The Bremen Autonomous Wheelchair "Rolland" is a smart wheelchair robot intended to support handicapped and elderly people. Rolland and other service robots differ from classical industrial robots in many ways. The two most fundamental characteristics are: First, Rolland is mobile and has to navigate in environments often optimized for humans but not for robots. A prerequisite for mobility and navigation is being able to localize oneself in the environment. And second, Rolland is simultaneously controlled by the user and by the technical system with interdependently changing priorities: it is a so-called "shared-control system". As a consequence, two primary goals are to be pursued for Rolland: it needs an "adequate" self-localization approach. And, the human-robot interaction has to be realized in a way such that conflict situations are avoided. Having emerged in the 1980s, the self-localization of mobile robots is still a hot research topic. It can be considered to be solved for small- and middlescale environments provided that adequate sensor equipment (e. g. laser range finders) is available. But there are still a lot of open problems in mobile robot self-localization. Some of them are especially relevant for the service robotics domain: low-cost solutions (a dollar 5,000 laser range finder almost doubles the price of a wheelchair) are to be found that work in large-scale environments (e. g. hospital or university buildings spread over a campus area). This thesis presents RouteLoc, a new self-localization approach that needs only minimal input (i. e. no expensive proximity sensors are required) to absolutely localize a robot even in large-scale environments. The algorithm is tested in real world experiments with Rolland on the campus of the Universitat Bremen. Among others, the so-called "kidnapped robot problem" is solved. The second topic, the shared-control aspect of manned service robots, is no less challenging. So-called "mode confusion" situations have to be avoided. In the aviation psychology community, the problem of mode confusion has already been discussed for about a good decade. However, the notion as such has never been rigorously defined. In addition, the pertinent publications so far cover almost exclusively the pilot-autopilot interaction. This thesis presents a new, rigorous view of mode confusion. A framework based on existing formal methods is established for separately modeling the technical system, the user's mental representation of it, and their safety-relevant abstractions. As a result, an automated model checking approach can be applied to detect mode confusion potential already in the design phase. In a case study, the obstacle avoidance skill of Rolland is checked for mode confusion potential with tool support. This thesis is rounded off with a review on the state of the art in smart wheelchair robotic research and with a brief history of Rolland.