Design and Experimental Validation of an Autonomous USV for Sensor Fusion-Based Navigation in GNSS-Denied Environments

Samuel Cohen-Salmon & Itzik Klein · 2025

Read Design and Experimental Validation of an Autonomous USV for Sensor Fusion-Based Navigation in GNSS-Denied Environments by Samuel Cohen-Salmon & Itzik Klein (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-23445.txt.

Introduction

Testing navigation software for boats usually needs a boat, and satellite positioning is often switched off in the scenarios that matter most. Samuel Cohen-Salmon and Itzik Klein built MARVEL, a small autonomous surface vessel designed as a cheap, portable, seaworthy testbed. It was built for navigation in GNSS-denied environments, where satellite signals cannot be trusted. It carries layered sensors: electromagnetic and Doppler velocity logs, inertial units, and a precise real-time satellite receiver for ground truth.

Sea trials showed the vessel stable, maneuverable, and adaptable in challenging conditions, streaming synchronized data from its redundant sensors at high frequency. The authors offer MARVEL as an open-ended platform. Other researchers can validate sensor fusion and AI algorithms in real water rather than only in simulation, without the cost of a full research vessel.

Catalog record

CREATORSamuel Cohen-Salmon & Itzik Klein
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-23445

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Design and Experimental Validation of an Autonomous USV for Sensor Fusion-Based Navigation in GNSS-Denied Environments Samuel Cohen-Salmon & Itzik Klein Abstract This paper presents the design, development, and experimental validation of MARVEL, an autonomous unmanned surface vehicle built for real-world testing of sensor fusion-based navigation algorithms in GNSS-denied environments. MARVEL was developed under strict constraints of cost-efficiency, portability, and seaworthiness, with the goal of creating a modular, accessible platform for high-frequency data acquisition and experimental learning. It integrates electromagnetic logs, Doppler velocity logs, inertial sensors, and real-time kinematic GNSS positioning. MARVEL enables real-time, in-situ validation of advanced navigation and AI-driven algorithms using redundant, synchronized sensors. Field experiments demonstrate the system’s stability, maneuverability, and adaptability in challenging sea conditions. The platform offers a novel, scalable approach for researchers seeking affordable, open-ended tools to evaluate sensor fusion techniques under real-world maritime constraints. I Introduction Unmanned surface vehicles (USVs) are becoming essential tools across a range of maritime applications. These platforms have been deployed in domains including environmental monitoring, oceanographic data collection, port security,search and rescue, and offshore inspection [ 1 , 2 , 3 ] . Depending on their mission profiles, USVs vary widely in size, architecture, propulsion, and endurance. From compact one-person deployable units to large, long-endurance ocean-going vessels, the diversity of USV designs continues to grow [ 4 , 5 ] . Hull types such as monohulls, catamarans, and trimarans each offer distinct trade-offs in stability, payload capacity, and hydrodynamic efficiency [ 6 , 1 , 7 ] . Similarly, the choice of propulsion systems depends on the mission. Jet drives offer high maneuverability for patrol applications, while propeller-based configurations provide consistent performance for survey and research tasks [ 8 , 2 ] . Increasingly, hybrid and renewable-powered USVs—using solar panels, wave energy, or sails—extend operational endurance with minimal supervision. These developments reflect a broader shift toward autonomy and sustainability in maritime systems [ 3 ] . A critical enabler of USV autonomy is their navigation system. Accurate and reliable navigation is essential not only for mission execution, but also for safety. Navigation systems typically combine multiple sensors, such as global navigation satellite systems (GNSS), inertial measurement units (IMU), Doppler velocity logs (DVL), and electromagnetic (EM) speed logs [ 9 , 7 ] . These sensors work together through sensor fusion algorithms to estimate the vehicle’s state in real time, particularly when GNSS signals are unavailable. In recent years, deep-learning algorithms are incorporated within the navigation algorithms [ 10 , 11 ] . One of the first applications of deep learning to underwater navigation was construction of missing DVL beams, which in turn enabled the estimation of the vehicle velocity vector [ 12 , 13 , 14 ] . In USVs, deep learning has been applied to enhance inertial navigation systems and dead-reckoning accuracy, as well as to interpret sonar data for improved perception and environmental mapping [ 15 , 16 ] . On the surface, USVs benefit from deep learning integration in navigation, guidance, and control pipelines, supporting operations in complex maritime environments. Reinforcement learning, in particular, has enabled USVs