Martin Riedmiller
University of Karlsruhe, ILKD, Germany
Sebastian Buck
University of Karlsruhe, ILKD, Germany
Sergio Dilger
University of Karlsruhe, ILKD, Germany
Ralf Ehrmann
University of Karlsruhe, ILKD, Germany
Artur Merke
University of Karlsruhe, ILKD, Germany
Download article
Published in: RobocCup-99 Team Descriptions. Simulation League
Linköping Electronic Conference Proceedings 4:13, p. 59-63
Linköping Electronic Articles in Computer and Information Science vol. 4 4:13, p. 59-63
Published: 1999-12-15
ISBN:
ISSN: 1650-3686 (print), 1650-3740 (online)
The following paper describes the design principles of the Karlruhe Brainstormers team for the RoboCup Simulator League. The basic motivation behind our approach is to broadly apply Machine Learning techniques. In particular; our longterm goal is to apply Reinforcement Learning techniques to autonomously learn team playing capabilities. This longterm goal determined the structure of the decision module; which has to choose between several available high-level moves based on evaluation functions. We plan to reach the final autonomously learning agent in several stages. The current version uses a hybrid decision module with both rule-based and learning components.
[1] A. G. Barto; S. J. Bradtke; and S. P. Singh. Learning to act using realtime dynamic programming. Arti cial Intelligence; (72):81-138; 1995.
[2] M. Riedmiller. Autonomously learning neural controllers. VDI-Verlag; 1996. Dissertation University of Karlsruhe (in german).
[3] Peter Stone and Manuela Veloso. Team-partitioned; opaque-transition reinforcement learning. In M. Asada and H. Kitano; editors; RoboCup- 98: Robot Soccer World Cup II. Springer Buch Verlag; 1998.
