Information relative map going toward constant time SLAM

Abstract

The paper presents the Information Relative Map algorithm for solving SLAM. Instead of estimating directly the relative quantities as in other relative mapping approaches, the proposed algorithm estimates the canonical quantities, the information vector and information matrix, using the Information filter. The estimation algorithm has constant time complexity without any approximation or linearization. The correlation between observed quantities are fully taken into the estimation. Furthermore, only independent relative quantities from observations are mapped so that the required computation is significantly reduced. The algorithm is empirically evaluated by testing on more than 100 simulated problem instances and the real world Victoria park dataset. The comparison with an existing implementation of the FastSLAM and EKF algorithms clearly shows a better performance in map precision and speed. © 2008 Springer-Verlag Berlin Heidelberg.