Master Thesis - Past Projects - Abstract

Användarvänligt skapande av strategier för intelligenta agenter

ID: LiTH-IDA-Ex-03/072-SE

This thesis deals with the issue of taking advantage of the knowledge of domain experts when creating intelligent autonomous agents. Utilizing human expertise in the agents' domain is difficult because the experts do not have the necessary skills to create agents by themselves. In this report an experimental platform for developing teams of agents is proposed and illustrated with RoboCup. This is a soccer competition where teams of intelligent autonomous agents meet. The development environment is designed to support cooperation between experts from different fields; programmers, agents experts and experts in soccer strategy. The end result is a strategy editor which together with a preexisting agent creation environment becomes a complete development environment for creating agents, from low level skills to team level strategies. The strategy editor attempts to emulate the traditional way to instruct soccer players, by working as a whiteboard where the strategies can be expressed with simple symbols. The strategy is compiled to individual behavior based agents without an internal representation of the strategy. The development environment supports continuous improvement and change of the expressed strategies. The idea is that it is essential for the development environment to change with feedback from the domain expert to fully utilize his or her knowledge. The strategy editor showed its potential by allowing people without knowledge of the underlying agent architecture to create strategies for the Robocup team of agents. The strategies and particularly the speed with which they could be created proved to be the main contributing factor for bringing a RoboCup team to the semi finals at the simulation league competitions of RoboCup´99.

Keywords: autonomous agents, Robocup, strategy editor

Author(s): Johan Ydren

Contact: Simin Nadjm-Tehrani