Distributed particle filters via barycenters in 2-Wasserstein space

Abstract

We develop a new version of distributed particle filters that exploits the novel theory of 2Wasserstein barycenters. We consider a wireless sensor network deployed over a vast geographical region where each sensor makes local observations and transmits a target state estimate to the Fusion center (FC). Subsequently, the FC produces a global target state estimate based on the transmitted data. We propose the One-step Particle Filters (OPFs) and the Iterative Particle Filters (IPFs) to accommodate scenarios where the sensors communicate at the end stage or at each stage, respectively. Moreover, we present a comprehensive study of the convergence results of the One-step Particle Filters. Finally, we validate our algorithms using synthetic experiments and demonstrate the effectiveness of our proposed approaches.