Image Jacobian modelling for uncalibrated visual servoing of robots using support vector regression

A new modelling method of image Jacobian estimation is presented for uncalibrated visual servoing of robots, in which a support vector regression (SVR) technique is used for non-linear mapping between target image features and robot joint angles, and an image Jacobian expression is derived from the SVR equation with Gaussian kernel. The experiments of robot visual servoing with both eye-in-hand and eye-to-hand camera configuration are conducted using the SVR-Jacobian estimator which has been automatically trained by self-learning under the control of computer, two kinds of experimental results have shown that the robot visual servoing converges at the desired goal and high target-tracking accuracy can be acquired with the proposed modelling method.