Online estimation technique for Jacobian matrix in robot visual servo systems

Abstract

Image Jacobian matrix is always required in image-based robot visual servo systems. With online estimation techniques for image Jacobian matrix, a precise system model would not be needed and the complex calibration process could be avoided. In this paper, the concept of total Jacobian matrix is proposed, which is mainly used to deal with the case that making a robot track a moving target. Filtering techniques especially particle filtering are utilized for online estimation of total Jacobian matrix, and thus a novel uncalibrated robotic visual servoing method is implemented. Both online estimation algorithms for total Jacobian matrix are experimented based on Kalman filter and particle filter respectively. The visual servo results on a 2 degree-of-freedom robot system show that the algorithm based on particle filter gives us much better performances than that based on Kalman filter.