Exploring Topological Environments

Simultaneous localization and mapping (SLAM) addresses the task of incrementally building a map of the environment with a robot while simultaneously localizing the robot relative to that map. SLAM is generally regarded as one of the most important problems in the pursuit of building truly autonomous mobile robots. This thesis considers the SLAM problem within a topological framework, in which the world and its representation are modelled as a graph. A topological framework provides a useful model within which to explore fundamental limits to exploration and mapping. Given a topological world, it is not, in general, possible to map the world deterministically without resorting to some type of marking aids. Early work demonstrated that a single movable marker was sufficient but is this necessary? This thesis shows that deterministic mapping is possible if both explicit place and back-link information exist in one vertex. Such 'directional lighthouse' information can be established in a number of ways including through the addition of a simple directional immovable marker to the environment. This thesis also explores non-deterministic approaches that map the world with less marking information. The algorithms are evaluated through performance analysis and experimental validation. Furthermore, the basic sensing and locomotion assumptions that underlie these algorithms are evaluated using a differential drive robot and an autonomous visual sensor.