Clustering technology is very important in ad hoc networks and sensor networks from the view point of reducing the traffic load and energy consumption. In this paper, we propose a new structure formation mechanism as a tool for clustering. It meets the key clustering requirements including the use of an autonomous decentralized algorithm and a consideration of the situation of individual nodes. The proposed mechanism follows the framework of autonomous decentralized control based on local interaction, in which the behavior of the whole system is indirectly controlled by appropriately designing the autonomous actions of the subsystems. As an application example, we demonstrate autonomous decentralized clustering for a two-dimensional lattice network model, and the characteristics and adaptability of the proposed method are shown. In particular, the clusters produced can reflect the environmental situation of each node given by the initial condition.