Localization in Indoor Environments Using a Panoramic Laser Range Finder

This work presents a self-localization system suited for robots roaming in everyday indoor environments such as offices or residential buildings. Thus, it has to cope with environments that are typically not a priori modelled or in any way specially prepared, but are places where people move around and objects change position from time to time. The proposed localization system utilizes a panoramic laser range finder with a viewing angle of nearly 360◦ that provides planar range scans of the immediate surroundings in a fixed height parallel to the floor. The sensory input is preprocessed by extracting line segment features, which indicate the walls or other vertical boundaries in the environment. The following step is a scan matching procedure: Two preprocessed scans, which are in general taken from different positions in the surroundings, are shifted and rotated against each other in such a way that they optimally coincide. This is accomplished by employing a modified version of the so-called DP-Algorithm. This algorithm is wellknown in the pattern recognition domain and provides the optimal assignment of line segments solely on basis of the two scans without relying on any additional source of information or a priori knowledge of the environment. Based on this assignment of line segments, localization is realized by setting up and solving an overdetermined equation system. A further result from this is an estimate of the respective position uncertainty, which is to be propagated through the successive localization steps. The position estimates together with the sensory observations can be combined to build a map of the environment. In this context the robust recognition of known places as performed by the DP-Algorithm based scan matching is utilized to reduce accumulated uncertainty. The constructed environmental map is an attributed graph, which allows for different degrees of refinement and the use of topological relations side by side with geometrical positions. The approach has been implemented on a mobile robot and tested in a real world office environment under varying conditions.