Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту
The article presents the concept and architecture of digital twins (DT) in the tasks of autonomous swarm navigation for unmanned aerial vehicles (UAVs) controlled by artificial intelligence. Study demonstrated that the effective operation of a drone swarm under conditions of disrupted or absent comm...
Gespeichert in:
| Datum: | 2025 |
|---|---|
| Hauptverfasser: | , , , |
| Format: | Artikel |
| Sprache: | Englisch |
| Veröffentlicht: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2025
|
| Schlagworte: | |
| Online Zugang: | https://journal.iasa.kpi.ua/article/view/342991 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Institution
System research and information technologies| _version_ | 1866303049806381056 |
|---|---|
| author | Zgurovsky, Michael Pankratova, Nataliya Golinko, Igor Grishyn, Kostiantyn |
| author_facet | Zgurovsky, Michael Pankratova, Nataliya Golinko, Igor Grishyn, Kostiantyn |
| author_sort | Zgurovsky, Michael |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2025-11-09T00:01:30Z |
| description | The article presents the concept and architecture of digital twins (DT) in the tasks of autonomous swarm navigation for unmanned aerial vehicles (UAVs) controlled by artificial intelligence. Study demonstrated that the effective operation of a drone swarm under conditions of disrupted or absent communication with the ground center is enabled by the functional distribution of DT components between the ground center and onboard levels of AI agents. Mathematical models of ground center’s DT provide strategic modeling, training, mission simulation, and post-mission analysis, while onboard AI agents focus on local adaptation, diagnostics, environmental reconstruction, and cognitive behavior control. Special attention is paid to the interface module of the DT, which provides asynchronous interaction with the ground infrastructure. A functional division on the swarm-level, environment, mission, telemetry, and agent-level DTs is proposed. The effectiveness of the “Learn–Simulate–Deploy–Adapt” cycle for continuous improvement of swarm systems in the context of electronic warfare (EW) and dynamic operational environments was justified. The results were partially supported by the National Research Foundation of Ukraine, grant No. 2025.06/0022 “AI platform with cognitive services for coordinated autonomous navigation of distributed systems consisting of a large number of objects”. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2025.3.02 |
| first_indexed | 2025-11-09T02:11:02Z |
| format | Article |
| fulltext |
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn, 2025
Системні дослідження та інформаційні технології, 2025, № 3 19
UDC 004.8:681.518:629.7.052
DOI: 10.20535/SRIT.2308-8893.2025.3.02
DIGITAL TWINS IN AI-CONTROLLED NAVIGATION TASKS
FOR AUTONOMOUS UAV SWARM
M.Z. ZGUROVSKY, N.D. PANKRATOVA, I.M. GOLINKO, K.D. GRISHYN
Abstract. The article presents the concept and architecture of digital twins (DT) in
the tasks of autonomous swarm navigation for unmanned aerial vehicles (UAVs)
controlled by artificial intelligence. Study demonstrated that the effective operation
of a drone swarm under conditions of disrupted or absent communication with the
ground center is enabled by the functional distribution of DT components between
the ground center and onboard levels of AI agents. Mathematical models of ground
center’s DT provide strategic modeling, training, mission simulation, and post-
mission analysis, while onboard AI agents focus on local adaptation, diagnostics,
environmental reconstruction, and cognitive behavior control. Special attention is
paid to the interface module of the DT, which provides asynchronous interaction
with the ground infrastructure. A functional division on the swarm-level, environ-
ment, mission, telemetry, and agent-level DTs is proposed. The effectiveness of the
“Learn–Simulate–Deploy–Adapt” cycle for continuous improvement of swarm sys-
tems in the context of electronic warfare (EW) and dynamic operational environ-
ments was justified. The results were partially supported by the National Research
Foundation of Ukraine, grant No. 2025.06/0022 “AI platform with cognitive ser-
vices for coordinated autonomous navigation of distributed systems consisting of a
large number of objects”.
Keywords: digital twin, swarm intelligence, autonomous navigation, unmanned ae-
rial vehicles (UAVs), cognitive artificial intelligence platform, decentralized control,
simulation modeling, simultaneous localization and mapping (SLAM), behavior
trees (BT), electronic warfare.
