A Hybrid Physics-Data-Driven Navigation Method for AUVs Fusing Hydrographic Information With INS/DVL Integration

In GNSS-denied underwater environments, the navigation accuracy of inertial navigation systems (INSs) aided by Doppler velocity logs (DVLs) is often severely degraded by complex hydrographic conditions. To address this, this article proposes a hybrid physics and data-driven integrated navigation method that fuses real-time hydrographic information to enhance performance. The approach establishes a layered error-suppression framework. At the physical layer, conductivity-temperature-depth (CTD) sensors are used to construct real-time sound speed profiles, applying Snell's law-based ray tracing to correct DVL sound-speed scale factors and refraction-induced geometry errors. At the data-driven layer, a hybrid CNN-MLP network intelligently estimates and compensates for residual ocean-current velocity disturbances remaining after physical corrections. Sea trial results validate the complementary contributions of the hybrid architecture: the physical layer effectively corrects acoustic geometric errors (reducing velocity RMSE by ∼35%), while the data-driven layer compensates for dynamic environmental residuals, ultimately reducing velocity RMSE by approximately 60% and suppressing position error by around 80%. This study validates that combining physics-based models with learning-based corrections effectively overcomes drift caused by DVL bottom-lock loss and time-varying currents, enabling high-accuracy, longendurance autonomous underwater vehicle (AUV) operations.