A Qualitative Approach to Sensor Data Fusion for Mobile Robot Navigation

Sensor based navigation is fundamental to any mobile robot. Conventional statistical approaches to the navigation problem maintain an exact global description of environment geometry. However, in practise, the behaviour of real physical sensors and the observations they make of the environment make such central geometric representations extremely fragile. To overcome such problems, this paper proposes the use of qualitative models of physical sensor observations. These aim to describe the world in terms of local sensor-centric representations of the observed environment. Each representation exploits those landmarks most natural to the physical sensor involved and no explicit geometric representation of the world is assumed. This leads naturally to a navigation process defined in terms of relationships between different sensor observables; an intrinsically more robust mechanism than found in conventional navigation algorithms. The representation and navigation methodology proposed is illustrated using sonar data from a real vehicle.