A further advantage of the filtering techniques is that Markov localization does not require a detailed map of the environment. Instead, it suffices to provide only an outline which merely includes the aspects of the world which are static.
Figure 15(a) shows a ground plan of our department building, which contains only the walls of the university building. The complete map, including all movable objects such as tables and chairs, is shown in Figure 19. The two Figures 15(b) and 15(c) illustrate how the distance filter typically behaves when tracking the robot's position in such a sparse map of the environment. Filtered readings are shown in grey, and the incorporated sensor readings are shown in black. Obviously, the filter focuses on the known aspects of the map and ignores all objects (such as desks, chairs, doors and tables) which are not contained in the outline. [Fox1998] describes more systematic experiments supporting our belief that Markov localization in combination with the distance filter is able to accurately localize mobile robots even when relying only on an outline of the environment.
