Discovering behavior patterns in muti-agent teams

Abstract

Three of the most important open research areas in multi-agent cooperative systems are the construction of models related with the communication between agents, the multiple interactions and the behaviors adopted by agents during a task [5] . This work deals with the discovery of behaviors in multi-agent teams, more precisely in soccer teams. It is an extension of a previous work presented in [1] . The extension is focused on the discovery of tactical plays adopted by soccer-agents during a match within the context of formations. Due to the nature of team work in soccer-agent domains, the discovery of tactical behaviors should be done within the context of team formations. Nevertheless, the dynamic nature and the multiple interactions between players at each instant of the game difficult the tracking of formations, which at the same time difficult the discovery of tactical plays. In this work is proposed an original and efficient way of discovering tactical plays supported by a robust tracking of formations, even though they are submitted to dynamic changes of the world, based on the construction of topological structures. Successful results, derived from the analysis of an important number of matches of the RoboCup Simulation league matches, valid the efficiency of the model presented in this work. © 2008 Springer-Verlag Berlin Heidelberg.