A multi-robot cooperation system should be a distributed system which is robust and fault tolerant. It should also be scalable as guaranteed by a scalable communication protocol and a cooperative control architecture. Moreover, robots should be implemented with a flexible navigation controller which can cope with a dynamically varying environment. Above all, it should be able to assign their missions flexibly, adaptively, and coherently. In this dissertation, a multi-robot cooperation system is proposed that efficiently addresses the issues of a scalability. To solve the above mentioned problems, three solutions are developed. First, a limit-cycle navigation method is proposed to improve the ability of obstacle avoidance. This method can overcome disadvantages of an existing potential field method and can easily cope with multi-obstacle situations. Moreover, it can guarantee that the desired posture at a designated position of the robot can be achieved. Second, for a distributed robot system, a novel communication protocol, CSMA/R is proposed. It can guarantee the performance in spite of addition or removal of robots. Third, a scalability-oriented control architecture, SoCA is proposed, which can guarantee the total performance of a system against any change in system size and a dynamically varying environment. Under the conditions that communication can be broken at any time, a communication channel is not always available and a robot does not have perfect ability, the proposed multi-robot cooperation system can guarantee a scalability without the need for centralized control by giving up its mission or conceding to other robots in the system.