NeuRSS: Enhancing AUV Localization and Bathymetric Mapping with Neural Rendering for Sidescan SLAM

TL;DR

NeuRSS tackles AUV localization drift and coarse bathymetric mapping from sidescan sonar by integrating neural rendering with SLAM. It uses an implicit SIREN-based bathymetry model to provide elevation priors that alleviate elevation degeneracy in SSS-SLAM, enabling iterative refinement of both vehicle pose and bathymetric maps. The framework extends a Lambertian sidescan model to account for nadir and shadows, and employs submap-based relative pose estimation with an elevation prior derived from neural rendering, followed by iSAM-based global optimization. Experiments on two field datasets demonstrate that neural-rendered elevation priors substantially improve localization accuracy and bathymetric fidelity compared to traditional priors, with iterative refinements yielding further gains. NeuRSS shows promise for offline high-fidelity mapping and suggests future directions for real-time fusion with MBES data and automatic data association.

Abstract

Implicit neural representations and neural rendering have gained increasing attention for bathymetry estimation from sidescan sonar (SSS). These methods incorporate multiple observations of the same place from SSS data to constrain the elevation estimate, converging to a globally-consistent bathymetric model. However, the quality and precision of the bathymetric estimate are limited by the positioning accuracy of the autonomous underwater vehicle (AUV) equipped with the sonar. The global positioning estimate of the AUV relying on dead reckoning (DR) has an unbounded error due to the absence of a geo-reference system like GPS underwater. To address this challenge, we propose in this letter a modern and scalable framework, NeuRSS, for SSS SLAM based on DR and loop closures (LCs) over large timescales, with an elevation prior provided by the bathymetric estimate using neural rendering from SSS. This framework is an iterative procedure that improves localization and bathymetric mapping. Initially, the bathymetry estimated from SSS using the DR estimate, though crude, can provide an important elevation prior in the nonlinear least-squares (NLS) optimization that estimates the relative pose between two loop-closure vertices in a pose graph. Subsequently, the global pose estimate from the SLAM component improves the positioning estimate of the vehicle, thus improving the bathymetry estimation. We validate our localization and mapping approach on two large surveys collected with a surface vessel and an AUV, respectively. We evaluate their localization results against the ground truth and compare the bathymetry estimation against data collected with multibeam echo sounders (MBES).

Figures (9)

• Figure 1: Illustration of landmark elevation degeneracy with pure $y\textrm{-}$translation motion, in the sensor forward-lateral-down frame. Here we show two submaps from two parallel survey lines with 3 landmarks (green dots). In the NLS optimization, we usually fix pose $\bm{x}_a$, namely fixing the solid red circles (indicating the range measurements). Without any priors on landmarks, $\bm{x}_b$ and all the landmarks can move together in $y\textrm{-}z$ plane, in this case, $\bm{x}_b$ positive translation along the $y\textrm{-}$axis to $\bm{x}_b^{'}$, landmarks moving up along the $z\textrm{-}$axis (from solid strokes to dashed strokes), where all SSS range and bearing measurements are still fulfilled.
• Figure 2: (a) Illustration of the gradient descent approach to find the intersection between the elevation arc and the seafloor, parameterized by SIREN. (b) An example of a SSS image in Dataset 1. (c) An example of a SSS image in Dataset 2, showing the sinkhole on the seabed.
• Figure 3: Factor graph formulation of the proposed framework. Left: Pose graph of the global optimization. Right: factor graph for submap-based relative pose estimation with the elevation prior.
• Figure 4: MMT survey vessel Ping (a) and Hugin AUV (b).
• Figure 5: Mean and standard deviation of the RTE at $thres2=0.5, 0.6, 0.7, 0.8, 0.9$. Black bars are the DR errors before optimization. Red bars and green bars are errors after optimization using SIREN estimates and linear interpolation of altimeter readings as the elevation prior, respectively. Blue bars are the errors without using any elevation priors.