| A fundamental capability of a truly autonomous mobile robot is its ability to build a map of its environment using the data gathered during navigation. The problem of learning how to build such a map is usually called the Simultaneous Localization and Mapping (SLAM) problem, as the locations visited by the robot need to be known in order for it to build a map. This thesis develops a statistical solution to the SLAM problem.; We formalize the SLAM problem using a Graphical Representation. The data correspond to the locations visited by the robot, obtained from a noisy odometer, and the distances to the closest obstacles from each of those locations, obtained from a noisy laser sensor. The map corresponds to an occupancy grid. We introduce the set of true locations visited by the robot, and true distances to obstacles, as latent variables.; Our formulation of the problem leads naturally to the estimation of the posterior distribution of maps given the data. We exploit particular factorizations of this distribution which allow us to implement three versions of importance sampling. We present the results obtained by these algorithms when applied to a data set obtained by a robot navigating inside Wean Hall Building at Carnegie Mellon University. One of our algorithms obtains successful results, and it emphasizes the need of including laser information in order to correct odometry error. (Abstract shortened by UMI.)... |
