A New Robust Filter for Nonlinear System Based on Multiple Scaled Multivariate Skew-t Distribution

Abstract

The accuracy of the state estimation for nonlinear system may be reduced if each dimension of the measurement noises has different heavy tail and skewness characteristic. In order to deal with the thorny problem, the multiple scaled multivariate skew-t distribution (MSMST), which is able to be formulated as a Gaussian, truncated Gaussian-Gamma hierarchical form, is first proposed to described such heavy-tailed and skew measurement noise (HTSMN). A new robust nonlinear filter is developed by utilizing variational Bayesian (VB) technique on the basis of hierarchical nonlinear Gaussian model (HNGM), which is constructed on the basis of the MSMST distribution. A typical example of maneuvering target tracking show that the superiorities of the proposed algorithm.