Vision-based Nonlinear Pose Estimation And Control Of Robots

With the development of science and technology,robots frequently appear in people's lives,and their application range has gradually expanded from the industrial field to life services,medical and health and other fields.The working environment of robots is becoming ever more complex and unstructured.As the "eye" of the robot,the visual sensor is an indispensable component for the robot to operate in an unstructured working environment.Therefore,visual servo system has become one of the hot research directions in the field of robot control,which is of great significance to improve the robot's adaptability to complex environments and autonomous control ability.Because of the robustness to system parameter uncertainty and image measurement noise,image-based visual servo technology has been extensively used in robot control.However,the traditional visual servo with image Jacobian matrix has some problems,such as small stability domain,easy to loss feature points and produce unexpected trajectory.Therefore,for the image-based visual servo systems,how to enhance the performance of the visual servo system,to ensure that feature points are always in view of camera,and to avoid obstacles in the unstructured environment are particularly important.To solve the above problems,this thesis has conducted in-depth research on vision-based robot pose estimation and control.The main research contents are as following:1.A continuous Riccati observer of position estimating of feature point is proposed,for the depth could not be perceived by monocular camera.The result of the observer could be used to improve the performance of visual servo system.Firstly,an extended Kalman filter is developed by using the robot motion information.Then,a continuous Riccati observer with variable gain is designed using the Lyapunov method,and a continuous Riccati equation for solving the observer gain is given.Finally,simulations and experiments verify the effectiveness of the proposed method.2.Two dynamic pose estimation methods respectively based on position and image are proposed,for the estimation of relative orientation between the camera and the target object.Firstly,when the three-dimensional information of the feature points is known,a nonlinear pose compensation filter in the SE(3)space is designed,and the performance of the observer at the undesired equilibrium point is improved by the switching strategy.Secondly,a moving horizon estimator is designed to estimate the relative orientation by using the image information when the position information of the feature points is unknown.Finally,simulations and experiments verify the effectiveness of the proposed methods.3.A visual servo control method using image Jacobian based model predictive control strategy is proposed to solve the problem of feature points leave the visual field in visual servo.Firstly,the image Jacobian matrix is utilized to predict the motion information of feature points,and the problem of visual field loss is expressed as an inequality constraint.Then,by defining the prediction performance funtion,the minimization problem with inequality constraints is given.Furthermore,a LOQO interior point method is utilized to solve the optimization problem with inequality constraints,and the iterative solution of the model predictive controller is given.Finally,simulations and experiments verify the effectiveness of the proposed method.4.A cascade visual servo control method using 3D motion model is proposed to overcome locally effect caused by image Jacobian.Firstly,a Asymptotically stable pose controller is designed on the virtual spherical surface by Lyapunov method.Secondly,a switching control strategy is presented to improve the attitude control performance of the controller.And then,the system is decomposed into a cascade structure.Furthermore,with the rotation control signals are already determined by the attitude controller,a model predictive control method for the translation signal with visual field constraints is proposed.Finally,the effectiveness of the proposed algorithm is verified by simulations and experiments.5.A path planning method using numerical optimization approach for a redundant manipulator is proposed,when there are obstacles in the workspace.Firstly,the obstacle avoidance constraint is expressed as a constraint located outside the envelope ellipse of the obstacle,and then the obstacle avoidance path planning problem is transformed into a kind of numerical optimization problem.Furthermore,considering the inverse kinematics solution of redundant manipulator,an inverse kinematics solution method based on iterative optimization is proposed to ensure the obstacle avoidance of the redundant manipulator in the joint space by transforming the path in the workspace into the path in the joint space.Finally,the effectiveness of the proposed algorithm is verified by simulations.