Fully distributed nonlinear state estimation using sensor networks

Abstract

This paper studies the problem of fully distributed state estimation using networked local sensors. Specifically, our previously proposed algorithm, namely, the Distributed Hybrid Information Fusion algorithm is extended to the scenario with nonlinearities involved in both the process model and the local sensing models. The unscented transformation approach is adopted for such an extension so that no computation of Jacobian matrix is needed. Moreover, the extended algorithm requires only one communication iteration between every two consecutive time instants. It is also analytically shown that for the case with linear sensing models, the local estimate errors are bounded in the mean square sense. A simulation example is used to illustrate the effectiveness of the extended algorithm.