A Landmark-Aided Navigation Approach Using Side-Scan Sonar

TL;DR

This work tackles GPS-denied underwater navigation by leveraging identifiable seafloor landmarks detected with Side-scan Sonar (SSS). It introduces a Bayesian landmark-aided navigation framework that combines a prediction step using the unscented transform with a particle-based update that performs probabilistic data association of SSS detections to known landmarks, enabling real-time operation despite the nonlinear slant-range geometry. The approach integrates a joint data-association model (MOU) and gating to handle false positives and missed detections, and evaluates performance through forward simulation and field experiments on two platforms across two sites, showing that landmark sightings bound localization error (mean RMSE around 1.7–1.8 m in field tests) and that the method remains computationally feasible for real-time deployment. The work demonstrates the practicality of low-cost onboard sensing for ocean monitoring and lays out concrete steps toward automatic landmark detection and SLAM-enabled operation in unknown environments, with significant implications for scalable, autonomous underwater data collection.

Abstract

Cost-effective localization methods for Autonomous Underwater Vehicle (AUV) navigation are key for ocean monitoring and data collection at high resolution in time and space. Algorithmic solutions suitable for real-time processing that handle nonlinear measurement models and different forms of measurement uncertainty will accelerate the development of field-ready technology. This paper details a Bayesian estimation method for landmark-aided navigation using a Side-scan Sonar (SSS) sensor. The method bounds navigation filter error in the GPS-denied undersea environment and captures the highly nonlinear nature of slant range measurements while remaining computationally tractable. Combining a novel measurement model with the chosen statistical framework facilitates the efficient use of SSS data and, in the future, could be used in real time. The proposed filter has two primary steps: a prediction step using an unscented transform and an update step utilizing particles. The update step performs probabilistic association of sonar detections with known landmarks. We evaluate algorithm performance and tractability using synthetic data and real data collected field experiments. Field experiments were performed using two different marine robotic platforms with two different SSS and at two different sites. Finally, we discuss the computational requirements of the proposed method and how it extends to real-time applications.

Figures (10)

• Figure 1: Hydrone surface platform (a) and Iver3 underwater vehicle (b) used for sss data collection and performance evaluation.
• Figure 2: A sss sensor takes cross-track measurements. The maximum range of the SSS sensor, $r_{\mathrm{max}}$, shows the farthest point observed by the sss. When a landmark is present, its respective slant range, $r_s$, can be extracted from the sss image. This slant range corresponds to a horizontal distance, $r_h$. The vehicle altitude above the seafloor, $\gamma$, is the third side of this triangle and is needed to obtain $r_h$ from $r_s$.
• Figure 3: sss image with landmarks annotated by boxes. The horizontal axis shows the slant range. Negative distance corresponds to the port side, and positive distances correspond to the starboard side. This SSS image shows approximately 2 minutes of data, and the maximum slant range at each side is 20m. The center frequency of the transmitted pulse is $1800$ kHz, and the ping rate is 30 Hz, meaning that each line of pixels is about 33 ms apart. The dark section at the center of the image is referred to as the nadir. This data was collected at the Scripps Institute of Oceanography using the sensing platform shown in Fig. 4.
• Figure 4: Flow diagram describing the order of operations in the navigation algorithm. Blocks shown in blue depict the main processing steps and their results. The sub-process in red shows details of the particle-based update step, which includes probabilistic association to handle mou. The probabilistic association of landmarks and detections is performed in the weight computation portion of the update step.
• Figure 5: (a) Hydrone surface vehicle with integrated Ark Scout Mk IIsss collecting data in Mission Bay, CA. White and red buoys mark deployed landmarks. (b) Iver3 vehicle with EdgeTech SSS collecting data at the Scripps Institution of Oceanography in La Jolla, CA.