Robot soccer is frequently used to validate coordination strategies in multi-robot systems. This is because coordination among the robots allows a robot soccer team to perform better through coordinated behaviors like passing the ball between teammates, intercepting the ball when opponent players pass it between them, or taking possession of the ball. This requires the coordination strategies focused on robot soccer teams to be designed in line with the conditions of the game like the position of the ball, of teammates or of opponent players. This paper presents the architecture for robot soccer team coordination, involving the dynamic assignment of roles among the players. This strategy is divided into tactics, which are selected by a hierarchical state machine. Once a tactic has been selected, it is assigned roles to players, depending on the game conditions. Each role performs defined behaviors selected by the hierarchical state machine. To carry out the behaviors, robots are controlled by the lowest level of the hierarchical state machine. The architecture proposed is designed for robot soccer teams with a central decision-making body, with a global perception. The strategy presented is validated by simulation.