Action Selection of Ball Trapping in the Soccer Simulation using Decision Tree

Abstract

In a group game using a ball such as soccer, the passing behavior among players is important for achieving cooperative behavior. To acquire the passing behavior, conventional approaches mainly apply search and machine learning to the decision making of the players who perform the ball kicking. On the other hand, the position and posture of the pass receiver when receiving the ball have not been sufficiently studied. This paper proposes a decision tree based learning method to select a more advantageous ball trapping behavior. We use the RoboCup Soccer Simulator as an experimental environment to evaluate the performance of the action selection model, which is acquired from the game logs.