173
1 INTRODUCTION
Naval forces operate among the most complex
engineered systems developed by modern societies.
Warships, submarines, and associated maritime
systems are characterised by long service lives, harsh
and uncertain operating environments, strict safety
requirements, and continuously evolving mission
profiles. In parallel, Artificial Intelligence (AI) has
gained prominence as a general-purpose technology
with both civilian and military significance.
Contemporary naval forces are therefore witnessing
rapid and transformative technological changes that
affect nearly all aspects of naval warfare and maritime
security. Advances in unmanned systems, automation,
and digitalisation, together with other disruptive
technologies such as AI, are fundamentally reshaping
how naval capabilities are designed, developed, and
employed. This transformation is not a distant prospect
but an ongoing process that is already influencing
operational doctrines and force structures. As global
competition in military technology intensifies,
remaining passive or delaying adoption of these
technologies is no longer a viable option for modern
navies. AI has emerged as one of the primary drivers
of this technological evolution. Its ability to process
vast amounts of data, identify patterns, support
decision making, and enable autonomous or semi-
autonomous systems provides significant advantages
across the full spectrum of naval activities. Recent
military conflicts, most notably the Russia-Ukraine
war, have demonstrated the practical utility of AI-
enabled systems in operational environments. AI has
been employed extensively in unmanned platforms for
Integrating Artificial Intelligence into Naval Capability
Development
A. Ljulj
1
, I. Štambuk
1
& V. Slapničar
2
1
Dr. Franjo Tuđman Defense and Security University, Zagreb, Croatia
2
University of Zagreb, Zagreb, Croatia
ABSTRACT: The rapid advancement of artificial intelligence (AI) is transforming naval capabilities, reshaping
ship design, lifecycle management, operational decision-making, and autonomous maritime systems. Naval
platforms are among the most complex engineered systems, characterised by long service lives, safety-critical
functions, and demanding operational environments, making AI integration both strategically attractive and
technically challenging. This paper presents an engineering-oriented review of AI applications in the naval
domain, focusing on their role across the capability development lifecycle. To illustrate practical implementation,
a Random Forest regression model is developed to support early-stage prediction of the block coefficient of naval
ships. The review highlights significant opportunities associated with AI integration, including enhanced
decision-making, improved design efficiency, and increased operational effectiveness. However, successful AI
adoption requires technological advancement alongside organisational adaptation, strong governance, and
sustained investment in human expertise. AI should therefore be understood not as a replacement for naval
engineering expertise, but as a force multiplier that augments analytical capacity and accelerates innovation
across the maritime domain.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 20
Number 1
March 2026
DOI: 10.12716/1001.20.01.19
174
tasks such as object detection, tracking, target
recognition, mission planning, and real-time decision
support. These developments underscore the growing
importance of AI not only in combat operations but
also in upstream processes such as capability
development, system design, acquisition, and lifecycle
management.
In this context, the main objective of this work is to
review and analyse the influence of artificial
intelligence on naval capability development across a
broad range of domains. These domains include naval
ship design, naval project management, maintenance
and sustainment, education and training, simulations,
operational employment of naval forces, including
intelligence, surveillance, reconnaissance, and the
implementation of maritime unmanned systems. The
literature review was conducted using a structured
narrative approach aimed at capturing recent and
relevant developments in artificial intelligence
applications within naval capability development.
Scientific and professional publications were identified
through searches of major academic databases,
including Scopus, Web of Science, IEEE Xplore, and
ScienceDirect, supplemented by targeted searches of
defence and maritime research repositories. The
review focused primarily on publications from 2015 to
2025 to reflect contemporary technological maturity,
with emphasis on peer-reviewed journal articles,
conference proceedings, and authoritative technical
reports. Sources addressing purely generic or non-
maritime AI applications were excluded unless they
provided transferable methodologies or concepts
applicable to naval engineering contexts. The selected
literature was analysed qualitatively and synthesised
thematically to identify dominant application areas,
enabling technologies, maturity levels, and technical,
organisational, and regulatory challenges. This
approach enables a comprehensive yet focused
assessment of AI integration trends and challenges
specific to naval engineering and maritime defence
systems. In addition to the literature review, a practical
use case is presented to illustrate the real-world
application of AI methods in naval engineering.
Specifically, the use case demonstrates the prediction
of a ship’s block coefficient using a Random Forest
Model, representing one of the many AI-based tools
applicable to data regression problems in naval ship
design. The available literature is systematically
analysed and synthesised to provide a comprehensive
overview of current research trends, technological
capabilities, and implementation challenges. By
consolidating findings from diverse sources, this paper
aims to identify both the opportunities and limitations
associated with AI integration into naval capability
development, with particular attention to application
maturity and potential impacts on operational
effectiveness, cost efficiency, and decision-making
processes.
The remainder of the paper is structured as follows.
Section 2 presents definitions and classifications of
artificial intelligence relevant to the naval domain.
Section 3 provides a comprehensive review of existing
literature related to AI applications in naval project
management, naval ship design, maintenance and
logistics, training and education, simulations, AI-
enabled naval operations, and maritime unmanned
systems. Section 4 addresses key challenges associated
with integrating AI into naval capabilities, including
organisational adaptation, platform integration,
lifecycle sustainment, training requirements, and
considerations for future development. Section 5
presents a detailed use case of AI-based Random Forest
modelling for block coefficient prediction in naval ship
design. Finally, Section 6 summarises the key findings,
discusses implications for naval planners and
strategists, provides recommendations, and outlines
directions for future research.
2 DEFINING AND CLASSIFYING AI
TECHNOLOGIES
2.1 General importance of AI
Artificial Intelligence has emerged as a pivotal
technology in the military sphere, with leading nations
recognising its potential for rapid integration across
defence and national security domains. AI is
transforming the global security environment by
enhancing military effectiveness while simultaneously
accelerating the pace and complexity of emerging
threats, thereby influencing collective defence, crisis
management, and cooperative security. As advances in
AI diffuse quickly from civilian to military
applications, control over intelligent systems is likely
to confer a decisive advantage in future warfare.
Consequently, the military application of AI has the
potential to disrupt existing balances of power and
challenge strategic stability and deterrence among
major powers. The following text presents several
statements from the literature highlighting the
importance of AI. In [51] United States views its pursuit
of AI primacy as a critical element in safeguarding the
nation against military, scientific, economic, and
political threats. Under Xi Jinping, the People's
Liberation Army (PLA) is directed to become a world-
class military by mid-century, with AI seen as central
to this goal [25]. Ukraine aspires to be among the top
three countries in the world by 2030 in both AI
development and practical adoption [41]. The UK must
adopt and exploit AI at pace and scale for Defence
advantage, establishing AI as one of our top priorities
and a key source of strategic advantage [48]. The
Russian view on AI importance considered through
citation “The creation and development of systems is
currently becoming one of the most important areas of
scientific and technological progress, the very
fundamental technology that can radically change the
nature of not only armed struggle, but also the whole
essence of power confrontation between states,
including economic, information and cyber war” [6].
2.2 Definitions and classification of AI
In accordance with [39], AI is a set of interrelated
technologies used to solve problems and perform tasks
that, when performed by humans, require thinking. In
addition, Artificial Narrow Intelligence (ANI), also
referred to as weak AI or narrow AI, is the only type of
artificial intelligence we have successfully realised to
date. Narrow AI is goal-oriented, designed to perform
singular tasks, i.e., facial recognition, speech
recognition/voice assistants, driving a car, or searching
the internet, and is highly intelligent at completing the
specific task it is programmed to do. Machine Learning
175
(ML) uses statistical techniques to enable computer
systems to recognise patterns in data without explicit
programming. This pattern recognition can be used as
a basis for AI implementation. Machine learning can be
achieved through a range of methods that may not be
specific to a task.
Artificial Intelligence is the computational science
field of research that focuses on machine learning and
smart decision-making. It is a major component of
robotics R&D. This includes contributions from fields
such as machine learning, natural language processing,
pattern recognition, cluster algorithm improvement,
and agent technology [38]. AI is the capability
provided by algorithms of selecting optimal or
suboptimal choices from a wide possibility space, to
achieve goals by applying strategies that can include
learning or adapting to the environment [51]. In
accordance with [41] NIAG (NATO Industrial
Advisory Group) Study Group SG-238 AI definition is:
Artificial Intelligence refers to systems designed by
humans that, given a complex goal, act in the physical
or digital world by perceiving their environment,
interpreting the collected structured or unstructured
data, reasoning on the knowledge derived from this
data, and deciding the best action(s) to take (according
to pre-defined parameters) to achieve the given goal.
