Numerically Robust Maximum Correntropy Criterion Kalman Filtering Using the UD Covariance Factorization Approach

The paper explores the maximum correntropy criterion Kalman filtering (MCC-KF) approach to estimating an unknown dynamic state of linear state-space models in the presence of non-Gaussian uncertainties and/or outliers. A widespread application of the MCC-KF estimator for solving practical problems has highlighted a special attention to the numerical stability issue. In this paper, we give a solution to an open problem formulated in our recent work for the class of UDU
^T
factorization-based MCC-KF methods. More precisely, we reveal a new stable MCC-KF implementation that outperforms the existing Cholesky factorization-based algorithms for estimation quality. The results of numerical experiments are also provided.