A Novel Lie Group Framework-Based Student’s <i>t</i> Robust Filter and its Application to INS/DVL Tightly Integrated Navigation
Siyuan Du, Fengchi Zhu, Zhao Wang, Yulong Huang, Yonggang Zhang
Abstract
The inertial navigation system and doppler velocity log (INS/DVL) tightly integrated navigation system is widely used in autonomous underwater vehicle. However, due to the changeable and complex marine environment, the DVL measurement noise may contain a significant amount of outlier interferences, and the system noise also exhibits inherent uncertainty. Although the existing variational-Bayesian-based robust Student’s <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">t</i> Kalman filter (VB-RSTKF) can mitigate the impact of the outliers to a certain extent, the establishment of its state-space model is based on traditional error variable, which makes the estimation consistency poor. In addition, its estimation accuracy is affected by the inaccurate prior prediction error covariance matrix and state noise covariance matrix (SNCM). In this paper, in order to improve the estimation consistency, a novel unified Lie group navigation framework is proposed, and the invariant error-based state and measurement equation are established respectively. Then, the expectation maximum (EM)-VB-based RSTKF (EMVB-RSTKF) is proposed to improve the algorithm estimation accuracy within the inaccurate prior SNCM, where the prediction error scale matrix is estimated by using the EM method and the SNCM is indirectly estimated online based on the estimate of prediction error covariance matrix. Finally, the results of simulations and semi-physical filed experiments show that the proposed algorithm has significant advantages in convergence speed and estimation accuracy.