AI systems can also be designed to teach people how
to adapt their behaviour by analysing how the
environment is affected by their previous actions.
NATO Bilateral Strategic Command (BI-SC) final
report on Joint Air Power Capabilities (JAPC) turned to
the definition of the NIAG SG-231: Artificial
Intelligence is the ability of a nonbiological system to
achieve any complex goal through processes
comparable to human cognitive processes, such as
perception, deduction, recognition, memorisation, and
learning. Machine learning is one of the most
important technical approaches to AI and the basis of
many recent commercial applications of AI. Modern
machine learning is a statistical process that starts with
data and tries to derive a rule or procedure that
explains the data or can predict the future. Deep
Learning is a subspecialty of Machine Learning that
has yielded some notable successes in the development
and prototyping of autonomous systems/vehicles.
Deep learning uses structures loosely inspired by the
human brain, consisting of a set of units (or “neurons”).
Each unit combines a set of input values to produce an
output value, which in turn is passed on to other
neurons. Deep learning networks use many layers, and
often use many units at each layer, to enable the
recognition of extremely complex, precise patterns in
data [7].
In conclusion, artificial intelligence encompasses a
range of interconnected technologies that enable
nonbiological systems to perceive, reason, learn, and
act to achieve complex goals in ways comparable to
human cognition. While current AI is primarily
realised as Artificial Narrow Intelligence (ANI),
focused on specific tasks, its capabilities are largely
driven by machine learning and deep learning
techniques that extract patterns from data to support
intelligent decision-making. Together, these
approaches form the foundation of modern AI
applications across domains such as robotics,
autonomous systems, and data-driven problem-
solving.
3 CONTEMPORARY USES OF AI IN THE
DEVELOPMENT OF NAVAL CAPABILITIES
3.1 AI in ship design and project management
Artificial Intelligence may enhance defence contracting
processes, including naval projects, by addressing
inefficiencies in the US DoD’s (Department of Defense)
slow and complex procurement system [43]. The study
proposes using AI, particularly natural language
processing (NLP) and machine learning (ML), to
automate routine contract management tasks, analyse
and generate documentation, and reduce
administrative burden while improving consistency
and compliance. AI-driven tools could streamline
approvals, flag issues, and provide predictive insights
based on historical data to support better contract
strategies. Overall, adapting commercial AI contract
management systems for DoD use could modernise
procurement by increasing efficiency, reducing errors,
and accelerating acquisition timelines. Fig. 1 presents
the R.A.F.T. AI assistant, which is used throughout all
phases of the acquisition process.
Figure 1. Acquisition Challenges and R.A.F.T. Solutions
(Source: [43])
The paper [10] proposes a naval project
management system that integrates generative AI and
machine learning across the entire project lifecycle,
from design to launch. Generative AI is used to
automate key functions such as document analysis,
activity planning, cost and timeline forecasting, and
risk management, reducing manual effort and
improving accuracy. AI-driven modules can extract
regulatory and technical information, optimise
schedules and resource allocation using historical data
and optimisation algorithms, and enhance forecasting
by identifying patterns from previous projects. The
system also supports proactive risk management by
analysing performance data and regulatory changes.
The authors apply Adaptive Structuration Theory
(AST) to explain how managers can adopt these
technologies. A modular microservices and event-
driven architecture is proposed to ensure flexibility,
scalability, and maintainability in complex naval
projects.
The modernisation of the Indian Navy increasingly
involves cutting-edge technologies where AI plays a
significant role in future naval capabilities, a domain in
which micro, small, and medium enterprises (MSMEs)
and technology partners are expected to contribute
through research, design, and innovation. AI-enabled
combat systems and autonomous platforms are central
to next generation naval technologies, improving
situational awareness and decision-making at sea. AI
integration enhances the Navy’s operational
176
effectiveness by processing vast sensor data for faster
threat detection, predictive maintenance, and
optimised mission readiness. AI is considered a
cornerstone of naval modernisation efforts, powering
autonomous unmanned systems, automating complex
tasks, and empowering more agile and resilient
maritime operations [37].
The DAS (Daewoo Shipbuilding Scheduling)
project implemented a neural network (NN) based
person-hour estimator to provide reliable estimates of
man-hour requirements for different assembly tasks,
which traditional methods could not capture
accurately due to complex, nonlinear relationships
between work factors. This neural estimator was
integrated into the broader intelligent scheduling
architecture to improve planning accuracy and support
dynamic scheduling decisions in the shipyard. By
embedding neural networks within the scheduling
system, the DAS approach enhanced the overall
scheduling performance and productivity in
shipbuilding operations, contributing to more effective
long term and shop-floor scheduling outcomes [27].
Article [17] explains that artificial intelligence is a
core enabler of digital twins (DT) for warship systems,
allowing virtual models to evolve dynamically by
learning from real-time and historical sensor data.
Machine learning and AI algorithms are used to
analyse large volumes of shipboard data to support
predictive maintenance, anomaly detection (AD), and
performance deviation analysis, and moving digital
twins beyond static simulations. AI enhances digital
twins by continuously recalibrating behaviour models,
improving diagnostic accuracy, and enabling
prescriptive decision-making for maintenance and
operations. The integration of AI also enables higher-
level system understanding through data-driven
prediction and optimisation across the warship
lifecycle. AI transforms digital twins into intelligent,
adaptive systems that improve operational readiness,
lifecycle management, and decision support for
complex naval platforms.
The review [23] outlines diverse AI and machine
learning techniques applied to ship design, with deep
neural networks (DNN) widely used as surrogate
models to predict ship resistance at near CFD accuracy
but far lower computational cost. Generative models
such as generative adversarial networks (GAN) and
variational auto encoders (VAE), combined with
optimisation algorithms, support automated hull-form
exploration and resistance reduction. Physics-
informed and hybrid ML approaches, along with
methods such as support vector machine (SVM) and
ensemble trees (ET), enhance robustness when data are
limited, while genetic and evolutionary algorithms
remain key for multi-objective structural optimisation.
Emerging methods such as reinforcement learning
(RL) and graph neural networks (GNN) further expand
capabilities in sequential design decision making and
fast structural response prediction.
The authors develop a deep learning neural
network (DLNN) model to assist with the preliminary
design of ship hull structures by enabling real-time
prediction of total resistance based on geometric
modification parameters, dramatically reducing
reliance on computationally intensive CFD
calculations. The trained model demonstrated high
accuracy in predicting resistance within the design
space, with an average testing error below 4%, and can
be used interactively during early design phases. This
AI-based approach represents a foundational step
toward integrating artificial intelligence into naval
architecture workflows to streamline hull design and
performance evaluation [2].
An AI-based approach that employs neural
networks and case-based reasoning (CBR) to automate
metallographic analysis for assessing metal quality in
shipbuilding, improving traditional manual
diagnostics, is presented in [15]. Their multilayer
neural network accurately identifies and quantifies
metal microstructures, enabling the reliable
determination of metal grades. The developed AI-
driven software demonstrates high accuracy and offers
a practical tool to streamline quality assessment in
shipbuilding processes.
The paper [31] develops a multi-attribute concept
design procedure for generic naval vessels that
integrates a self-balanced concept design model with a
genetic algorithm (GA) (artificial intelligence
evolutionary computing) and Pareto optimal search to
generate and evaluate balanced design solutions. The
methodology uses a set of geometric, tactical, and
technical design variables to produce designs that are
evaluated on attributes such as life cycle cost and
overall measure of effectiveness, identifying a set of
non-dominated (Pareto optimal) solutions.
Using an artificial neural network (ANN) model,
the authors try to improve the ship preliminary design
process with the aim to speed it up using minimal
resources. Fig. 2 presents the training data set and
predicted design parameters such as length, breadth,
and displacement of a ship. The method used provides
very good accuracy of predicted data using only two
input parameters and shows the efficiency of the
applied AI method [33].
