The layered control architecture is designed for the need of the multirobot intelligent team formation.There are three levels:the cooperation task level, the coordination behavior level and the action planning level.The cooperation task level uses the potential grid method, which improves the safety of the path and reduces the calculation complexity. The coordination behavior level uses the reinforcement learning which can strengthen the robots' intelligence. The action planning level uses the fuzzy planning methods to realize the action matching.The communication model transfers the message in different level. This architecture shows not only the independence and the intelligence of the single robot but also the cooperation and the coordination among the robots.In each level, the task is distributed reasonably and clearly. Finally the feasibility of the architecture is verified further in the simulation of the experiment. The expansibility of the architecture is good and the architecture can be used in the similar system.