LSTM-based Dead Reckoning Navigation for Autonomous Underwater Vehicles

Autonomous Underwater Vehicles (AUVs) do need to be equipped with highly-accurate and robust navigation systems in order to perform challenging operations and complex missions. Nevertheless, since the Global Positioning System (GPS) is not a feasible, functional solution in the underwater scenario, the localization task is fulfilled by employing filtering techniques - as Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) - or Dead Reckoning (DR) strategies. Even though the specific algorithm architecture may vary, such methodologies usually rely on the direct linear speed measurements provided by specialized and expensive sensors, such as the Doppler Velocity Log (DVL). Thus, DVL failures or fallacies alongside DVL-denied environments may arise as unexpected causes for a severe malfunction of the whole navigation system. Motivated by the aforementioned considerations and the outstanding performance of Long Short-Term Memory (LSTM) neural networks in time-series regression problems, an LSTM-based DR approach has been developed to estimate the surge and sway body-fixed frame velocities, without canonically employing the DVL sensor.