Figure 2. Predicted ship design parameters using the ANN
model (Source: [33])
Grech La Rosa in [20] investigates how generative
artificial intelligence (GAI) can be integrated into the
early stages of concept ship design to support
designers in handling complex tasks and decision-
making that traditionally rely on approximation and
expert judgment. It explores the potential of GAI tools
to enhance various aspects of design workflows,
including weight grouping, payload catalogues,
technical analysis, and layout configuration to improve
performance, sustainability, and inclusivity. The
authors use a case study to examine how outputs from
GAI models (e.g., text and image generation) can
177
contribute constructively to design development and
evaluation. This exploratory work highlights the
opportunities and challenges of incorporating GAI into
naval architecture and suggests that such tools could
become valuable aids in conceptual design processes.
The work [9] reviews the use of artificial intelligence
across the ship lifecycle, including design,
manufacturing, operation, and maintenance. Machine
learning and optimisation methods improve decision
making, reduce trial-and-error processes, and
accelerate design cycles. In early design stages, AI
supports performance prediction and rapid
exploration of design alternatives. During
manufacturing, AI enhances CAD/CAM automation,
anomaly detection, and quality control, improving
efficiency and accuracy. In operation and maintenance,
predictive analytics help forecast failures and optimise
maintenance schedules. The article also highlights the
role of deep learning (DL) and reinforcement learning
(RL) in improving complex structural and
hydrodynamic optimisation tasks.
The study [35] reviews AI applications across all
phases of ship design, from concept development and
detailed analysis to production and performance
optimisation. It finds that machine learning, data
analytics, and automation are increasingly used to
reduce time and cost while improving the quality and
reliability of design outcomes beyond conventional
methods. AI is also applied in shipyards to enhance
production planning and resource allocation, as well as
to optimise operational performance, including fuel
efficiency and lifecycle costs. However, the authors
emphasise that further efforts are needed to bridge the
gap between research advances and practical
implementation.
The literature identifies three main categories of AI
applications in naval ship design and project
management: administrative and decision support
automation, data-driven design optimisation, and
lifecycle performance prediction. Initial adoption has
concentrated on administrative tasks such as
procurement, scheduling, and cost estimation, where
AI offers efficiency gains with relatively low risk. More
advanced uses involve machine learning and
generative methods to optimise design spaces,
particularly in early design stages where uncertainty is
high and traditional tools are computationally
demanding. To conclude, the most effective
implementations combine AI with established naval
architecture methods, emphasising hybrid and
physics-informed approaches while recognising that
lifecycle applications depend on strong data
integration and organisational readiness.
3.2 AI in naval operations
The systematic review in [47] examines the integration
of AI into naval operations, particularly in surveillance
and reconnaissance for early threat detection. It
highlights techniques such as deep learning,
convolutional neural networks (CNN), and other
machine learning models that improve object
recognition, anomaly detection, and real-time
maritime data analysis. The study shows that
combining AI with radars, satellites, UAVs, and
predictive models enhances early warning capabilities
and enables more proactive defence responses, with
applications including unmanned vessels and nano
drones. While AI significantly strengthens operational
effectiveness, the authors emphasise the need for
continued investment in infrastructure, training, and
interoperability standards [47].
The paper [13] provides a survey of current AI
advancements in the Navy. It presents practical
examples of AI use in the Navy, including
productivity, navigation, logistics, threat detection,
and training. As autonomy increased, research
attention shifted toward human-machine interaction,
explainable AI (XAI), and trust in AI-enabled systems.
Talpur et al. in [46] review recent deep learning–
based AI techniques for maritime security, focusing on
improved surveillance, threat detection, and
situational awareness. They note that traditional
methods such as radars, satellite imagery, and patrol
vessels face coverage gaps and data overload, which
deep learning models, including convolutional neural
networks, recurrent neural networks (RNN), and
transformers, help address by analysing satellite, AIS,
SAR, radar, and unmanned sensor data. These systems
enhance vessel classification, anomaly detection, and
multimodal data fusion (DF), supporting identification
of illegal activities and advanced functions such as real-
time tracking and behaviour recognition. The authors
conclude that deep learning is transforming maritime
domain awareness, while emphasising the need to
overcome data, computational, and interpretability
challenges.
Vasankari & Saastamoinen in [49] investigate the
use of multi-agent reinforcement learning (MARL) to
support tactical decision making in complex littoral
naval combat environments, where real-world data are
limited, and conditions are dynamic and partially
observable. They model engagements as partially
observable stochastic games and implement double
deep Q networks (DDQN) and proximal policy
optimisation (PPO) algorithms to learn effective
tactical policies under uncertainty. Through
simulation-based training, agents can explore and
evaluate alternative strategies without relying on
extensive empirical datasets. The results indicate that
MARL can reinforce existing doctrines while also
generating novel strategic options, highlighting its
potential to reshape naval decision-making processes.
Johnson in [24] examines how AI can enhance the
naval tactical kill chain by supporting complex, time-
critical decisions under uncertainty. Using models
such as the OODA (Observe-Orient-Decide-Act) loop
and F2T2EA (Find-Fix-Track-Target-Engage-Assess),
the study maps 28 tactical functions to relevant AI
approaches, including machine learning, data fusion,
and cognitive AI (CAI) for tasks to support capabilities
such as target detection, classification, and engagement
assessment. It also considers broader approaches like
explainable AI and human-machine teaming to
improve decision support and operational
performance. While AI shows promise for
strengthening kill chain effectiveness, further research
is needed to refine methods, ensure integration, and
guarantee safe deployment.
A systematic review examining how AI is used to
enhance cybersecurity in maritime environments,
particularly for protecting vessels, ports, and maritime
178
communication networks from sophisticated cyber
threats, is presented in [34]. The review identifies and
analyses key AI techniques such as machine-learning-
based intrusion detection systems (IDS), anomaly
detection, predictive threat modelling (PTM), and AI-
enhanced zero-trust architectures that improve real-
time threat detection and automated responses. While
AI-driven approaches have shown notable
effectiveness in identifying cyberattacks and
improving defensive capabilities, the article highlights
significant challenges, including data scarcity, lack of
maritime-specific datasets, and vulnerability of AI
models to adversarial attacks. To overcome these
limitations, future research is recommended to focus
on large scale maritime datasets, adversarial
robustness, explainable AI (XAI), and integration with
advanced technologies like federated learning,
blockchain, and quantum cryptography.
Schubert et al. in [40] emphasise that AI can enhance
military command and control (C2) decision support,
particularly in time critical and complex scenarios.
Based on user-centred workshops with military
personnel, the study identifies key areas for AI
support, including building a common operational
picture, threat analysis, and evaluating alternative
courses of action. Techniques such as information
fusion (IF), natural language processing (NLP), and
learning based analysis can improve data
categorisation, anomaly detection, and planning
through simulation-assisted evaluation. While AI is
not intended to replace human commanders, it can
significantly increase analytical speed and depth and
offer operational advantages if effectively integrated
into C2 systems.
Francavilla & Armstrong in [18] analyse how AI is
transforming naval warfare by enhancing decision
making, surveillance, and combat systems through
real-time sensor processing and integration of manned
and unmanned platforms. They outline varying levels
of autonomy, from human-in-the-loop to human-out-
of-the-loop systems, stressing the need for human
oversight to ensure ethical and legal compliance. AI
applications, including autonomous vehicles and
geospatial systems such as SatShipAI, improve
maritime intelligence, anomaly detection, navigation,
logistics, and predictive maintenance. The authors
conclude that successful integration of AI in naval
operations depends on robust training, ethical
frameworks, international cooperation, and sustained
technological investment.
Absalon in [1] argues that integrating AI into
submarine platforms can enhance operational
effectiveness by improving prediction, detection, and
decision-making capabilities. AI-driven predictive
maintenance increases reliability and availability,
while machine learning improves sonar data analysis
and contact classification. AI-based tactical decision
aids and sensor fusion support faster, data-informed
manoeuvring and combat system responses, reducing
crew cognitive load. However, successful
implementation requires careful design of human-
machine interfaces, trust in autonomous systems, and
rigorous operational testing.
The deep learning approaches for classifying
warship images using a unique dataset from the UK
National Museum of the Royal Navy are investigated
in [3]. Several pre-trained convolutional neural
networks were evaluated, achieving high accuracy in
both coarse and fine classification tasks. The use of a
Grad-CAM enhanced model interpretability, enabling
archivists and curators to better understand
classification decisions. It could aid cultural heritage
preservation through automated cataloguing of
historical photographs and assist naval training by
providing visual recognition tools. The methodology
can be extended to broader maritime and defence
applications, such as identifying ship types in
intelligence, surveillance, and reconnaissance (ISR)
imagery, monitoring naval fleets, and supporting
maritime domain awareness.
