A Comparison of the EKF, SPKF, and the Bayes Filter for Landmark-Based Localization

Abstract

The conventional approach to nonlinear state estimation, the Extended Kalman Filter (EKF), is quantitatively compared to the performance of the relative newcomer, the Sigma-Point Kalman Filter (SPKF). These approaches are applied to the problem of localization of a mobile robot using a known map, and compared under the context of the practical best performance of a Bayes Filter-type method using a particle filter with a very large number of particles.