The Design and Implementation of a Multi-Agent Soccer Simulator as a Tool for AI Research and Education

Abstract

The Design and Implementation of a Multi-Agent Soccer Simulator as a Tool for AI Research and Education Copyright c © 2006 by Iwein J.J. Borm (1174398) Man-Machine Interaction Group Faculty of EEMCS Delft University of Technology Members of the Supervising Committee drs. dr. L.J.M. Rothkrantz (chair) dr. ir. C.A.P.G. van der Mast ir. H.J.A.M. Geers This thesis describes a multi-agent soccer simulator based on a highly simplified soccer model, and its application in an introductory AI course. The soccer model allows agents to start on a high level and removes many uncertainties that occur in physical soccer robots. Because of these simplifications, the system is suitable for researching high-level strategic behavior, such as cooperation and team work. The simulator features a three-layer model representing the physical environment, robot and behavior. This layered approach makes the system extensible. Reference teams of increasing complexity are available, and help to detect flaws in strategies, as well as pose a serious challenge for users to beat. Coursework was developed, that guides students through the process of creating a team capable of defeating the reference team, in order to teach students about basic AI concepts such as multi-agent systems, ad-hoc networks, rule-based reasoning and cooperative agents. The coursework has been used in a first-year undergraduate AI course at Delft University of Technology, and was found to be motivating and very educative.