Seminar: A Hybrid Genetic Programming Based Decision Making System for Multi-Agent Systems of RoboC

Amir Tavafi
M.Sc. Candidate
Supervisor: Dr. Wolfgang Banzhaf

A Hybrid Genetic Programming Based Decision Making System for Multi-Agent Systems of RoboCup Soccer Simulation


Abstract

In this thesis, a hybrid genetic programming approach is proposed for decision making system in the complex multi-agent domain of RoboCup Soccer Simulation. In the past, genetic programming was rarely used to evolve agents in this domain due to the difficulties and restrictions of the soccer simulation domain. The proposed approach consists of two phases each of which tries to cover the other's restrictions and limitations. The first phase will produce some evolved individuals based on a GP algorithm with an off-game evaluation system and the second phase uses best individuals of the first phase as input to run another GP algorithm to evolve players in the real game environment where evaluations are done during real-time runs of the simulator. It is observed that the evolved individuals after the second phase are able to outperform the same team with a decision making system which is not evolved.

 

Contact

Department of Computer Science

230 Elizabeth Ave

St. John's, NL A1B 3X9 CANADA

Tel: (709) 864-2530

Fax: (709) 864-2552

becomestudent@mun.ca