Gemini in RoboCup-2000
We implemented “Gemini” a client program for the SoccerServer. The objective of this program is testing a lot of learning methods on multi-agent environments. In the current implementation,Gemini can select the most effective strategy for an enemy, using reinforcement learning. Furthermore, we are trying to implement a meta-level learning, which turn each learning function on or off according to whether the learning succeed or not.
- 1.Kaelbling L. P., Littman M. L. and Moore A. W. “Reinforcement Learning: A Survey” Journal of Artificial Intelligence Research 4,pages 237–285 1996.Google Scholar
- 2.Kimura H., Yamamura M. and Kobayashi S. “Reinforcement Learning in Partially Observable Markov Decision Processes: A Stochastic Gradient Method” Journal of Japanese Society for Artificial Intelligence, Vol.11, No.5 pages 761–768 1996Google Scholar
- 4.M. Ohta and T. Ando “Cooperative Reward in Reinforcement Learning” Proc. of 3rd JSAI RoboMech Symposia pages 7–11 April 1998.Google Scholar