AI is becoming a central component of modern
naval operations, enhancing capabilities across
surveillance, decision-making, combat, cybersecurity,
and training. It improves maritime domain awareness
through real-time anomaly detection, vessel
classification, and predictive analytics, while also
supporting tactical decision making and the
operational kill chain by enabling simulations,
reinforcement learning, and human-machine teaming.
AI integration into platforms such as submarines and
autonomous vessels enhances predictive maintenance,
sensor fusion, and operational efficiency, though
effective deployment requires trust, rigorous testing,
and well-designed human-machine interfaces. Beyond
operational uses, AI contributes to cybersecurity and
cultural heritage management, showing its versatility
across defence and analytical domains. Overall, AI can
significantly strengthen naval effectiveness, but its
success depends on investments in infrastructure,
training, ethical oversight, and seamless integration
across systems.
3.3 Use of AI in unmanned naval vehicles
The state-of-the-art path planning techniques for
unmanned surface vehicles (USVs) are reviewed in
[54]. It highlights how artificial intelligence methods
such as neural networks and reinforcement learning
are increasingly integrated to enhance navigation
autonomy and decision-making in complex marine
environments. AI-based approaches, including deep
Q-networks and hybrid machine learning models, help
USVs dynamically adjust paths in response to obstacles
and environmental conditions by learning optimal
strategies from sensor data. These intelligent
algorithms improve the ability to handle unknown or
changing scenarios compared with traditional search
and optimisation methods, enabling real-time obstacle
avoidance and more efficient trajectory generation. The
review identifies limitations in current AI-driven
methods, such as insufficient modelling of sea factors,
and suggests future research directions that
incorporate environmental dynamics into AI path
planners to further boost performance.
The development of autonomy in unmanned
surface vehicles with particular emphasis on intelligent
collision avoidance manoeuvres that aim to reduce
reliance on human intervention is examined in [8]. It
highlights that integrating AI techniques, such as
neural networks, fuzzy logic (FL), and other intelligent
control and optimisation methods, can enhance USV
perception, decision making, and compliance with the
International Regulations for Preventing Collisions at
179
Sea (COLREGs) by enabling automatic recognition and
response to dynamic obstacles. These AI-driven
approaches address challenges inherent in open
marine environments, including the classification of
static versus moving obstacles and the adaptation of
avoidance strategies in real time. It underscores the
need for further research to improve the robustness
and fail-safety of intelligent collision avoidance
systems as part of broader autonomy goals for USVs.
The overview [53] provides a comprehensive
presentation of developments and challenges in
autonomous berths of unmanned surface vehicles
(USVs), emphasising that intelligence and automation
are critical for reducing human intervention and
improving safety in complex marine environments. It
discusses how advanced sensing, multi-sensor fusion,
and intelligent decision-making systems, often
supported by artificial intelligence techniques, enable
USVs to perceive berthing scenarios, assess risk, and
plan collision-free approaches without human control.
The paper also highlights the importance of adaptive
autonomy levels and real-time environmental
modelling in handling uncertainties and dynamic
conditions during autonomous berthing operations.
Fig. 3 shows the neural network berthing control
process.
Figure 3. Neural network berthing control process (Source:
[53])
A novel application of deterministic artificial
intelligence (DAI) for autonomous control of remotely
operated ocean vehicles, showing how integrating
model-based AI with optimal learning enhances
vehicle guidance and control performance, is shown in
[36]. Unlike traditional stochastic or feedback-only
controllers, the proposed DAI approach embeds the
physical governing equations of the system into the
control law and combines this with learning to adapt
to dynamic changes, enabling highly precise heading
control. Simulation results for a Seabotix vLBV 300
remotely operated vehicle demonstrate that the DAI
controller achieves millidegree accuracy on initial
heading commands and significantly reduces error
over subsequent manoeuvres. The work suggests that
physics-aware AI control can substantially improve the
autonomous and remote operation of underwater
vehicles in challenging environments.
The concept of human-centred explainable artificial
intelligence (XAI) for marine autonomous surface
vehicles (ASVs), emphasising the importance of
interpretability, understandability, and trust in AI
systems for broader real-world deployment beyond
expert users, is introduced in [50]. It argues that as
ASVs become more common, AI models must not only
perform effectively but also be explainable and aligned
with user values to support safety and user confidence.
Drawing on examples from recent research, the
authors structure human-centred XAI through
cognitive processes such as analogy, visualisation, and
mental simulation to illustrate how AI decisions can be
made more transparent to diverse stakeholders. The
work highlights that improving explainability not only
aids developers but is essential for trust and interaction
among passengers, other vessels, and remote operators
in autonomous maritime contexts.
An extensive review of route planning and collision
avoidance algorithms developed for unmanned
surface vehicles, covering both simulated and real-
world applications from the early 2000s to the present,
is shown in [21]. It categorises these methods into
global and local planning approaches and highlights
trends in algorithm evolution, including the increasing
use of hybrid strategies that combine different
techniques to improve performance. The authors
emphasise the importance of validating algorithms in
real maritime environments as well as simulations to
fully assess reliability, adaptability, and operational
effectiveness. Key algorithmic paradigms such as
artificial potential fields (APF), reinforcement learning,
and fuzzy logic are identified as particularly promising
based on their evaluation in diverse scenarios.
A ship target detection method tailored for
unmanned surface vehicles (USVs) using the deep
learning object detector EfficientDet, aiming to
improve perception performance in complex maritime
environments where reflections, haze, and lighting
variations make detection challenging, is proposed
within [28]. The authors highlight that the method is
effective for identifying both static and dynamic ships
on the water surface, improving USV situational
awareness. They also discuss potential applications in
autonomous navigation and maritime threat
assessment, suggesting the approach provides a useful
benchmark for USV perception systems. Fig. 4 shows
the image of the ship targets from the USV.
Figure 4. The image of the ship targets from the USV. (Source:
[28])
A Universal Autonomous Control and
Management System (UACAMS) for a multipurpose
unmanned surface vessel that enables operation from
manual remote control to fully autonomous missions
are presented in [45]. AI is applied through layered
autonomy (LA) modules that perform sensor data
fusion, environment perception, adaptive path
planning, and collision avoidance. The system
continuously interprets inputs from radar, LiDAR,
sonar, and navigation sensors to support real time
autonomous decision making without human
180
intervention. AI-based control logic allows the vessel
to dynamically modify its behaviour in response to
environmental changes and mission objectives, which
was validated during real-world sea trials.
Artificial Intelligence is becoming central to
autonomy in unmanned naval vehicles, especially
through neural networks, reinforcement learning, and
hybrid control methods that improve navigation,
collision avoidance, and perception. These systems
enable adaptive decision-making, real-time obstacle
avoidance, and more efficient trajectory planning in
complex and dynamic marine environments.
However, many AI models remain limited by weak
integration of sea state dynamics, physical constraints,
and real-world uncertainties, reducing their reliability
outside simulations. Physics-informed and
deterministic AI offer improvements by combining
learning with physical system models, enhancing
precision, stability, and control performance. At the
same time, explainable and human-centred AI is
increasingly important to ensure trust, safety, and
regulatory acceptance. Despite clear progress, wider
deployment requires more robust environmental
modelling, explainability, and validation in real
operational conditions.
3.4 Discussion
Table 1 provides an overview of how artificial
intelligence is being integrated across different areas of
naval capability development, spanning design
activities, operational use, and unmanned systems.
Synthesising the reviewed literature according to the
thematic framework defined in the above sections
reveals distinct patterns in the maturity and role of
artificial intelligence across naval capability
development domains. In ship design and project
management, AI applications are predominantly
decision support tools that augment early-stage design
exploration, scheduling, and cost forecasting,
reflecting relatively high technical readiness and low
certification risk. In contrast, operational and
command and control applications prioritise real-time
data fusion, pattern recognition, and decision support
under time pressure, where performance gains are
significant, but trust, explainability, and human
oversight remain critical constraints. For unmanned
naval vehicles, the literature demonstrates the highest
levels of AI maturity, with autonomy, perception, and
path planning functions increasingly validated
through simulation and sea trials. Across all domains,
the review highlights a consistent transition from
isolated, task-specific AI solutions toward more
integrated, lifecycle-spanning systems, while
simultaneously underscoring that full operational
acceptance depends on human-AI teaming,
verification and certification frameworks, and
organisational adaptation rather than algorithmic
performance alone.
