Avoidance of singular localization environment using model predictive control for mobile robots

Abstract

Localization is important to achieve safety motion of autonomous mobile robots. In this paper, we propose a model predictive control which prevents vehicles from falling into a singular localization environment with less features. We assume to use a laser range finder (LRF) as a sensor to obtain two dimensional point-cloud data of the surrounding environment and apply map-matching method for localization. We can calculate a covariance matrix of localization error using Fisher information matrix. The maximum eigenvalue of this covariance matrix is used as an index of uncertainty of localization. Then, we extract regions having large uncertainty for localization to construct prohibited regions. The robot avoids singular environments and regions with significantly low estimation accuracy by considering the prohibited regions represented as inequality constraints for model predictive control (MPC). We show the effectiveness of the proposed method through numerical simulations which simulate singular environments indicating a corridor or a large floor where only degenerated information is available.