Decentralized Mapping of Robot-Aided Sensor Networks

Abstract

A key problem in deploying sensor networks in real-world applications is that of mapping, i.e. determining the location of each sensor such that subsequent tasks such as tracking, control and planning can be performed. In this paper, we present a robust decentralized algorithm for mapping the nodes in a sparsely connected sensor network using rangeonly measurements and motion from a mobile robot. Our approach utilizes an Extended Kalman Filter (EKF) in polar space, which lets us model the nonlinearities within the rangeonly measurements using Gaussian distributions. We extend this unimodal centralized EKF to a multi-modal decentralized framework. Each node within the network estimates its position along with its neighbor? position and uses a message-passing algorithm to propagate its belief to its neighbors. Thus, the global network localization problem is solved in pieces, by each node independently estimating its local network. We demonstrate the effectiveness of our approach using simulated and real-world experiments with little to no prior information about the node locations.