Separating Moving Objects from Landmarks

Navigation and localization are still one of the most fundamental tasks to be accomplished by mobile autonomous robots. One of the main purposes of the navigation and localization process is to build a precise, usually allocentric spatial static representation (e.g. [S. Thrun and Schulz, 2000] ). Although robots are able to carry more and more powerful sensors, the question is, which informations are needed for localization and navigation. One way to do these tasks with only a minimal amount of resources is via landmarks. Furthermore it is an easy and failsafe way to do so. Localization can be done with only a single 180◦ degree camera, and a navigation by the change of the landmark ordering is very robust against misinterpretations and errors. This technique uses the fact that, seen from the agent, landmarks are switching locations only in a certain way( [Wagner, Visser, and Herzog, 2004] ). With an additional timer the robustness of this technique can be further increased. But with timing and the use of angles between the landmarks, it is also possible to measure the distances between the landmarks and the agent. Furthermore this technique can be extended to detect moving objects and to compute the speed and direction of them. Localization and navigation in times of GPS-receivers is no challenging task anymore. But in order to localize the agent must be outdoor and localization is done globally. To navigate relative to some local objects all these objects need to be properly located on a map. Furthermore it give only its own position and does not help localize other, potentially moving objects. For this several authors such as Lewitt and Lawton, Schlieder and Wagner ([Levitt and Lawton, 1990], [Schlieder, 1993], and [Wagner, Visser, and Herzog, 2004]) have considered navigation via landmarks. These approaches do not rely on a precise allocentric map. In contrast to Lewitt and Lawton, the approach of Wagner overcomes the problem of a incorrect localization via the socalled round-view. In addition it needs less information than the approach of Schlieder. Here we first describe a way how an agent can localize itself with landmarks. These landmarks can be, for example Copyright c © 2010, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. P Q