Cooperative Enclosing and Grasping of an Object by Decentralized Mobile Robots Using Local Observation

This paper discusses the design of a decentralized capturing behavior by multiple mobile robots. The design is based on a gradient descent method using local information. The task of capturing a target is divided into two subtasks; the enclosing subtask and the grasping subtask. An analysis of the convergence of the local control policy in the enclosing subtask is provided, while ensuring that the neighborhood relation of the robot system is preserved. In the grasping subtask, the force-closure condition in decentralized form is used to design a local objective function. A local estimation of the shape of the object is introduced so that each robot can decide how to move on the basis of only the available local information. The proposed local control policies were evaluated using simulations and the flexibility of the system was verified owing to the decentralized nature of the system. The enclosing subtask was implemented using multiple mobile robots with local observation from omni-directional CCD cameras.