A novel switching Gaussian-heavy-tailed distribution based robust fixed-interval smoother

• Published in: Signal processing 2022-06, Vol.195, p.108492, Article 108492
• Main Authors: Fu, Hongpo, Cheng, Yongmei
• Format: Article
• Language: English
• Subjects: Dynamic outliers; Fixed-interval smoother; Nonlinear system; Variational Bayesian inference
• ISSN: 0165-1684; 1872-7557
• DOI: 10.1016/j.sigpro.2022.108492

•Present a novel robust fixed-interval smoother for the nonlinear systems with non-stationary heavy-tailed distributed noises,.•Model the non-stationary heavy-tailed noises by a new switching Gaussian-Heavy-Tailed distribution.•Present robustness when the state and measurement noises are inaccurate Gaussian distributed and/or heavy-tailed distributed.•Demonstrate the superior performance of the proposed smoother as compared with cutting-edge smoother by the experiments with the synthetic data and real data. In nonlinear systems, the stochastic process and measurement noises may be non-stationary heavy-tailed distribution due to the dynamic outliers induced by unreliable sensors and complicated environments. The main purpose of this paper is to address the problem by establishing a new switching Gaussian-heavy-tailed (SGHT) distribution. We model the noise with the help of switching between the Gaussian and the newly designed heavy-tailed distribution. Then, utilizing two auxiliary parameters satisfying categorical and Bernoulli distributions respectively, we construct the SGHT distribution as a hierarchical Gaussian presentation. Furthermore, applying variational Bayesian inference, a novel SGHT distribution based robust fixed-interval smoother is derived. The experiment results of the synthetic data and real vehicle localization dataset demonstrate the superior performance of the proposed smoother as compared with cutting-edge smoother.