A Variational Bayesian Maximum Correntropy Extended Kalman Filter for Robust GPS/INS Vehicle Localization

Abstract

This paper presents a novel variational Bayesian maximum correntropy extended Kalman Filter for robust GPS/inertial navigation system integration in outdoor vehicle localization applications. The proposed method addresses two critical challenges in real-world navigation systems: time-varying sensor noise characteristics as well as process (IMU) and measurement (GPS) outliers. By combining the adaptivity of variational bayesian approximation for noise covariance estimation with the robustness of maximum correntropy criterion for outlier rejection, the proposed approach provides superior performance compared to conventional methods. The error-state formulation efficiently handles the nonlinear nature of attitude representation while maintaining computational tractability. Extensive experimental validation using real-world datasets demonstrates significant improvements in estimation accuracy. Results show our method achieves sub-meter positioning accuracy in all three axes (0.354–0.445 m in unknown measurement noise covariance conditions, 0.322–0.506 m during GPS outliers) while effectively handling both gradual sensor degradation and sudden measurement anomalies. The proposed filter demonstrates particularly impressive robustness against GPS outliers, maintaining stable estimation when conventional filters exhibit significant performance degradation, making it highly suitable for vehicle navigation in challenging environments.