Simultaneous Position and Velocity Estimation of AUVs with Unknown External Inputs Based on Variational Bayesian

This paper proposes a variational Bayesian filter for simultaneous position and velocity estimations of autonomous underwater vehicles (AUVs) with unknown external inputs. To cope with complex underwater dynamics and unpredictable external input disturbances, we model the system dynamics using a random walk approach. Instead of relying on fixed a priori noise statistics as required by standard filters, the variational Bayesian (VB) framework is employed to jointly infer the external inputs and their unknown, time-varying noise statistics. This adaptive approach allows the proposed unknown external input variational Bayesian filter (UEVBF) to enhance robustness against unknown input disturbances while retaining optimal estimation performance as the extended Kalman filter (EKF) in the absence of input disturbances. Furthermore, it simplifies traditional modeling by decoupling control variables from state variables, thereby enhancing computational efficiency. The effectiveness of the algorithm is verified through underwater vehicle navigation simulations. The results show that the UEVBF has an estimation accuracy equivalent to that of EKF under normal conditions and is superior to that of EKF and particle filter (PF) in the presence of unknown external inputs. In addition, the proposed UEVBF exhibits significantly faster computational speed than PF, which makes it a practical solution for real-time AUV navigation.