Autonomous Underwater Vehicles (AUVs) are widely employed for deep sea exploration and underwater operations, but their navigation performance is often degraded in complex environments due to time-varying measurement noise, abnormal observations, and Doppler Velocity Log (DVL) outages. To address these challenges, this paper proposes an integrated SINS/DVL/PS navigation framework that combines an Adaptive Huber and Sage-Husa Extended Kalman Filter (AHR-EKF) with a Nonlinear AutoRegressive with eXogenous inputs (NARX)-based velocity prediction model. The AHR-EKF effectively suppresses outliers and adapts to time-varying noise, thereby enhancing filter stability and state estimation accuracy. During DVL outages, the NARX model predicts short-term AUV velocity using propeller speed, velocity increments from the navigation system, and attitude information as exogenous inputs. This data-driven approach compensates for lag and mismatch in propeller-based velocity measurements, while capturing both short-term fluctuations and overall velocity trends. Simulations and sea trials were conducted to validate the method. In the simulation experiment during DVL outages, the V-NARX method achieved east and north positioning of RMS errors of 8.397 m and 6.530 m, compared with 24.699 m and 10.218 m for the V-RPM method. In the sea trial, the V-NARX method achieved east and north RMS errors of 41.160 m and 28.023 m, respectively, compared with 52.820 m and 67.057 m for V-RPM, corresponding to reductions of 22.1% and 58.2%. The proposed method maintains trajectory continuity and effectively suppresses rapid INS error accumulation during DVL outages, significantly enhancing emergency navigation capability under DVL outages. Although its positioning accuracy does not match that of normal DVL operation, the method provides a practical and reliable engineering solution for continuous AUV navigation when DVL is unavailable.
🌊
