Distributed Matrix Weighted Fusion Gaussian Filter

Abstract

This letter is concerned with the distributed fusion estimation problem for nonlinear systems. Generally, optimal distributed weighted fusion criterion can be given by the form of analytic solution under the case of linear systems. However, this classical fusion criterion is not applicable to nonlinear systems because the error cross-covariances among local filters cannot be obtained. In this case, the statistical linear regression (SLR) method is firstly proposed to approximate the error cross-covariances among local Gaussian filters. To further reduce the performance loss caused by linearization errors in SLR, a measurement-dependent maximum likelihood criterion is developed to adjust the error cross-covariance matrices. Finally, a mobile robot localization example is employed to show the effectiveness and advantages of the proposed methods.