Spatial navigation with uncertain deviations

Abstract

We consider geometrical scenes with obstacles and landmarks that can't necessarily be distinguished and generalize the notion of panoramas (Sch93; Her94), introduced in the qualitative Spatial Reasoning (QSR) approaches to robot navigation. We study various notions of motion strategies in the accessibility graph associated with the local panoramas under uncertain deviations, a natural model of uncertainty for motion planning and navigation. We show that randomized motion strategies can be better than deterministic ones with a finite memory and stress the usefulness of random decisions for qualitative spatial reasoning.