Composition of Vector Fields for Multi-Robot Manipulation via Caging

Abstract

This paper describes a novel approach for multirobot caging and manipulation, which relies on the team of robots forming patterns that trap the object to be manipulated and dragging or pushing the object to the goal configuration. The controllers are obtained by sequential composition of vector fields or behaviors and enable decentralized computation based only on local information. Further, the control software for each robot is identical and relies on very simple behaviors. We present our experimental multi-robot system as well as simulation and experimental results that demonstrate the robustness of this approach.