Mobile Robot Navigation Using Active Vision

Active cameras provide a navigating vehicle with the ability to fixate and track features over extended periods of time, and wide fields of view. While it is relatively straightforward to apply fixating vision to tactical, short-term navigation tasks, using serial fixation on a succession of features to provide global information for strategic navigation is more involved. However, active vision is seemingly well-suited to this task: the ability to measure features over such a wide range means that the same ones can be used as a robot makes a wide range of movements. This has advantages for map-building and localisation. The core work of this thesis concerns simultaneous localisation and map-building for a robot with a stereo active head, operating in an unknown environment and using point features in the world as visual landmarks. Importance has been attached to producing maps which are useful for extended periods of navigation. Many map-building methods fail on extended runs because they do not have the ability to recognise previously visited areas as such and adjust their maps accordingly. With active cameras, it really is possible to re-detect features in an area previously visited, even if the area is not passed through along the original trajectory. Maintaining a large, consistent map requires detailed information to be stored about features and their relationships. This information is computationally expensive to maintain, but a sparse map of landmark features can be handled successfully. We also present a method which can dramatically increase the efficiency of updates in the case that repeated measurements are made of a single feature, permitting continuous real-time tracking of features irrespective of the total map size. Active sensing requires decisions to be made about where resources can best be applied. A strategy is developed for serially fixating on different features during navigation, making the measurements where most information will be gained to improve map and localisation estimates. A useful map is automatically maintained by adding and deleting features to and from the map when necessary. What sort of tasks should an autonomous robot be able to perform? In most applications, there will be at least some prior information or commands governing the required motion and we will look at how this information can be incorporated with map-building techniques designed for unknown environments. We will make the distinction between position-based navigation, and so-called context-based navigation, where a robot manoeuvres with respect to locally observed parts of the surroundings. A fully automatic, real-time implementation of the ideas developed is presented, and a variety of detailed and extended experiments in a realistic environment are used to evaluate algorithms and make ground-truth comparisons. 1