Distributed Multi-Sensor Fusion for Improved Collaborative GPS-Denied Navigation

Abstract

This paper addresses the problem of determining high accuracy absolute and relative 3-D positions of mobile sensor network nodes by fusing inertial and radio frequency (RF) ranging measurements to support collaborative navigation in GPS-denied environments. We summarize four multi-sensor fusion algorithms based on optimization and Extended Kalman Filtering (EKF) techniques, focusing on a novel distributed iterative EKF formulation. We derive a general error scaling law, termed the “Teamwork Effect,” e(n, s)?e (1, s)/ n , where e (1, s) is the error of each node’s individual location estimate as a function of s (e.g. time or distance traveled) and e (n, s) is the improved location error a node can expect when collaborating as a member of a size n network, under idealized assumptions. Simulations of a variety of operational scenarios show that each approach can localize both the absolute and relative positions of a mobile network with high accuracy, even under non-ideal conditions involving greatly varying INU and RF sensor noise models. In addition to validating the predicted n error reduction of the “Teamwork Effect,” the simulations demonstrate additional collaborative effects of the multisensor fusion algorithms. An “Anchor Effect” enables the flexible use of minimal numbers of optional RF navigation reference beacons to greatly reduce long-term error growth. A “Reset Effect” provides an automatic reset of location estimate uncertainties from larger to lower values under certain common operational conditions, resulting in position estimates that improve, rather than degrade, over time. By comparing the trade off between performance and computational complexities, we conclude that our distributed iterative EKF algorithm is a good candidate for implementation in a real-time high-performance navigation system supported by a lowbandwidth tactical network.