Research on Underwater Integrated Navigation Method Based on BiLSTM and Adaptive Factor Graph Optimization

Underwater autonomous vehicles (AUVs) are essential for marine exploration and engineering, but their navigation systems face challenges in complex underwater environments, including sensor degradation, noise interference, and the absence of GNSS signals. This paper proposes a BiLSTM-based adaptive factor graph optimization (AFGO) method to enhance both the accuracy and robustness of AUV navigation. The method integrates measurements from multiple sensors, including SINS, DVL, and LBL, and dynamically adjusts observation weights by leveraging the BiLSTM to learn the statistical properties of sensor errors. By utilizing a sliding-window factor graph, the system continuously adapts to sensor failures and environmental noise, ensuring high estimation accuracy and robustness over time. Experimental results demonstrate that the proposed method significantly outperforms traditional factor graph algorithms in three representative underwater scenarios, especially under conditions of sensor degradation, highlighting its ability to maintain reliable performance and rapidly recover from observation loss.