Robust multi-sensor fusion for unmanned underwater vehicle navigation

The multi-sensor fusion of the inertial navigation system, underwater transponder positioning (UTP) system, Doppler velocity log, and depthometer represents an optimal approach to maintaining bounded navigation errors for Unmanned Underwater Vehicles (UUVs). However, this fusion process is susceptible to divergence due to three key challenges. Firstly, conventional time-of-arrival ranging methods for UTP necessitate strict time synchronization, which is difficult to maintain over extended periods, such as several months or years. Secondly, errors in the underwater sound velocity can significantly impact the accuracy of UTP ranging. Thirdly, in challenging underwater acoustic environments, the error statistics of UTP measurements often deviate from a Gaussian distribution and are characterized by numerous outliers. To address these issues, we propose a novel UTP measurement model based on round-trip time and implement an online estimation of the effective sound velocity. Additionally, we employ a Gaussian mixture model to approximate arbitrary non-Gaussian noise distributions and design a robust filter accordingly. The effectiveness and robustness of the proposed algorithms are validated through semi-physical field experiments conducted with a UUV prototype in real oceanic conditions.