DCNet: A Data-Driven Framework for DVL Calibration

TL;DR

This work introduces DCNet, a data-driven end-to-end calibration framework for DVL measurements in AUV navigation. By using a two-headed neural network with 1D and dilated 2D convolutions, DCNet jointly processes DVL and GNSS-RTK data to estimate five DVL error models, including vector-scale and vector-bias terms. On real Snapir AUV data, EM5 (vector scale and bias) with DVL2 achieved about 70–72% RMSE improvements and roughly 80% faster calibration compared with a baseline scalar-scale approach, demonstrating a practical path to high-accuracy, low-cost DVL calibration on nearly constant-velocity trajectories. The results indicate significant potential for broader deployment of low-cost DVLs in marine robotics, enabling accurate navigation with shorter calibration times.

Abstract

Autonomous underwater vehicles (AUVs) are underwater robotic platforms used in a variety of applications. An AUV's navigation solution relies heavily on the fusion of inertial sensors and Doppler velocity logs (DVL), where the latter delivers accurate velocity updates. To ensure accurate navigation, a DVL calibration is undertaken before the mission begins to estimate its error terms. During calibration, the AUV follows a complex trajectory and employs nonlinear estimation filters to estimate error terms. In this paper, we introduce DCNet, a data-driven framework that utilizes a two-dimensional convolution kernel in an innovative way. Using DCNet and our proposed DVL error model, we offer a rapid calibration procedure. This can be applied to a trajectory with a nearly constant velocity. To train and test our proposed approach a dataset of 276 minutes long with real DVL recorded measurements was used. We demonstrated an average improvement of 70% in accuracy and 80% improvement in calibration time, compared to the baseline approach, with a low-performance DVL. As a result of those improvements, an AUV employing a low-cost DVL, can achieve higher accuracy, shorter calibration time, and apply a simple nearly constant velocity calibration trajectory. Our results also open up new applications for marine robotics utilizing low-cost, high-accurate DVLs.

Figures (9)

• Figure 1: Five DVL error models divided into baseline models EM1 - EM4, and our proposed EM5.
• Figure 2: Block diagram of our DCNet architecture used to estimate one of the proposed EM1-EM5 error terms.
• Figure 3: Illustration of two 2D convolution kernels sized $2\times2$ and how the dilated kernel processes both GNSS and DVL velocity vectors. Figure \ref{['2d_dil_procc_vels_fig']} shows the $3\times1$ dilated 2D convolution kernel processes both X axes of the DVL and GNSS velocities simultaneously. Figure \ref{['2d_dil_conv_subfig']} illustrates the $3\times1$ dilated 2D kernel, while Figure \ref{['2d_reg_conv_subfig']} shows a $1\times1$ 2D kernel, which in fact is referred to as not dilated.
• Figure 4: Block diagram showing the closed loop MSE loss calculation procedure after the proposed approach estimates the error terms.
• Figure 5: The "Snapir" AUV being pulled out of the water after a mission.