In this thesis, a multi-agent system for robot soccer is designed to be controlled by Supervisor controller which is consist of Supervisor, Attack controller, and Defense controller. It is designed with Petri Net, and analyzed by a Petri Net's theory. After that, real soccer-playing robots are made. Robot soccer game has very dynamic and uncertain characteristics because the ball and robots move fast and constantly. Furthermore, there exist competitions between agents of one team and hostile agents of the opposite team. It is a basic problem to determine what role the agent should take in the given situation. Particularly, the soccer-playing robot should take an appropriate action according to its role. While playing real robot-soccer game, there are insufficient things at determining roles for robots. So, Q-learning is used for Supervisor which determine roles.