Brainstormers 2D — Team Description 2009
Thomas Gabel, Martin Riedmiller · Scientific Publication Server (Frankfurt University of Applied Sciences) · 2009
The main focus of the Brainstormers' effort in the RoboCup soccer simulation 2D domain is to develop and to apply machine learning techniques in complex domains. In particular, we are interested in ap- plying reinforcement learning methods, where the training signal is only given in terms of success or failure. Our final goal is a learning system, where we only plug in win the match - and our agents learn to gener- ate the appropriate behavior. Unfortunately, even from very optimistic complexity estimations it becomes obvious, that in the soccer simulation domain, both conventional solution methods and also advanced today's reinforcement learning techniques come to their limit - there are more than (108×50) 23 different states and more than (1000) 300 different poli- cies per agent per half time. This paper outlines the architecture of the Brainstormers team, focuses on the use of reinforcement learning to learn various elements of our agents' behavior, and highlights other advanced artificial intelligence methods we are employing.