Autonomous navigation among large number of nearby landmarks using FastSLAM and EKF-SLAM - A comparative study

This paper compares two commonly used algorithms to solve Simultaneous Localization and Mapping (SLAM) problem in order to safely navigate an outdoor autonomous robot in an unknown location and without any access to a priori map. EKF-SLAM is considered as a classical method to solve SLAM problem. This method, however, suffers from two major issues; the quadratic computational complexity and single hypothesis data association. Large number of landmarks in the environment, especially, nearby landmarks, causes extensive error accumulation when the robot is traveling along a desired path. The multi-hypothesis data association property and the linear computational complexity are essential features in FastSLAM method. Those features make this method an alternative to overcome mentioned issues. The FastSLAM algorithm uses Rao-Blackwellised particle filtering to estimate the path of the robot and EKF-SLAM method to estimate locations of landmarks. In case of FastSLAM applications, however, observation noise needs to be reconsidered if the motion measurements are noisy while the range sensor is noiseless. This study suggests optimization of a specific situation of FastSLAM algorithm in case of noise discrepancy.