Table 1. AI applications in the naval capability area (Source: The authors)
Section
Application description
Supporting AI techniques
Ship Design
& Project
Management
Automation of procurement documentation, compliance checks,
and analysis of historical contracts
NLP, ML
Optimised activity planning and resource allocation across
complex projects
GAI, ML, AST
Prediction of project costs and delivery timelines
GAI, ML
Early identification of risks and requirement conflicts
ML, GAI
Estimation of person-hours and dynamic production scheduling
NN
Hydrodynamic and structural optimisation of hull forms
DNN, GAN, GA, RL, GA, DLNN,
VAE, SVM, GAI, ET, GNN, ANN
Automated metallographic analysis for material quality
assessment
NN, CBR
Support of predictive maintenance and anomaly detection
DT
Naval
Operations
Detection and classification of maritime objects and threats
CNN, DL, ML
Fusion of multisensory data for real time situational awareness
ML, DF
Detection of illegal activities such as smuggling or illegal fishing
DL, AD, RNN
Learning and evaluation of tactical options in combat scenarios
MARL, DDQN, PPO
Target detection, tracking, and engagement assessment
ML, CAI, XAI, DF, RNN
Detection and mitigation of cyber threats in maritime systems
IDS, AD, XAI, PTM
Course of action analysis and operational planning support
NLP, IF
Unmanned
Naval
Vehicles
Autonomous route planning in dynamic marine environments
NN, RL, DDQN
Real time, COLREGs compliant collision avoidance
FL, NN, PSO, APF, RL, DL
Collision free docking and manoeuvring in confined spaces
NN, DF
Precision heading and motion control of unmanned vehicles
DAI
Detection and classification of vessels in complex sea conditions
CNN, DL
Transparent AI decision making to support trust and safety
XAI
Adaptive mission execution using multisensory data
LA, ML
181
4 CHALLENGES IN AI INTEGRATION
Despite the increasing maturity of artificial intelligence
technologies and their growing adoption across naval
capability development, their integration into naval
platforms and organisations remains constrained by a
range of persistent challenges. These challenges cover
technical, organisational, legal, and ethical domains
and are particularly pronounced in naval
environments characterised by safety-critical
operations, long platform lifecycles, and complex
human-machine interactions. Unlike conventional
deterministic software, AI systems exhibit
nondeterministic, data-dependent, and adaptive
behaviour, which fundamentally challenges
established naval engineering, acquisition, and
certification paradigms [17], [22]. This section,
therefore, focuses on the key barriers that continue to
limit large scale and operationally unrestricted
deployment of AI in naval systems.
4.1 Verification, validation, and certification
Verification, validation, and certification remain the
most significant technical barriers to the operational
use of AI-enabled naval systems. Existing naval
certification frameworks are largely designed for
deterministic systems with predictable behaviour and
traceable requirements. In contrast, machine learning
based systems derive their functionality from training
data and statistical inference, making their behaviour
difficult to formally verify across all operational
conditions [22], [25]. This challenge is particularly
acute in safety-critical naval applications such as
navigation, collision avoidance, and platform control.
While AI-driven autonomy has demonstrated
promising performance in controlled and experimental
settings, certification for unrestricted maritime
operations remains limited due to the difficulty of
guaranteeing safe behaviour under uncertain
environmental conditions, sensor degradation, and
adversarial interference [30], [29]. Emerging mitigation
approaches, including physics-informed and hybrid
models, aim to constrain learning based behaviour
within known physical laws and engineering limits,
thereby improving consistency and certifiability [25].
Explainable artificial intelligence further contributes by
improving transparency and traceability, which are
essential for both certification and operational trust [3].
Nevertheless, comprehensive certification
methodologies for AI-enabled naval systems remain an
open research and regulatory challenge.
4.2 Integration with legacy platforms and digital
infrastructure
A second major challenge concerns the integration of
AI technologies with legacy naval platforms. Many
warships and submarines currently in service were
designed before the advent of data-centric and AI
enabled architectures and therefore lack the sensor
integration, computing capacity, data accessibility, and
open interfaces required for effective AI deployment
[11], [26]. Although newer platforms increasingly
adopt modular and open system architectures,
retrofitting legacy platforms often requires extensive
and costly upgrades to sensors, data buses, onboard
processing, and cybersecurity infrastructure. In
addition, AI systems increase cyber risk due to their
reliance on large data flows, continuous connectivity,
and frequent software updates. As a result, AI
integration must be accompanied by robust
cybersecurity measures and secure data pipelines to
mitigate risks such as data manipulation, poisoning,
and unauthorised access [5], [7].
Digital twin concepts illustrate both the
opportunities and the challenges of integration. While
digital twins can support predictive maintenance and
lifecycle optimisation, their effectiveness depends on
reliable, high-quality, and continuous data streams
from the physical platform, conditions that are often
difficult to achieve in legacy naval systems [17], [42].
4.3 Human-AI interaction and organisational adaptation
Human-AI interaction represents a critical
nontechnical challenge that directly affects operational
effectiveness. Naval operations remain fundamentally
human-centric, with responsibility for command
decisions and the use of force resting with human
operators. Poorly designed interfaces, insufficient
transparency, or over-reliance on automation can
degrade situational awareness and undermine trust in
AI systems [50]. Research on autonomous maritime
systems emphasises the importance of maintaining
appropriate levels of human oversight, particularly for
systems operating in human-in-the-loop or human-on-
the-loop modes [30], [8]. Human-centred design and
explainable AI are therefore essential to ensure that
operators understand system limitations, uncertainty,
and failure modes [3]. Beyond the technical interface,
organisational adaptation poses an equally significant
challenge. AI adoption often requires changes in
doctrine, training, and decision-making processes.
Defence acquisition studies indicate that without
institutional alignment and workforce upskilling, even
technically mature AI solutions may fail to deliver
operational benefits. Resistance to algorithmically
supported decision-making remains a notable barrier
in hierarchical naval organisations.
4.4 AI integration in naval acquisition and sustainment
AI integration challenges existing naval acquisition
and sustainment models, which are traditionally
optimised for hardware-centric systems with long
development and upgrade cycles. In contrast, AI
systems evolve continuously through data updates,
retraining, and algorithm refinement [43]. This
mismatch complicates requirements definition, testing,
and contractual arrangements. Although AI-based
tools offer potential benefits in procurement efficiency
and contract management, their adoption requires
regulatory adaptation, improved data governance, and
revised intellectual property frameworks [43].
Furthermore, lifecycle sustainment of AI systems
introduces new challenges, as model performance may
degrade over time due to changing operational
conditions and adversary behaviour. Continuous
monitoring, validation, and retraining are therefore
required capabilities that are not yet well integrated
into traditional naval maintenance practices [16], [42].
182
4.5 Ethical, legal, and strategic constraints
Ethical, legal, and strategic considerations impose
important constraints on AI integration in naval
systems. Increased autonomy raises questions of
accountability, escalation control, and compliance with
international humanitarian law, particularly in
contexts involving the potential use of force [4], [12].
While most navies retain human authority over lethal
decisions, even nonlethal AI applications may have
indirect lethal consequences, necessitating clear
governance and oversight mechanisms. National and
international AI strategies increasingly emphasise
responsible, transparent, and human-centred AI
development [41], [19], [25]. For naval forces, aligning
rapid technological innovation with ethical principles
and legal obligations remains a central challenge in
maintaining legitimacy and strategic stability.
5 USE CASE OF AI-BASED RANDOM FOREST
MODELING FOR BLOCK COEFFICIENT (CB)
PREDICTION IN NAVAL SHIP DESIGN
This section showcases an example of using a Random
Forest-based artificial intelligence model to predict the
block coefficient (Cb) of naval ships. Accurately
determining the block coefficient is essential during the
initial design phase, as it significantly impacts lightship
weight, total displacement characteristics,
hydrodynamic resistance, and propulsion power
needs. Thus, early and reliable predictions of Cb lead
to better-informed design choices and enhance overall
vessel performance. The model proposed is crafted
using a database of main ship dimensions that
encompasses a broad spectrum of naval vessel types,
including coastal patrol vessels (CPV), fast attack crafts
(FAC), offshore patrol vessels (OPV), corvettes (COR),
frigates (FRI), destroyers (DST), and aircraft carriers
(AC). This variety adds complexity but also increases
the model's applicability to practical scenarios.
