Our metric Markov localization technique, including both sensor filters, has been implemented and evaluated extensively in various environments. In this section we present some of the experiments carried out with the mobile robots Rhino and Minerva (see Figure 1). Rhino has a ring of 24 ultrasound sensors each with an opening angle of 15 degrees. Both, Rhino and Minerva are equipped with two laser range-finders covering a 360 degrees field of view.
The first set of experiments demonstrates the robustness of Markov localization in two real-world scenarios. In particular, it systematically evaluates the effect of the filtering techniques on the localization performance in highly dynamic environments. An additional experiment illustrates a further advantage of the filtering technique, which enables a mobile robot to reliably estimate its position even if only an outline of an office environment is given as a map.
In further experiments described in this section, we will illustrate the ability of our Markov localization technique to globally localize a mobile robot in approximate world models such as occupancy grid maps, even when using inaccurate sensors such as ultrasound sensors. Finally, we present experiments analyzing the accuracy and efficiency of grid-based Markov localization with respect to the size of the grid cells.
The experiments reported here demonstrate that Markov localization is able to globally estimate the position of a mobile robot, and to reliably keep track of it even if only an approximate model of a possibly dynamic environment is given, if the robot has a weak odometry, and if noisy sensors such as ultrasound sensors are used.