Probabilistic Methods for Improved Self-consistency of Doppler Velocity Log Range-Based Bathymetry

We develop a probabilistic approach to improve the self-consistency of Doppler Velocity Log (DVL) derived bathymetric maps that models the interactions of acoustic beams with uneven seafloors. DVLs measure velocity and altitude over the seafloor using Doppler shift and time-of-flight (TOF) of reflected acoustic signals, and are integral to underwater vehicle navigation. The proposed method combines acoustic propagation and reflection theory with Gaussian process regression (GPR) to model terrain bias in DVL range measurements. We compare three bathymetric mapping methods using DVL range data gathered during field trials with the University of Southampton’s Smarty200 AUV. Our results demonstrate that accounting for terrain bias improves the self-consistency of maps, implying potential use of this method to identify loop closures for enhanced underwater navigation.