Research On Robotic Arm Visual Servo Control Based On Image

Robotic arm servo control is an important research direction in robotics field.With the rapid development of industry,the application of industrial robotic arm is more and more widely.The uncalibration image-based visual servoing strategy can effectively avoid the problems of inverse kinematics and camera calibration.The servo control system of this method has the advantages of simple structure,lower computational complexity and higher control accuracy.This paper focuses on the problems of image Jacobian matrix online identification and the mapping between image feature space and robot motion space.The main content of the thesis is divided into two parts.Part one,the Extreme Learning Machine(ELM),which is a training method of single hidden layer feed forward neural network,is used to in the strategy of image Jacobian estimation.The ELM is used to optimize the image Jacobian estimation algorithm which based on Kalman filter(KF).The algorithm names KFELM.And a robot vision servo system based on fuzzy logic(FL)gain adaptive and KFELM based image Jacobian matrix estimation is designed.This method solves the problems of slow convergence rate and low estimation accuracy of image Jacobian matrix in KF-based robotic arm visual servo system.Part two,a hybrid model of multivariate adaptive regression spline(MARS)and online sequence Extreme Learning Machine(OS-ELM)is used to predict the velocity of the end-effector of a manipulator.Multivariate adaptive regression splines is used to evaluate the importance of training sample features.Compared with the methods based on state estimation,this method avoids the problems of kinematics,inverse kinematics,depth information and improves the real-time performance of the algorithm.Around the above research,the innovation of this article is as follows:(1)Aiming at the problems that the KF-based uncalibrate hand-eye system is unstable to the system disturbance noises and the statistical error of the noises and the system convergence rate is slow,an image-based visual servo(IBVS)control system with KFELM and FL gain adaptive is proposed.Firstly,an online identification model of image Jacobian matrix based on KF is established.Then,in order to solve the KF algorithm robustness problem of errors which cause by disturbances in robot system and statistical errors in noise which due to colors noise,an error compensation model based on ELM is established to compensate the error of the suboptimal estimation by KF.Finally,according to the feedback of image features error,a gain adaptive method of FL speed controller based on image feature error feedback is designed and a visual servo control scheme based on FL and KFELM(FL-KFELM)is proposed.The proposed method is applied to the PUMA 560 6-DOF manipulator control with Eye-in-Hand model.The experimental results show that the FL-KFELM based visual servo comparing to other algorithms has significant advantages in the convergence rate and the estimated accuracy of the image Jacobian matrix.(2)Aiming at the problems of getting the exact image Jacobian matrix hardly,convergence rate slow and poor real-time visual servo in the classical IBVS system,a hybrid regression model MOS-ELM,which is applied into the IBVS system,based on MARS and OS-ELM is proposed.In order to improve the predictive performance of OS-ELM,MARS is used to evaluate the importance of input features,then key features as input features of OS-ELM training.Compared with the simulation results of OS-ELM and some improved OS-ELM,the proposed MOS-ELM is proved to have better generalization performance and stability.Finally,the proposed method is applied to the PUMA 560 6-DOF manipulator with Eye-in-Hand model to verify the excellent performance of the method.