Agent learning in simulated soccer by fuzzy Q-learning

Multiagent systems have emerged as an active subfield of artificial intelligence in the past few years. Soccer simulation provides a rich and challenging multiagent real-time domain. This paper employs fuzzy Q-learning to learn offending behaviors in the neighborhood of the goal in simulated soccer. Attacking players can choose one action out of actions such as shoot, pass, and dribble according to distances and angles to the goal and one of opponent defending players. The learning results are compared with those by Q-learning. Through computer simulations, we show that fuzzy Q-learning is effective in learning good offensive behaviors in simulated soccer.