Two new methods of Kalman filters family for nonlinear SLAM

In this paper, reducing the linearization error of Kalman filters family for nonlinear simultaneous localization and mapping (SLAM) problem is investigated and two new methods are presented to reduce the linearization error. Inaccuracy of the formula used to calculate the slope of linear approximation of observation function (H_k) leads to severe linearization error in Kalman filters family. These new methods named Mean Iterated Extended Kalman Filter (MIEKF) and Mean Stepwise Extended Kalman Filter (MSEKF) lessen the linearization error by revising the formula used to compute H_k. Simulation results, utilizing `Car Park Dataset´ demonstrate the effectiveness and reliability of our proposed methods. They illustrate that the best estimation accuracy belongs to MIEKF-SLAM and the method named MSEKF-SLAM comes to the second place in estimation accuracy point of view. In addition, our proposed methods are computationally efficient. Thus, as to both linearization error and computational complexity, MIEKF and MSEKF are more effective than other mentioned Kalman filters in this paper.