to learn adaptive navigation strategies through interaction with dynamic surroundings, offering a promising alternative to traditional model-based controllers [ 17 , 18 ] . Other approaches, employ deep learning methods to adaptively estimate the process noise covariance in the inertial DVL fusion process [ 19 , 20 , 21 ] . Such algorithms rely on high-quality and diverse datasets for training and testing. In addition, most literature works employ simulation or semi-experimental approaches. In the latter, a single sensor is used to record actual data serving as the ground truth. The unit under test is a generated sensor created by adding a noise model to the ground truth (GT). These two gaps pose a challenge for researchers aiming to evaluate navigation algorithms under real-world conditions. To address these limitations, we present the design and development of MARVEL — a modular AI research vessel for experimental learning. MARVEL is a cost-effective, fully autonomous USV built to support real-time sensor fusion experimentation and autonomous navigation research. Its modular architecture supports easy integration of multiple redundant sensors, including DVLs, EM logs, IMUs, and GNSS. The system was designed to be portable, seaworthy, and reconfigurable, enabling consistent performance across a range of sea conditions. Unlike many commercially available platforms, MARVEL is tailored for high-frequency data acquisition, extended deployments, and affordable customization. It was developed using off-the-shelf components, open-source software, and a low-cost chassis modified for stability and robustness. Through iterative field testing, the platform demonstrated high reliability, maneuverability, and adaptability in both calm and challenging marine environments. Once the system architecture was finalized, software development focused on implementing advanced navigation, guidance, and control algorithms tailored for real-world maritime conditions. The testing phase consisted of laboratory-based validation of electronic, control, and power subsystems, followed by real-world trials under progressively challenging sea states. These field tests provided critical insights into MARVEL’s performance, allowing for iterative refinements and system optimizations. Through a combination of modular design, open-source software, and cost-efficient hardware selection, MARVEL was developed to serve as a versatile and scalable research platform, supporting a wide range of sensor fusion experiments in autonomous marine navigation. The rest of the paper is organized as follows: Section II presents the scope and design objectives of the platform. Section III describes the system design, including hardware architecture and sensor integration. Section IV details the assembly process, testing methodology, and experimental validation and Section V concludes this work. II Core Design Principles As illustrated in Figure 1 , the development was guided by six core principles: accessibility, ease of use, low cost, seaworthiness, autonomy, and open-source compatibility. While seaworthiness and autonomy were key technical targets, the platform was also intentionally designed to be accessible, easy to operate, and low-cost to ensure it could serve a wide range of users and research institutions. Accessibility was addressed by using widely available hardware, modular open-source software, and clear documentation, allowing the system to be operated and adapted by students and early-career researchers. Ease of use was achieved by simplifying setup, calibration, and control processes, streamlining field deployment even for users with limited technical backgrounds. The low-cost approach relied on off-the-shelf components, reused lab materials, and open-source tools, resulting in a robust yet budget-conscious solution. The primary goal was to design, build, and validate a cost-effective, fully autonomous USV capable of operating in diverse marine environments while meeting specific operational requirements. The USV needed to be seaworthy, capable of maintaining stability in wave heights up to 2\text{\,}\mathrm{m} , while sustaining a steady course at a minimum speed of 2\text{\,}\mathrm{kn} , even under wind conditions reaching 20\text{\,}\mathrm{kn} . Portability was another fundamental consideration, requiring the system to be easily handled by a two-person team for seamless transportation, deployment, maintenance, and retrieval in field experiments. Sensor integration was at the core of MARVEL’s design, allowing for the seamless inclusion of multiple navigational and environmental sensors. Ensuring high maneuverability was a critical design objective, requiring precise control under dynamic maritime conditions. Differential thrust control, coupled with advanced navigation algorithms, enabled station-keeping capabilities and precise path tracking. Cost optimization was also prioritized throughout the project, leveraging pre-existing hardware components and minimizing financial investment by repurposing available resources.

Read the complete text