Noise Covariance Identification Based Adaptive UKF with Application to Mobile Robot Systems

Abstract

A novel adaptive unscented Kalman filter (UKF) based on dual estimation structure is proposed. The filter is composed of two parallel master-slave UKFs, while the master one estimates the states and the slave one estimates the diagonal elements of the noise covariance matrix for the master UKF. By estimating the noise covariance online, the proposed method is able to compensate the errors resulting from the change of the noise statistics. Such a mechanism improves the adaptive ability of the UKF and enlarges its application scope. Simulations conducted on the dynamics of an omni-directional mobile robot indicate that the performance of the adaptive UKF is superior to the standard one in terms of fast convergence and estimation accuracy.