Map building, localization, and structure estimation problems in mobile robotics

This dissertation focuses on many relevant problems in mobile robot building a model (map) of the indoor environment. It is a very basic task in mobile robotics. First, two dimensional map building techniques are discussed. A 2D map is a floor map that describes the shape of a horizontal 2D space in the environment. The existing approaches are sensor specific. A comprehensive geometric feature based map has been developed using line segments, circular arcs, and point clusters. These features are unified models for the information from various 2D range sensors. Fast local and global localization techniques have been developed by taking the advantage of the strong geometric information of the proposed map.; Building three dimensional map is still a very open problem in mobile robotics. A new three dimensional map building approach has been proposed. 3D structures in the neighborhood of the geometric features in the 2D map are estimated and attached to the 2D map as a 3D description of the part of the environment that the mobile robot interacts. Monocular vision and 2D range are used together to estimate the structure of interest from multiple viewpoints. In the structure estimation, 2D range information is utilized to help building the structure models and provide the structure hypotheses. The structure estimation accuracy has also been improved by fusing monocular vision and 2D range in an Extended Kalman Filter scheme.; The proposed techniques have been implemented on a Nomad XR4000 mobile robot in various experiments and the results validate their effectiveness.