Random Forest is an ensemble machine learning
method that builds numerous decision trees and
combines their results to create a more precise
prediction [14]. Random Forest regression is more
reliable than traditional linear models for complex,
high-dimensional problems, as it handles missing data,
captures nonlinear relationships effectively, and
delivers greater accuracy and prediction stability.
Additionally, it can process large datasets with
relatively few parameters while automatically
identifying the most important features to further
improve accuracy. As a form of ensemble learning, it
integrates the collective knowledge of various decision
trees, thereby enhancing both predictive accuracy and
stability [52]. High predicted accuracy is provided by
Random Forest Models, which are well known for their
resilience and adaptability in a variety of regression
and classification tasks. They offer comprehensible
metrics of feature relevance, are resistant to overfitting,
and need little data preprocessing. Specifically,
Random Forests measure each input variable's relative
contribution, making it possible to identify the features
that have the biggest impact on the model's
predictions. There are various crucial steps in the
Random Forest Model's training and prediction
process. Initially, bootstrap sampling is used to create
several subsets of the original dataset by sampling with
replacement, which encourages variation among the
trees and permits some observations to appear more
than once. Second, each decision tree is built by
evaluating a random subset of input features at each
node to find the best split. Third, the bootstrapped
datasets and randomised feature selection are used to
separately train many decision trees. Lastly, each tree
generates a unique forecast for unseen data, and these
predictions are aggregated to form the overall model
output, usually by averaging for regression tasks. In
this study, a Python-based Random Forest regression
model was developed to estimate the block coefficient
(Cb) of naval ships using ship length (L), breadth (B),
draft (T), and speed (v) as input features. The database
consists of 80 naval ships, and it is divided into training
and testing subsets using a standard 70/30 split for
model development and validation, respectively,
although alternative ratios may also be applied. A 20-
row excerpt from the database is presented in Table 2.
Model training was performed using the training
subset, followed by performance evaluation on the test
dataset, and the feature importance values were
subsequently computed. Reference Cb’s from the
dataset for all 80 ships, and predicted Cb for six ships
in Table3 from the Random Forest Model are shown in
Fig. 5.
A similar Random Forest Model applied to
merchant ship performance prediction [44] was
evaluated using the coefficient of determination (R²),
which measures the agreement between predicted and
reference block coefficient values. That study reported
a stable R² value of approximately 0.912, indicating
strong predictive capability compared to traditional
block coefficient estimation methods. In the present
naval ship model, the R² value is lower (0.8502),
primarily due to the relatively small dataset size and
the wide variety of ship classes, which increases
regression complexity. Model performance could be
improved by expanding the database with additional
naval ship examples. However, as this model is
intended primarily for illustrative purposes, it is
retained in its current form.
Table 2. Excerpt of 20 rows from the naval ship database
(Source: The authors)
Type
Class
State
Lpp [m]
B [m]
T [m]
V [kn]
Δ [t]
Fn
Reference Cb
CPV
Omiš
HRV
39.11
7.50
2.02
28.00
260.95
0.74
0.43
CPV
Mirna
HRV
29.67
6.62
1.76
28.00
138.60
0.84
0.39
FAC
Kralj
HRV
49.66
7.87
2.12
36.00
365.10
0.84
0.43
FAC
Kralj
HRV
50.22
7.87
2.12
37.00
390.00
0.86
0.45
FAC
Končar
HRV
41.66
7.75
1.89
40.00
263.50
1.02
0.42
FAC
Saar 4
IZR
52.78
7.01
2.40
34.00
450.00
0.77
0.49
FAC
Saar 4.5
IZR
56.15
7.01
2.80
33.00
498.00
0.72
0.44
OPV
Gawron
POL
86.63
12.42
3.60
29.50
2150.00
0.52
0.54
OPV
River
GB
82.36
11.96
3.80
25.00
2000.00
0.45
0.52
OPV
Holland
RNN
98.64
14.72
4.55
21.50
3750.00
0.36
0.55
OPV
Cassiopea
ITA
72.62
10.86
3.60
21.00
1500.00
0.40
0.52
OPV
Saar 62
IZR
56.42
6.99
2.70
32.00
500.00
0.70
0.46
COR
Braunschwei
K130
81.10
11.82
3.50
26.00
1840.00
0.47
0.54
COR
Bosphorus
ADA
90.50
12.82
3.95
30.00
2400.00
0.52
0.51
COR
Khamronsin
THA
56.40
7.30
2.50
25.00
630.00
0.55
0.60
COR
Fatahillah
IND
76.44
9.88
3.30
30.00
1450.00
0.56
0.57
FRI
FREMM
FRA/ITA
129.22
17.80
6.00
27.00
6000.00
0.39
0.42
FRI
Sachsen
GER
130.13
15.13
5.30
29.00
5800.00
0.42
0.54
FRI
Alvaro de Bazan
SPA
133.50
16.55
4.75
28.00
6250.00
0.40
0.58
FRI
Iver Huitfeldt
NOR
126.22
17.58
6.00
28.00
5290.00
0.41
0.39
183
Ship length (L, 0.3046) is the most significant
variable, followed by draft (T, 0.2917), width (B,
0.2672), and speed (v, 0.1365), according to feature
importance analysis. Regarding the factors that
determine hull fullness, these findings are consistent
with accepted naval architecture concepts. Table 3
reports the referenced and predicted block coefficient
values for six typical test ships. Lyashenko in [32] offers
a concise tutorial on how to use Random Forest Models
in Python and a comparison with similar machine
learning techniques.
Figure 5. Cb predicted values by Random Forest Model and
referenced Cb (Source: The authors)
Table 3. Cb predictions for 6 test ships (Source: The authors)
Type
Class
State
Lpp [m]
B [m]
T [m]
V [kn]
Δ [t]
Fn
Cb
RFM Cb
PV
Diana
DAN
39.13
7.22
2.20
25.00
280.00
0.66
0.44
0.45
FRI
Thetis
DAN
102.19
12.96
6.00
21.80
3500.00
0.35
0.43
0.47
FRI
Absalon
DAN
124.67
17.55
6.30
24.00
6600.00
0.35
0.47
0.47
FRI
La Fayette
FRA
113.75
13.86
4.10
25.00
3500.00
0.39
0.53
0.54
AC
J. F. Kennedy
USA
300.00
40.00
11.00
30.00
83981.00
0.28
0.62
0.58
DST
Visakhapatnam
IND
148.33
15.66
6.50
33.50
7400.00
0.45
0.48
0.51
6 CONCLUSIONS
This review has demonstrated that artificial
intelligence is becoming a foundational enabler across
the full spectrum of naval capability development. The
literature clearly indicates that AI technologies are no
longer confined to isolated decision support
applications but are increasingly embedded
throughout naval ship design, project management,
maintenance and sustainment, education and training,
simulations, operational employment, and maritime
unmanned systems. Data-driven and hybrid AI
approaches are complementing traditional physics-
based and rule-based naval engineering methods,
enabling improved performance prediction, design
space exploration, and lifecycle optimisation. The
analysis shows that AI has reached relatively high
maturity levels in domains such as unmanned vehicle
navigation, collision avoidance, perception, and path
planning, while applications in ship design
optimisation, digital twins, and lifecycle management
are progressing rapidly. Conversely, AI integration
into safety-critical combat systems, autonomous
decision making, and fleet-wide command and control
remain constrained by verification, validation,
certification, and ethical considerations. The reviewed
literature consistently emphasises that AI adoption in
naval contexts is shaped as much by organisational,
legal, and cultural factors as by technological
readiness. For naval planners and strategists, the
findings underline that AI integration is not merely a
technological upgrade but a systemic transformation of
how naval capabilities are conceived, developed, and
employed. AI-enabled tools can significantly enhance
decision-making speed, operational awareness, and
resource efficiency. However, these benefits can only
be realised if AI is incorporated early in the capability
development lifecycle and aligned with doctrinal,
organisational, and training frameworks. The review
highlights the importance of viewing AI as a force
multiplier that augments human expertise rather than
replacing it. Effective human-AI teaming,
explainability, and trust are critical for operational
acceptance, particularly in complex and contested
maritime environments. Strategically, navies that fail
to adapt acquisition processes, digital infrastructures,
and workforce competencies risk falling behind peers
who successfully leverage AI to accelerate innovation
and operational effectiveness. At the same time,
premature or poorly governed deployment of AI-
enabled systems may introduce new vulnerabilities,
escalation risks, and legal challenges.
