Image Jacobian matrix identification based on Kalman filtering and visual predictive control of manipulator

Abstract

In order to overcome modeling accuracy restriction in robot control, a sort of visual predictive control algorithm is proposed. The predictive model is based on image Jacobian matrix, which is estimated on-line through Kalman filtering. The tricky problem of constraint handling is taken into account, such as mechanical constraint and visibility constraint. The validity of the algorithm is illustrated through the experiments on NAO's arm, and the rapid identification of object is achieved by using Naomark label. Experimental results demonstrate that the proposed algorithm is feasible and the control performance is satisfactory.