Distributed, Sparse and Asynchronous C-Means for Robust Coverage with Networked Robots

In this paper we present a novel coverage framework for a network of mobile robots that aim at covering a finite and discrete set of points of interest. The proposed framework is non-exclusive, in that each point of interest can be covered by multiple robots with different intensities, thus resulting in a partly overlapping segmentation of the points of interest in groups covered each by a specific robot. This property improves robustness of our solution for the fact that each point of interest may remain covered by secondary robots, in case the main one becomes unable to fully perform its intended functions. Moreover, the intensity of the associations represents valuable meta-information that can be the basis for implementing complex interaction tasks among robots and points of interests, e.g., in order to define priorities or to elect a leader in case of emergency at some specific point of interest. The proposed algorithm is distributed, asynchronous, and near-optimal, and it is based on the C-means, originally proposed for non-exclusive clustering and for cluster centroid identification of a set of observations. We also cope with limited sensing range of robots, which cannot be addressed by traditional C-means, thus leading to an unfeasible coverage solution. In fact, the proposed sparse C-means can enforce robots to take into account only neighboring points of interest.