Noises Adaptive Filtering Structure for Robotic Positioning Using Monocular Vision

Abstract

This paper has developed a new monocular vision feedback control strategy based on the adaptive filtering structure, for robotic positioning without noise parameters. In real robotic workshops, the perfect statistic knowledge of the the system noise and observation noise are not easy to be derived, thus a noise adaptive filtering structure is firstly studied for the nonlinear mapping on-line identification between robotic vision and motor spaces. In our finding, the Kalman recursive filtering gain is improved by a feed-forward neural network, in which the neural estimator dynamic adjustment its weights to minimization the error of vision-motor mapping estimation, do need not the knowledge of noise variances. Finally, the proposed vision servoing control based on adaptive filtering has been successful implemented in robotic positioning tasks, and the experimental results demonstrate its validity and practicality for a six-degree-of-freedom (DOF) robotic system.