A Variational Bayesian-Based Maximum Correntropy Adaptive Kalman Filter for SINS/USBL Integrated Navigation System

The strap-down inertial navigation system and ultra-short baseline (SINS/USBL) integrated system is a highly promising tool for navigating the unmanned underwater vehicles (UUVs). The measurement in the deep sea often contains unknown, time-varying noise and outliers. The traditional Kalman filter may face challenges in achieving high-precision underwater navigation due to its limited robustness and adaptivity. Although the robust Kalman filter has been developed and can effectively handle non-Gaussian noises in most cases, it may still suffer a significant loss in accuracy under non-stationary noise conditions. This study presents an adaptive robust Kalman filter that integrates the maximum correntropy criterion (MCC) with the variational Bayesian (VB) method to effectively mitigate the effects of complex noise. The proposed method achieves adaptivity by employing the VB method to estimate measurement noise covariance while enhancing robustness by mitigating outliers using the variable kernel bandwidth MCC strategy. According to simulation and offshore experiments, the proposed method provides superior estimation accuracy compared to related adaptive and robust algorithms.