Based on the synthesised literature, several
recommendations can be made. First, naval
organisations should pursue a phased and domain-
specific approach to AI integration, prioritising
applications with clear operational value and
manageable certification risks. Second, investment in
digital infrastructure, including data governance,
cybersecurity, and modular system architecture, is
essential to enable scalable AI deployment across
legacy and future platforms. Third, explainable and
physics-informed AI methods should be promoted to
support verification, validation, certification, and
operator trust, particularly in safety-critical
applications. From an organisational perspective,
navies should adapt acquisition and lifecycle
management frameworks to accommodate
continuously evolving AI systems, including
mechanisms for iterative testing, retraining, and
sustainment. Strengthening collaboration with
industry, research institutions, and small and medium-
sized enterprises is also critical to maintaining access to
innovation. Ethical, legal, and governance
considerations must be embedded into AI
development from the outset to ensure compliance
with international law and to preserve legitimacy in
naval operations. Despite significant progress,
substantial research gaps remain. Future work should
focus on developing standardised methodologies for
the verification and certification of AI-enabled naval
systems operating in open and adversarial maritime
environments. Additional research is needed on
human-AI interaction, particularly regarding
workload management, trust, and decision
accountability in multi-domain naval operations.
Moreover, empirical validation of AI applications
through large-scale trials and real-world operational
data remains limited and should be expanded. In the
end, interdisciplinary research addressing the
strategic, ethical, and legal implications of increasing
autonomy in naval warfare will be essential as AI
capabilities continue to mature. Effectively addressing
184
these challenges will determine whether AI becomes a
reliable and trusted component of future naval
capability development.
REFERENCES
[1] C. J. Absalon, “Integrate AI on submarines to predict,
detect, and decide,” Proceedings: U.S. Naval Institute,
2025, [Online]. Available:
https://www.usni.org/magazines/proceedings/2025/july/
integrate-ai-submarines-predict-detect-and-decide.
[Accessed: Nov. 14. 2025].
[2] Y. Ao, Y. Li, J. Gong, S. Li, “An artificial intelligence-aided
design (AIAD) of ship hull structures,” Journal of Ocean
Engineering and Science, 2021, -
doi.org/10.1016/j.joes.2021.11.003
[3] B. Baesens, A. Adams, R. Pacheco-Ruiz, A. S. Baesens, S.
Vanden Broucke, “Explainable deep learning to classify
Royal Navy ships,” Electronics, 2021, -
doi.org/10.1109/ACCESS.2023.3346061
[4] C. Barrera, I. Padrón Armas, F. Luis, O. Llinas, G. N.
Marichal Plasencia, “Trends and challenges in unmanned
surface vehicles: From survey to shipping,” TransNav,
2021, - doi.org/10.12716/1001.15.01.13
[5] M. Blănaru, “The need for an integrated model of smart
warfare,” Bulletin of Carol I National Defence University,
2024, - doi.org/10.53477/2284-9378-24-03
[6] V. M. Burenok, “Artificial intelligence in the military
confrontation of the future,” Voennaya Mysl, 2021, -
doi.org/10.21557/MTH.70170498
[7] S. H. Calfee, “Delivering advanced unmanned
autonomous systems and artificial intelligence for naval
superiority,” Center for Strategic and Budgetary
Assessments, 2025, [Online]. Available:
https://csbaonline.org/uploads/documents/CSBA8260_(
Navy_APO_Report)_web_3.pdf. [Accessed: Nov. 21,
2025].
[8] S. Campbell, W. Naeem, G. Irwin, “A review on
improving the autonomy of unmanned surface vehicles
through intelligent collision avoidance manoeuvres,”
Annual Reviews in Control, 2012, -
doi.org/10.1016/j.arcontrol.2012.09.008
[9] L. Cao, J. Shen, D. Bian, “Research on the application of
artificial intelligence in ship lifecycle assurance,” In
Proceedings of the 5th International Conference on
Automation Control, Algorithm and Intelligent Bionics,
2025, - doi.org/10.1145/3760269.3760290
[10] D. Cortés, K. León, “AI for comprehensive naval project
management: From render to launch,” Ship Science and
Technology, 2025, - doi.org/10.25043/19098642.276
[11] A. Cucinschi, “Evolution of warships in the digital age,”
The Bulletin of "Carol I" National Defence University,
2024, - doi.org/10.53477/2284-9378-24-47
[12] M. L. Cummings, “Artificial intelligence and the future
of warfare,” International Security Department and US
and the Americas Programme, 2025, [Online]. Available:
https://www.chathamhouse.org/sites/default/files/public
ations/research/2017-01-26-artificial-intelligence-future-
warfare-cummings-final.pdf. [Accessed: Nov. 24, 2025].
[13] M. Dhiman, “The role of artificial intelligence and the
navy. Centre for Security Studies,” JSIA, 25., 2025,
[Online]. Available: https://jgu.s3.ap-south-
1.amazonaws.com/jsia/Mehak+Dhiman+-
+Role+of+AI+and+the+Navy.pdf. [Accessed: Nov. 27,
2025].
[14] N. Donges, “Random forest: A complete guide for
machine learning. Built In,” 2025, [Online]. Available:
https://builtin.com/data-science/random-forest-
algorithm. [Accessed: Nov. 28. 2025].
[15] V. Emelianov, A. Zhilenkov, S. Chernyi, E. Zinchenko,
“Application of artificial intelligence technologies in
metallographic analysis for quality assessment in the
shipbuilding industry,” Heliyon, 2022, -
doi.org/10.1016/j.heliyon.2022.e10002
[16] P. Fernandez, “Designing submarines 4.0 with
CAD/CAM/CAE tools: The application of Industry 4.0 in
naval engineering,” In Proceedings of the International
Conference on Naval Architecture, 2020,
https://library.imarest.org/record/7680?v=pdf
[17] S. Ferreno-González, A. Munin‐Doce, M. M. González,
L. S. Caamaño, V. Diaz ‐ Casas, “ Digital twins for
warship systems: Technologies, applications, and
challenges, ” Journal of Marine Science and
Engineering, 2023, - doi.org/10.1002/9781119892199.ch7
[18] M. Francavilla, J. F. Armstrong, “The Future Of Maritime
Interactions In The Age of AI,” The defence horizon
journal,” 2025, https://tdhj.org/blog/post/maritime-
interactions-ai/
[19] Government of Australia, “Navy, RASAI strategy 2040:
Robotics and artificial intelligence strategy. National
strategy document,” 2025, [Online]. Available:
https://www.navy.gov.au/about-navy/strategic-
planning/robotics-autonomous-systems-artificial-
intelligence-strategy. [Accessed: Dec. 04, 2025].
[20] A. Grech La Rosa, P. Simpson, R. Zammit, “Exploring the
opportunities of generative artificial intelligence in
concept ship design,” In Proceedings of the 15th
International Marine Design Conference (IMDC 2024),
2024, - doi.org/10.59490/imdc.2024.752
[21] S. H. Hashali, S. Yang, X. X. Xiang, “Route Planning
Algorithms for Unmanned Surface Vehicles (USVs): A
Comprehensive Analysis,” J. Mar. Sci. Eng., 2024, -
doi.org/10.3390/jmse12030382
[22] M. C. Horowitz, L. Kahn, M. Mahoney, “The future of
military applications of artificial intelligence: A role for
confidence-building measures?” Orbis, 2020, -
doi.org/10.1016/j.orbis.2020.08.003
[23] N. M. Htein, P. Louvros, E. Stefanou, M. Aung, N. Hifi,
E. Boulougouris, “AI-based optimization techniques for
hydrodynamic and structural design in ships: A review,”
Journal of Marine Science and Engineering, 2025, -
doi.org/10.3390/jmse13091719
[24] B. Johnson, J. M. Green, G. Burns, T. Collier, R. Cornish,
K. Curley, A. Freeman, J. Spears, “Mapping Artificial
Intelligence to the Naval Tactical Kill Chain,” Naval
technical journal, 2023,
https://nps.edu/documents/10180/142489929/NEJ+Hybri
d+Force+Issue_Mapping+AI+to+The+Naval+Kill+Chain.
pdf
[25] E. B. Kania, “Chinese military innovation in artificial
intelligence,” Centre for new American security, 2025,
[Online].Available:
https://www.uscc.gov/sites/default/files/June%207%20H
earing_Panel%201_Elsa%20Kania_Chinese%20Military%
20Innovation%20in%20Artificial%20Intelligence.pdf.
