Optimal Pose Correction Technique for SLAM using Structural Regularity

Abstract

Formulation of cost function and selection of optimization method are the major aspects of optimization based SLAM, which has been addressed in this paper. Two cost functions namely (i) Map Oblique Error and (ii) Map Spread Error have been formulated for correcting the reconstructed map in SLAM. They are based on the structural regularities of the indoor environment and the robot dynamics, which remains to be highly reliable information. The cost functions provide a metrics to qualify the reconstructed map, which also represents accurate unobservable localization error. A multi-stage exhaustive enumerative (MSEE) optimization technique has been proposed for minimization of these cost functions. It divides the conventional exhaustive enumerative (EE) method into multiple stages. At each stage the search space is converged to increase the resolution. This reduces the number of iterations by 80% to minimize the cost function and with the same solution resolution as that of EE method. A typical indoor environment with generic parallel and/or orthogonal features and a two wheeled mobile robot with a single LIDAR sensor are considered for this study. Experimental results illustrate the efficiently of the proposed technique in improving SLAM accuracy.