A Switching Framework for Robust Underwater Pose Estimation: Integrating IPNet with Vision-based SLAM

Abstract Autonomous underwater vehicles (AUVs) carry out a wide range of tasks in underwater environments. Maintaining continuous, high-precision pose estimation of the AUV over extended durations is critical for its mission success and operational safety. To address the limitations of vision-based simultaneous localization and mapping (SLAM) in underwater environments, particularly trajectory tracking failures caused by visual information scarcity, this paper proposes a novel pose estimator switching framework. The proposed framework enhances the robustness of AUV pose estimation by incorporating two key components. First, we designed a visual monitor that continuously assesses the work state of the SLAM system. Next, we developed a neural network-based pose estimator driven by Inertial Measurement Unit(IMU) data and combined with pressure measurements, which is called IPNet Estimator. When visual information becomes unavailable, the system seamlessly switches to this alternative estimator, ensuring uninterrupted pose estimation for the AUV. Finally, experiments conducted on a publicly available underwater dataset validate the feasibility and effectiveness of the proposed framework.