One of the most amazing phenomena widely observed in nature is self-assembly; living systems spontaneously form their body structure through the developmental process. While this remarkable phenomenon are not thoroughly understood in biology, the concept of self-assembly becomes undeniably indispensable also in artificial systems as they increase in size and complexity. Based on this consideration, this paper discusses the realization of self-assembly with the use of a multi-robotic system each of which has simple motile function. The main contributions of this paper are twofold: the first concerns a fully decentralized control method derived from the mutual entrainment between coupled nonlinear oscillators; and the second is related to the exploitation of physical interaction between agents stemming from a passive deformation mechanism, which allows an efficient movement of individual agents during the course of self-assembly. Here, form generation by self-assembly is considered as the result of time evolution toward the most dynamically stable state. Owing to this, in principle, the proposed method also satisfies significant ability of self-repair without making any modification to the proposed algorithm. In this paper we validate proposed method by exploiting real physical robotic agents. Experimental results show that stable and spontaneous self-assembly is achieved irrespective of the initial positional relationship between the agents.