Localization and Mapping for Service Robots: Bearing-Only SLAM with an Omnicam

Abstract

Localization and mapping are fundamental problems in service robotics since representations of the environment and knowledge about the own pose significantly simplify the implementation of a series of high level applications. For instance, nearly all relevant applications of service robots require navigation skills that allow for purposeful motions. Typical examples are fetch-and-carry tasks or floor coverage tasks. These are best implemented based on pose knowledge and on continuously updated maps of the environment. Thus, a key component towards widespread use of service robots is a localization capability that can vary from pose tracking over relocalization to even solving the most demanding socalled kidnapped robot problem. In the latter case the robot is carried to an arbitrary location during its operation and is expected to detect this and then relocalize itself. Of course, the difficulty of the localization problem significantly depends on the available information. Normally, localization requires some kind of map as reference and map building requires pose knowledge to consistently insert artifacts. A SLAM (simultaneous localization and mapping) problem arises when the robot does neither have access to a map of the environment nor does it know its own pose. The SLAM problem is more difficult than the mapping with known poses and it is more difficult than the localization problem based on a given map. A successful approach to overcome the chicken-and-egg problem of concurrently building a map and maintaining the robot pose is based on a probabilistic representation. The online SLAM problem maintains the robot pose and the map in a single state vector. The remaining challenge is to estimate the posterior over the current pose along with the map given all the measurements and controls. SLAM is of particular importance for service robotic applications since it significantly reduces deployment efforts and ensures continues updates as needed in dynamic environments. However, one cannot neglect the specific demands on service robots. For instance, in most applications of service robots the consumer neither accepts modifications of the environment (like artifical landmarks) nor complex and time consuming deployment efforts. Although a large body of work already proved that the SLAM problem is solvable even without deploying artifical landmarks, most approaches are based on range measuring devices. For most of the service robotics applications like floor