Uncertainty representation for mobile robots: Perception, modeling and navigation in unknown environments

Abstract

We present a fuzzy-sets based approach to the problem of mobile robot navigation in unknown environments. Fuzzy sets are used to represent the uncertainty that is inherent to the perception of the environment through the robot sensors. This uncertainty is then propagated in the process of map building so that not only a plausible spatial layout of the environment, but also the confidence on this layout, is obtained. The initial map built by the robot is then used for self-localization as it continues navigating in the same environment. The new information collected by the sensors is matched to the initial map and the transformation that brings them together is used to correct and bound the dead-reckoning errors. Uncertainty representation is a key aspect of this process since it allows to immediately detect pairs of coincident boundaries, thus, permitting real-time self-localization. The approach is illustrated by experiments in office environments.