Vision-based Monte Carlo Self-localization for a Mobile Service Robot Acting as Shopping Assistant in a Home Store * H.-M. Gross, A. Koenig, H.-J. Boehme, and Ch. Schroeter Department of Neuroinformatics, Ilmenau Technical University, 98684 Ilmenau, Germany Horst-Michael.Gross@tu-ilmenau.de Abstract We present a novel omnivision-based robot localiza- tion approach which utilizes the Monte Carlo Lo- calization (MCL) [2], a Bayesian filtering technique based on a density representation by means of par- ticles. The capability of this method to approximate arbitrary likelihood densities is a crucial property for dealing with highly ambiguous localization hypotheses as are typical for real-world environments. We show how omidirectional imaging can be combined with the MCL-algorithm to globally localize and track a mobile robot given a taught graph-based representation of the operation area. In contrast to other approaches, the nodes of our graph are labeled with both visual fea- ture vectors extracted from the omnidirectional im- age, and odometric data about the pose of the robot at the moment of the node insertion (position and heading direction). To demonstrate the reliability of our approach, we present first experimental results in the context of a challenging robotics application, the self-localization of a mobile service robot acting as shopping assistant in a very regularly structured, maze-like and crowded environment, a home store. 1 Introduction and motivation An interactive mobile service robot, e.g., a shopping assistant, should be able to actively observe its oper- ation area, to detect, localize, and contact potential users, to interact with them continuously, and to ad- equately offer its specific services. Typical service tasks we want to solve in our PERSES (PERsonal SErvice System) project are to guide the user to de- sired areas or articles within a home store (guidance function) or to follow him as a mobile information kiosk while continuously observing the user and his behavior (companion function) (see [3]). To accom- modate the challenges that arise from the specifics of our interaction-oriented scenario and the character- istics of the operation area, a very regularly struc- * Supported by a Thuringian Ministry of Science, Research, and Art Grant (PERSES & SERROKON-Projects) 100 m 45 m Figure 1: (Top) Location plan of our experimen- tal area, a large home store in Erfurt (toom Bau- Markt). The topology of the store is characterized by many similar, long hallways of equal width. Because of their very regular structure, most of the hallways can be distinguished only visually. (Bottom) exem- plary appearance of three hallways which can not be distinguished by distance sensors (sonar, laser) be- cause of identical geometrical features. The hallways and racks, however, show very characteristic views, which allow a vision-based self-localization. tured, maze-like and crowded environment, we place special emphasis on vision-based methods for both human-robot interaction and robot navigation. The motivation for this is outlined in the following: Functional and economical advantages: Mean- while, vision systems have become available as very powerful universal sensor systems with a good price- performance ratio such that they can be successfully utilized in a great number of robotics tasks - both in human-robot interaction and autonomous naviga- tion. Therefore, our low-cost prototype of a mobile and interactive shopping assistant currently under development will be equipped with an universally us- able omnidirectional vision-system instead of an ex- pensive laser rangefinder, which shows a number of limitations in human-robot interaction and naviga- Proceedings of the 2002 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems EPFL, Lausanne, Switzerland • October 2002 0-7803-7398-7/02/$17.00 ©2002 IEEE 256