Multiple hypothesis methods for robot self localisation.

The goal of the localisation system on a robot is to determine the location of that robot within an environment. The localisation problem used to evaluate the work within this thesis is based on the RoboCup soccer leagues. The basic localisation problem within these leagues involves an estimation of a location on a given map. This map being a soccer field. The measurements available for the localisation system include visual measurements obtained from processing the image obtained from the camera, as well as estimates of the robot actions. On the current soccer field used there are very few unique measurements available, leading to a need for multi-modal estimates. Typically the Kalman filter is unable to handle multi-modal estimates. Using the multiple-hypotheis Kalman filter method allows for muli-modal estimates by using multiple Kalman filters to represent multiple modes or hypotheses. This thesis explores the use of the Multiple Hypothesis Kalman Filter as a capable solution for the robot localisation problem in an environment with a limited number of uniquely identifiable landmarks. Additional Multiple Hypothesis Kalman Filter methods used in other applications are explored and compared. Finally a number of modifications and suggestions are made to the use of this technique in the context of a robot localisation system.