[Accessed: Dec. 10, 2025].
[26] J. Kim, S. Han, H. Lee, B. Koo, M. Nam, K. Jang, J. Lee,
M. Chung, “Trend Research on Maritime Autonomous
Surface Ships (MASSs) Based on Shipboard Electronics:
Focusing on Text Mining and Network Analysis,”
Electronics, 2024, - doi.org/10.3390/electronics13101902
[27] J. K. Lee,K. J. Lee, J. S. Hong, W. Kim, E. Y. Kim, S. Y.
Choi, H. D. Kim, O. R. Yang, H. R. Choi, “DAS intelligent
scheduling systems for shipbuilding,” AI Magazine, 1995,
[28] R. Li, J. Wu, L. Cao, “Ship target detection of unmanned
surface vehicle base on EfficientDet,” Systems Science
and Control Engineering, 2022, -
doi.org/10.1080/21642583.2021.1990159
[29] C. Lin, C. Wu Yu, N. Wang, “Current challenges in
design, modelling, control, and applications of
unmanned surface and underwater vehicles,” Applied
Ocean Research, Journal of Control and Decisionv, 2024,
- doi.org/10.1080/23307706.2024.2398383
[30] Y. Liu, R. Bucknall, “Path planning algorithm for
unmanned surface vehicle formations in a practical
maritime environment,” Ocean Engineering, 2015, -
doi.org/10.1016/j.oceaneng.2015.01.008
185
[31] A. Ljulj, V. Slapničar, I. Grubišić, “Multi-attribute
concept design procedure of a generic naval vessel,”
Alexandria Engineering Journal, 2020, -
doi.org/10.1016/j.aej.2020.04.038
[32] V. Lyashenko, “How to use random forest for regression:
Notebook, examples and documentation,” 2025, [Online].
Available: https://cnvrg.io/random-forest-regression/.
[Accessed: Dec. 29, 2025].
[33] A. Maimun, S. C. Loon, J. Khairuddin, “Artificial
intelligence for ship design process improvement: A
conceptual paper,” Journal of Naval Architecture and
Marine Engineering, 2023, -
doi.org/10.3329/jname.v20i3.69695
[34] T. Miller, I. Durlik, E. Kostecka, S. Sokołowska, P.
Kozlovska, R. Zwolak, “Artificial intelligence in maritime
cybersecurity: A systematic review of AI-driven threat
detection and risk mitigation strategies,” Electronics,
2025, - doi.org/10.3390/electronics14091844
[35] O. Odjada, O. Obruche, “The application of artificial
intelligence in ship design: A systematic review,” Finnish
Thesis Repository, 2025, [Online]. Available:
https://www.theseus.fi/handle/10024/901309. [Accessed:
Dec. 03, 2025].
[36] S. Osler, T. Sands, “Controlling remotely operated
vehicles with deterministic artificial intelligence,” IEEE
Journal of Oceanic Engineering, 2022, -
doi.org/10.3390/app12062810
[37] P. Panneerselvam, K. G. Ramkumar, “The role of MSMEs
in shipbuilding and future naval technologies,” Maritime
Technology and Research, 2020,
[38] J. Ray, K. Atha, E. Francis, C. Dependahl, J. Mulvelon, D.
Alderman, L. A. Ragland-Luce, “China’s industrial and
military robotics development,” U.S.–China Economic
and Security Review Commission, 2025, [Online].
Available:
https://www.uscc.gov/sites/default/files/Research/DGI_
China%27s%20Industrial%20and%20Military%20Roboti
cs%20Development.pdf. [Accessed: Nov. 15, 2025].
[39] Royal Australian Navy. “RAS-AI strategy 2040: Warfare
innovation navy,” 2025, [Online]. Available:
https://www.navy.gov.au/sites/default/files/2024-
02/RASAI-Strategy-2040.pdf. [Accessed: Dec. 20, 2025].
[40] J. Schubert, J. Brynielsson, M. Nilsson, P. Svenmarck,
“Artificial intelligence for decision support in command
and control systems (FOI-S-5904-SE),” Swedish Defence
Research Agency, 2025, [Online]. Available:
https://foi.se/download/18.41db20b3168815026e010/1548
412090368/Artificial-intelligence-decision_FOI-S--5904--
SE.pdf. [Accessed: Oct. 26, 2025]
[41] A. Shevchenko, O. V. Bilokobylsky, M. Vakulenko, A. S.
Dovbysh, “Regarding the draft strategy development of
artificial intelligence in Ukraine,” Artificial Intelligence,
2022, - doi.org/10.15407/jai2022.01.008
[42] J. I. Silvera, J. L. Muñoz, J. M. Luquero, A. Cajade, M.
Bustelo, “Navantia’s digital twin implementation
perspective in military naval platform life cycle,”
Technical white paper, 2025, [Online]. Available:
https://publications.sto.nato.int/publications/STO%20Me
eting%20Proceedings/STO-MP-MSG-171/MP-MSG-171-
P1.pdf. [Accessed: Nov. 17. 2025].
[43] V. Skots, “Application of artificial intelligence in the
Department of Defense to enhance the contracting
process timeline,” Defense Acquisition Research Journal,
2021,
[44] V. Slapničar, A. Ljulj, “A preliminary design model of a
merchant ship by adjusting empirical methods,” In K. J.
Spyrou & N. Themelis (Eds.), Innovations in Sustainable
Maritime Technology—IMAM 2025, 2025, -
doi.org/10.1007/978-3-032-01566-2_7
[45] A. Stateczny, P. Burdziakowski, “Universal autonomous
control and management system for multipurpose
unmanned surface vehicles,” Polish Maritime Research,
2019, - doi.org/10.2478/pomr-2019-0004
[46] K. Talpur, R. Hasan, I. Gocer, S. Ahmad, Z. Bhuiyan, “AI
in Maritime Security: Applications, challenges, future
directions, and key data sources,” Information, 2025, -
doi.org/10.3390/info16080658
[47] O. A. Torres Salcedo, G. Sarmiento Meneses, J. D. Vélez
Restrepo, “Artificial intelligence and its integration in
naval operations,” OnBoard Knowledge Journal, 2025, -
doi.org/10.3390/OBK1010000
[48] UK Ministry of Defence. “Defence artificial intelligence
strategy,” 2025, [Online]. Available:
https://assets.publishing.service.gov.uk/media/62a7543e
e90e070396c9f7d2/Defence_Artificial_Intelligence_Strate
gy.pdf. [Accessed: Oct. 27, 2025].
[49] L. Vasankari, K. Saastamoinen, “Strategizing the
shallows: Leveraging multi-agent reinforcement learning
for enhanced tactical decision-making in littoral naval
warfare,” Artificial Intelligence Applications and
Innovations, 2024, - doi.org/10.1007/978-3-031-63215-0_10
[50] E. Veitch, A. O. Alses, “Human-centered explainable
artificial intelligence for marine autonomous surface
vehicles,” Journal of Marine Science and Engineering,
2021, - doi.org/10.3390/jmse9111227
[51] A. Zhukov, A. Saveliev, D. Zhurenkov, S. Khachaturyan,
I. Kartasan, “International experience of state initiatives
in the development of artificial intelligence as a dual-use
technology,” Web of Conferences, 2024, -
doi.org/10.1051/e3sconf/202453103010/
[52] D. Wei, “10 must-know models for ML beginners:
Random forest,” 2025, [Online]. Available:
https://medium.com/@weidagang/demystifying-
machine-learning-models-random-forest-f992dc50b427.
[Accessed: Oct. 27, 2025]
[53] G. Wu, D. Li, H. Ding, D. Shi, B. Han, “An overview of
developments and challenges for unmanned surface
vehicle autonomous berthing,” Complex & Intelligent
Systems, 2023, - doi.org/10.1007/s40747-023-01196-z
[54] B. Xing, M. Yu, Z. Liu, Y. Tan, Y. Sun, B. Li, “A Review
of Path Planning for Unmanned Surface Vehicles,”
Journal of Marine Science and Engineering, 2023, -
doi.org/10.3390/jmse11081556