A layered approach for an autonomous robotic soccer system

Manuela Veloso Peter Stone Sorin Achim Computer Science Department Carnegie Mellon University Pittsburgh PA 15213-3891 fmmv,pstone,soring@cs.cmu.edu http://www.cs.cmu.edu/~fpstone,mmv,soring Introduction As robots become more adept at operating in the real world, it becomes more important to build teams of robots, capable of high-level collaborative and adversarial planning and learning in real-time situations. Robotic Soccer is an interesting emerging domain that is particularly appropriate for studying these issues (Kitano et al. 1995; Stone & Veloso 1996a; Kim 1996). We have been pursuing research in this domain using both a simulator and real physical agents. This paper is a very brief introduction to our work and the reader is referred to the references provided. We present the architecture of the physical system and introduce how actions are layered building upon each other to create strategic reasoning. We decompose the system's capabilities in di erent layers, namely behavioral, perceptual, and strategic. We view the strategic layer itself consisting of di erent levels. We have been using realistic simulation environments (Noda 1995; Sahota 1993) to learn basic collaborative strategic procedures. Our on-going research consists of extending and applying these robust strategic templates to the physical agents. A ground-breaking system for Robotic Soccer, and the one that served as the inspiration and basis for our work, is the Dynamo System developed at the University of British Columbia (Sahota 1993). This system introduces a decision making strategy called reactive deliberation which is used to choose from among seven hard-wired behaviors. In our approach, we structure the deliberation and reaction as a layered learning architecture. Other e orts have been pursued on applying learning to acquire speci c behaviors in di erent setups (e.g., (Asada et al. 1994)). We have chosen to focus on producing a simple, robust design that will enable us to concentrate our e orts on learning low-level behaviors and high-level strategies. Our current mini-robotic system is certainly usable for tasks other than Robotic Soccer, but since our main purpose in building the system was to work in the Robotic Soccer domain, we made most of our design decisions with this domain primarily in mind. Overall Architecture The architecture of our physical RoboSoccer system addresses the combination of high-level and low-level reasoning by viewing the overall system as the combination of the mini-robots, a vision camera over-looking the playing eld connected to a centralized interface computer, and several clients as the minds of the minirobot players. Figure 1 sketches the building blocks of the architecture.