Terrain-aided navigation (TAN) provides absolute positioning for underwater vehicles during long-term operations. However, conventional filter-based TAN systems typically do not explicitly model inertial measurement unit (IMU) biases as part of the state. To address the issue, this paper proposes a novel terrain-aided tightly coupled system based on invariant federated filter to attenuate the impact of IMU biases on navigation accuracy. Unlike conventional TAN architectures composed of three relatively independent modules, the proposed system directly fuses raw sensor measurements within a unified invariant federated filtering framework. In the filter, IMU biases and navigation errors are jointly augmented into the state vector. The corresponding state-space model is derived based on the Lie-group strapdown inertial navigation system error model. By explicitly separating IMU biases from navigation states, the proposed model improves estimation accuracy. Simulation results demonstrate that the proposed system provides accurate and robust positioning results for different initial misalignment angles, different initial position errors, and IMUs of different grades.
Accurate sound speed profiles (SSPs) are critical for unmanned underwater vehicle (UUV) under-ice ranging. In Liaodong Bay's ice-covered waters, sea-ice freezing and melting cause rapid temperature and salinity changes, while sparse observations and empirical formulas inadequately characterize local conditions, producing coupled errors in sound speed estimation and spatiotemporal reconstruction. This paper proposes an under-ice SSP reconstruction method integrating regional correction of the Coppens sound speed formula with Transformer-SIC-LoRA, a sea ice-conditioned Transformer using Sea Ice-Conditioned Low-Rank Adaptation (SIC-LoRA). The regional Coppens correction reduced mean bias error from 0.40 m/s to -0.02 m/s, validating effective calibration under limited samples. On independent test data, Transformer-SIC-LoRA achieved 0.43 m/s RMSE, improving approximately 48.2% over MLP and 60.9% over EOF methods. Under-ice simulations showed that with the UUV positioned 2 m below ice, the all-angle mean vertical ranging error was 0.50 cm, reaching only 2.47 cm at 70° beam angle. These results demonstrate that the method provides reliable sound speed correction for upward-looking multibeam under-ice measurements by UUVs in ice-covered waters.