The objective of this thesis is to develop a robot soccer strategy system that can cooperate, coordinate, and learn. In order to achieve cooperation, we use a hierarchical architecture is employed. A primary robot is chosen and a fuzzy evaluator is used to evaluate the state of the field. Then a strategy is determined for command other robots to execute the strategy. In order to achieve the coordination, a multi-obstacle avoidance algorithm is developed to avoid action conflict among robots. A machine learning method is developed to improve the strategy system itself through the learning process. The complete algorithm had been verified in the 2002 FIRA World Cup of soccer robot competition and showed excellent results. In fact, our team, Formosa , won the first place of 11 vs. 11 robot competition.