INTRODUCTION
In today’s rapidly evolving world, the application of DT has gained significant
momentum, becoming a key success factor across various industries. Virtual rep-
licas of physical objects, systems, or processes open up opportunities for real-time
analysis, modeling, and optimization. The DT enables companies to reduce costs,
predict malfunctions, improve the management of production processes, and de-
velop new products with minimal risks. This technology becomes particularly
crucial as industry, healthcare, transportation, and urban planning undergo digital
transformation [1].
For instance, in [2], the author conducted a study on the implementation of
DT in manufacturing using reinforcement learning models. Compared to tradi-
tional management methods, production efficiency was improved by 18%, energy
consumption was reduced by 12%, and system downtime was decreased by 15%.
DT toolkit enables significant advancements in intelligent management
across various fields of activity. Although the concept of DT existed for over two
decades, scientific discussions regarding its precise definition are still ongoing.
A comprehensive definition of a DT is presented in [3], which incorporates
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 20
Grieves’ definition [4] and distinguishes DTs from digital models and digital
shadows based on the presence of information flows between the physical system
and its digital counterpart. If there is no automated data flow between the physical
system and its digital representation, such object is considered a digital model, an
example could be a CAD model of a technical system (e.g., an aircraft). A digital
shadow refers to a digital object (model) that receives data from the physical one.
Its primary function is to automatically track certain changes in the physical sys-
tem in order to represent its properties. If data flows from the physical system to
the digital object and vice versa, then digital object is considered a DT, as chang-
es in the digital representation affect the physical system.
These examples of interaction between a physical system and a digital copy
enable to distinguish at a qualitative level the categories of concepts of a digital
model, a digital shadow and a DT, but they do not specify details regarding the
DT components. One of the foundational studies on the standardization of DT is
the Industrial Internet Reference Architecture (IIRA), proposed by the Industrial
Internet Consortium (IIC) [5]. This document provides guidelines for the devel-
opment of systems, solutions, and applications that incorporate DTs in industrial
and infrastructure domains. This architecture contains general definitions for in-
terested parties, the order of system decomposition, design patterns and a list of
terms. The IIRA model defines at least four types of interested parties: business;
use; operation; implementation..Each area focuses on the implementation of the
corresponding functional model of the DT, structure, interfaces, internal compo-
nent interactions, as well as on the system of DT models interaction with physical
object’s external elements. According to the IIRA model, information about the
DT includes (but is not limited to) a combination of the following categories:
physical model and data; analytical model and data; archives of time variables;
transaction data; master data; visual models and calculations. Thus, the concept of
DT has a multifaceted architecture and, therefore, complex mathematical support
for implementation.
A promising area for the application of DT toolkit is UAV control. An addi-
tional challenge in this domain is the cognitive coordination of UAV swarms dur-
ing flight. The autonomous navigation of UAV swarms is based on the integration
of two key system components:
a ground center with module DT designed for UAV training, validation,
and control;
an onboard AI-platform for UAV with cognitive services.
The ground center functions as a strategic control center, where training,
testing, and validation of neural networks, used on board of the drones, are carried
out [6]–[8]. This center hosts an infrastructure for simulating combat missions in
a virtual environment using onboard platform of UAVs [3], [5]. This approach
ensures a high degree of realism, allowing to test the system’s behavior under
load, estimate mission losses and adapt swarm architecture to changing condi-
tions.
The onboard segment of the UAVs is responsible for executing the mission
(scenario) defined by the ground center, either with continuous data exchange
with the ground center or in full autonomy mode. Also, it enables the swarm to
independently navigate, make real-time decisions, avoid obstacles, stabilize flight,
and coordinate swarm members without the need for constant communication
with the ground center.
Digital twins in AI-controlled navigation tasks for autonomous UAV swarm
Системні дослідження та інформаційні технології, 2025, № 3 21
The interaction between the onboard and ground centers constitutes a con-
tinuous cycle of adaptation, learning, and improvement. During the pre-mission
phase, the ground center trains the models, simulates the mission execution, com-
piles UAV operational algorithms, and uploads updated algorithms to the onboard
systems. During the mission, the drones operate autonomously but send telemetry
data to the center whenever communication is available. The center monitors the
mission and, if necessary, sends corrective commands. After mission, the col-
lected data are analyzed, checked for anomalies, models are refined, and a new
training cycle is being initiated.
The evolution of the cognitive component of the UAV swarm AI system is
realized through iterative model training using feedback obtained after mission.
