Robust Adaptive Nonlinear KF Under Hierarchically Gaussian Outliers

Abstract

Standard state estimation techniques are designed under the assumption that the system is perfectly known, which does not typically hold in practice. Under model mismatch the filter performance is significantly degraded, reason why robust estimators are relevant. In this contribution we address the nonlinear filtering problem under outliers, for which a skewed Gaussian scale mixture distribution is considered to obtain a flexible description that allows for a conditionally Gaussian representation. A variational Bayesian approach is used to approximate the joint posterior distribution of the states and latent variables, designing a robust nonlinear filter, where the skewness parameters are estimated by online expectation-maximization. An illustrative navigation example is provided to show the new filter’s advantages and limitations.