Experiments in Evidence Based Localisation for a Mobile Robot

Abstract

This paper addresses the problem of localisation in autonomous mobile robot navigation, i.e., the task of identifying places after prior exploration and mapbuilding by the robot. In particular, the work is concerned with the more general problem of relocalisation without using past experience (i.e., knowing roughly where you are to start with), referred to here as the lost robot problem. In the experiments presented here, the robot had to relocalise after being moved to a randomly chosen location, its sensors being disabled during that move. The robot therefore had no a priori knowledge of its position, and had to use current sensory perceptions and map knowledge alone to relocalise. A perception-based localisation method is presented which is resilient to the problem of perceptual aliasing (i.e., perceptual identity of distinct locations), and is capable of relocalising even in environments where no single place has a unique perceptual signature. During an exploration phase, ...