is called the tube-MPC.Then,a mobile robot vision formation of the leader-follower structure is established based on the research of single wheeled robot.Firstly,this thesis derives a visual tracking model of mobile robot,which combines the kinematics model of two-wheel mobile robot with an on-board camera.With the aid of transformations of image space and inertial space,the differential equations of visual tracking model are finally obtained.This model is the foundation of mobile robot control and multi-robot formation control.Moreover,it also can be used to estimate the relative position,orientation and velocity of the leading robot.Finally,the MPC method is used to control the visual formation system to reach the desired robot correlation.Based on the model predictive control,this paper realizes the visual servo of nonholonomic mobile robot,including visual stabilization,tube-based visual stabilization,and visual formation control.Finally,the validity and reliability of the visual servo are verified by three related experiments.