Distributed centroid estimation from noisy relative measurements

We propose an anchorless distributed technique for estimating the centroid of a network of agents from noisy relative measurements. The positions of the agents are then obtained relative to the estimated centroid. The usual approach to multi-agent localization assumes instead that one anchor agent exists in the network, and the other agents positions are estimated with respect to the anchor. We show that our centroid-based algorithm converges to the optimal solution, and that such a centroid-based representation produces results that are more accurate than anchor-based ones, irrespective of the selected anchor.