The problem of programming a robot t o carry out a systematic exploration of its environment using realistic sensors is considered in this paper. The robot is modelled as a single point moving in a twodimensional configuration space populaied with visually opaque and transparent obstacles. The robot is equipped with proximity sensors, a vision-based recognition system, and a method of odometry, all of which have some uncertainty associated with their measurements. By using visually distinctive configurations of features in the world as natural landmarks, a series of local maps is constructed. The geometrical relationships between mutually visible landmarks are used to build a relational map from this collection of local maps. This novel representation forms the basis for a systematic exploration algorithm. The approach has been implemented in simulation, and results are presented.
Camillo J. Taylor, David J. Kriegman