Research On Uncalibated Visual Servoing Control Method For Manipulator

With the application of robot in various fields,research on visual servoing technology has also been widely concerned.In this paper,an uncalibrated monocular robot is used as a model.It is not necessary to obtain the internal parameters and external parameters of the camera.The visual servoing task is realized only by the design and improvement of the control algorithm.Compared with the calibrated visual servoing system,not only the operation steps are simplified,but also the accuracy and anti-interference ability of the control algorithm are improved.Moreover,the uncalibrated visual servoing system is based on the robustness of the external environment.This paper solved the possible problems of pseudoinverse singularity,image noise,calibration error interference,and large calculation complexity with many feature points.First of all,this paper introduces the research background and current situation of visual servoing method,summarizes the common visual servoing structure in general,and analyzes their advantages and disadvantages respectively.Moreover,the transformation description of the visual servoing space coordinate system,the camera model and the kinematics modeling are introduced,which provides a theoretical basis for visual servoing research.Secondly,the uncalibrated visual servoing method using projection homography solves the problem that the calculation speed becomes slow when there are many feature points in the traditional method.In engineering,the increase of feature points will enhance the robustness of the image and improves the accuracy of the visual servoing task.In this paper,by using the homography mapping of the current image plane and the desired image plane,the task function is constructed and the homography Jacobian matrix with fixed dimension is generated,so that the calculation of the pseudoinverse of the Jacobian matrix does not change with the number of feature points.Compared with the traditional method,in the case of more feature points,the computational complexity of the Jacobian matrix is reduced,and the convergence speed is faster.In addition,two estimation methods of Jacobian matrix are introduced,which are kalman filter estimation method and adaptive estimation method respectively.Both can realize the estimation of Jacobian matrix and complete the visual servoing task.Thirdly,in order to solve the problem of pseudoinverse singularity problem,slow convergence rate,image noise and calibration error interference in visual servoing tasks,this paper proposes a novel visual servoing method using extreme learning machine and Q-learning.First,the extreme learning machine is used to approximate the pseudoinverse of the interaction matrix,avoiding the singularity of the interaction matrix,and effectively reducing the feature noise and camera calibration errors.Secondly,Q-learning method is used to adjust the gain to improve the convergence speed.Compared with other methods,extreme learning machine has better generalization performance,faster computing speed and unique optimal solution.In addition,Q-learning has the ability to self-learn,that is,without the need for empirical rules,the optimal decision can be obtained through training.Compared with the traditional fixed gain,Q-learning method adaptively tunes the gain,which increases the convergence speed of the visual servoing task.Finally,in order to verify the validity of the uncalibrated visual servoing method for the homography projection and the method based on the extreme learning machine and Q-learning,this paper uses the experimental system composed of the Denso robot platform to verify the experimental results respectively.