An Adaptive and Robust UKF Approach Based on Gaussian Process Regression-Aided Variational Bayesian
Xu Lyu, Baiqing Hu, Kailong Li, Lubin Chang
Abstract
In this study, both adaptive and robust one is proposed, considering that the properties of a classic unscented Kalman filter (UKF) can be degraded severely by the outliers measured in the contamination distribution and the effects of time-varying noise. An adaptive and robust UKF approach on Gaussian process regression- assisted Variational Bayesian is proposed. The Variable Bayesian (VB) method is used to statistically approximate the time-varying noise due to the strong nonlinearity and inaccuracy of the system model. At the same time, Gaussian Process Regression (GPR) has the advantage of machine learning. The observation information trained by the GPR model performs real-time prediction of the sliding window. Based on the above two points, the output source of the measurement information is estimated to realize the robustness of the filter measurement update. When the noise is uncertain and there are outliers in the measurement information, simulation and SINS/GPS integrated navigation of actual tests were performed, respectively. The test outcomes indicate that the algorithm has better nonlinear estimation performance than the traditional method, the effectiveness of the filtering algorithm proposed is verified.