Vision-based fast and reactive monte-carlo localization.pdf

Vision-Based Fast and Reactive Monte-Carlo Localization Thomas Ro?fer ? , Matthias Ju?ngel ? ? Bremer Institut fu?r Sichere Systeme, Technologie-Zentrum Informatik, FB 3, Universita?t Bremen. E-Mail: roefer@tzi.de ? Institut fu?r Informatik, LFG Ku?nstliche Intelligenz, Humboldt-Universita?t zu Berlin. E-Mail: juengel@informatik.hu-berlin.de. Abstract— This paper presents a fast approach for vision-based self-localization in RoboCup. The vision system extracts the features required for localization without processing the whole image and is a first step towards independence of lighting conditions. In the field of self-localization, some new ideas are added to the well-known Monte- Carlo localization approach that increase both stability and reactivity, while keeping the processing time low. I. INTRODUCTION THE Sony Four-Legged Robot League is one of the of-ficial leagues in RoboCup. In a match, two teams of four robots each compete against each other on a field of approximately 5× 3 m2 in size. All teams consist of robots of the same type, i. e. the Sony Aibo ERS 210 (cf. Fig. 1b). The Aibo is a four-legged robot with 20 degrees of free- dom, a color camera, and more than 30 further sensors. The specialty of the league is that, on the one hand, the movements of the robots are the most complex in RoboCup so far, and on the other hand, an on-board camera is the most central sensor. As the robots are only equipped with a 200 MHz processor, all algorithms used, e. g. for image- processing or self-localization, have to be highly efficient to run in real-time. II. GRID-BASED VISION As the main sensor of the robot is a camera, all objects on the RoboCup field are color coded. There are two-colored flags for localization (pink and either yellow, green, or sky- blue), the two goals are of different color (yellow and sky- blue), the ball is orange (as in all RoboCup leagues), and the robots of the two teams wear tricots in different colors (red and blue). A very common preproc