Data-Driven Strategies for Coping with Incomplete DVL Measurements

TL;DR

The paper tackles the challenge of incomplete DVL measurements in AUV navigation by comparing two data-driven regressor architectures, LiBeamsNet (1D-CNN) and MissBeamNet (LSTM), against a model-based average estimator. Using real Mediterranean Sea data collected with a Snapir AUV, both DL approaches outperform the baseline by over 16% in velocity prediction accuracy, with LiBeamsNet showing a slight edge in final performance and MissBeamNet displaying faster convergence. The work demonstrates that regression of missing DVL beams from past measurements can robustly sustain INS/DVL navigation during outages, improving reliability for deep-sea operations. This suggests practical value for deploying DL-based outage compensation in autonomous underwater missions, supported by public data and code availability.

Abstract

Autonomous underwater vehicles are specialized platforms engineered for deep underwater operations. Critical to their functionality is autonomous navigation, typically relying on an inertial navigation system and a Doppler velocity log. In real-world scenarios, incomplete Doppler velocity log measurements occur, resulting in positioning errors and mission aborts. To cope with such situations, a model and learning approaches were derived. This paper presents a comparative analysis of two cutting-edge deep learning methodologies, namely LiBeamsNet and MissBeamNet, alongside a model-based average estimator. These approaches are evaluated for their efficacy in regressing missing Doppler velocity log beams when two beams are unavailable. In our study, we used data recorded by a DVL mounted on an autonomous underwater vehicle operated in the Mediterranean Sea. We found that both deep learning architectures outperformed model-based approaches by over 16% in velocity prediction accuracy.

Figures (5)

• Figure 1: Illustration of scenarios where two or more DVL beams may be unavailable.
• Figure 2: The LiBeamsNet architecture, as introduced in cohen2022libeamsnet, presenting a 1DCNN for extracting features from $N$ past DVL beam measurements.
• Figure 3: The MissBeamNet architecture, as described in yona2023missbeamnet, employs an LSTM structure to analyze $N$ past DVL beam measurements, focusing on capturing longer-term dependencies.
• Figure 4: The loss values of both the train and test sets for LiBeamsNet as a function of the epoch number.
• Figure 5: The loss values of both the train and test sets for MissBeamNet as a function of the epoch number.