Model-consistent iterated least-squares initialization for UKF-based AUV tracking under an isogradient sound speed profile

Autonomous underwater vehicle (AUV) tracking is an important enabling technique for the Internet of Underwater Things (IoUT). However, depth-dependent acoustic propagation and inaccurate filter initialization can significantly degrade the transient performance of nonlinear AUV tracking. To address these issues, this paper develops a model-consistent iterated least-squares initialized unscented Kalman filter (ILS-UKF) method under an isogradient sound speed profile (SSP). In the initial localization stage, a curved-ray ILS estimator is used to obtain the initial AUV position and construct its measurement-informed local covariance from the weighted time-difference-of-arrival (TDoA) information. Specifically, the pairwise TDoA construction eliminates the common AUV transmission time term and therefore does not require synchronization between the AUV clock and the sensor-network clock. In the continuous tracking stage, the resulting position-covariance pair is used to initialize the UKF, which employs the same curved-ray TDoA model for recursive motion-state estimation. A posterior Cramér–Rao lower bound (PCRLB) and computational-complexity analysis are further provided for performance assessment. Simulation results show that the proposed method achieves lower transient tracking errors and faster convergence than the conventional UKF, and that the measurement-informed covariance contributes to improved transient tracking performance.