Compensating for Partial Doppler Velocity Log Outages by Using Deep- Learning Approaches

Autonomous underwater vehicles allow researchers to explore the ocean depths and play an important role in many marine applications. A Doppler velocity log (DVL) is commonly used in autonomous underwater vehicle navigation. It measures four beam velocities to estimate the vehicle velocity vector. When less than four beams are available, the accuracy of the velocity vector estimation degrades or, in some situations, an estimate is not available at all. In real-life scenarios such situations commonly occur when the autonomous underwater vehicles is operating in complex environments or when passing or over trenches in the seafloor. This paper proposes a deep learning approach to compensate for situations of partial beam measurements. To that end, past DVL beam measurements are plugged into a dedicated network to regress the missing beam velocity. Once obtained, it is combined with the other three measured DVL beams to estimate the vehicle velocity vector. To examine the proposed approach, a simulated dataset of an autonomous underwater vehicle equipped with a DVL, was generated. Our results show that the proposed approach is capable of accurately estimating the missing DVL beam and, as a result, improving the estimation of the vehicle velocity vector. Sea experiments, made with the Snapir autonomous underwater vehicle at the Mediterranean sea, shows that the proposed approach works well even with sea recorded data. There, an improvement of more than 57% in the accuracy of the velocity vector estimation was achieved.