The models are based on a combination of reinforcement learning, local decision-
making via BT, and neural network-based anomaly detection. This architecture
enables the system to self-learn and improve strategies without compromising
autonomy. A distinctive feature of the AI-system is the integration of memory and
logging mechanisms that accumulate data from mission to mission, forming the
foundation for the swarm’s cognitive adaptation. Thus, the system acquires the
capability for cognitive evolution — learning from its own experience to enhance
efficiency and resilience in the dynamic challenges of the modern battlefield.
ARCHITECTURE AND INTERACTION LOGIC OF DT FOR SWARM-
ORIENTED AUTONOMOUS UAV NAVIGATION
The DT is deployed at the ground center, while a limited interface module is im-
plemented onboard the UAV, which provides (Fig. 1):
data and telemetry buffering;
local scenarios adaptation in case of communication loss.
Fig. 1 schematically illustrates the fundamental architecture of interaction
between the ground-based DT and the onboard interface module of the drone,
which are the key components of the autonomous cognitive artificial intelligence
platform for swarm control. The ground DT performs strategic-level functions —
modeling swarm behavior, training neural networks, simulating combat missions
in a virtual environment, and conducting in-depth post-mission analysis. This en-
vironment acts as a virtual test bed where adaptive strategies are developed and
verified before their implementation on real platforms.
Fig. 1. Systemic interaction between the ground DT and the UAV’s onboard subsystem
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 22
On the other hand, the onboard interface module, integrated into each drone,
ensures autonomous system operation under conditions of partial or complete
communication loss with the ground center. It implements local adaptation to en-
vironmental changes, performs internal diagnostics of the system’s technical state,
and reconstructs the surroundings using onboard sensors and odometry [9], or
simultaneous localization and mapping (SLAM) algorithms [10], [14]. This ap-
proach enables each agent to make real-time decisions independently, ensuring
decentralized, flexible, and fail-safe swarm behavior.
Asynchronous interaction between the ground-based DT and the onboard
module emphasizes the key concept of drone independence during flight. Com-
munication between the two levels is not continuous and may occur only at spe-
cific moments, when external environmental conditions allow it. This design en-
ables the maintenance of autonomous navigation even in hostile environments,
including when electronic warfare (EW) systems are active. At the same time,
data accumulated during the mission is buffered and transmitted to the ground
center when communication becomes available for analysis and further model
retraining, starting the next cycle of cognitive improvement. Therefore, the struc-
ture illustrated in Fig. 1 represents a dynamic and distributed system, where the
ground and onboard components operate in synergy to ensure adaptability, resil-
ience, and ability to evolve for swarm systems.
Within the structure of an autonomous AI-platform for UAV swarms, the
DT module performs an asynchronous yet strategically significant function. Its
primary role is not to ensure continuous communication during missions, but ra-
ther to prepare, analyze, and update behavioral strategies during periods of time,
when no combat missions are carried out. This approach aligns with the require-
ments of autonomous navigation in combat scenarios and under EW system activity,
when communication with the ground center may be unavailable or undesirable.
Before the start of a mission, DT in the ground center allows for the testing
of scenarios, adaptive strategies, and behavioral models [11], which are subse-
quently uploaded to each drone’s onboard system. During flight, the drones oper-
ate fully autonomously, relying solely on local sensors, the cognitive core, and
adaptive algorithms. However, if communication is available, they exchange data
with the ground center. All data about behavior, telemetry, and decisions made
are recorded in internal buffers for further analysis. Table 1 presents the formal-
ization of the components of the UAV interface module.
T a b l e 1 . Formalization of the UAV interface module’functions
Function name Description Call time Data exchange format
Telemetry
buffering
Collection and storage of data for
subsequent transmission
After each
control cycle JSON / ROS message
SLAM or
odometry
module
Construction of a local environ-
mental map or spatial orientation
using camera image analysis
Real-time Local database
Fail-safe
monitor
Analysis of internal
system parameters
Every minute
or upon event
Log file /
Signal system
Behavior
controller
Adaptive switching between
branches of the BT
Upon event /
As planned
Internal FSM (finite
state machine) state
Swarm state
synchronizer
Exchange of critical information
with neighboring drones
Optionally,
peer-to-peer DDS / RTPS
Digital twins in AI-controlled navigation tasks for autonomous UAV swarm
Системні дослідження та інформаційні технології, 2025, № 3 23
Information is transmitted to the ground center’s DT after mission
completion or at designated evacuation checkpoints. This enables in-depth
analysis, model retraining, and updating the knowledge that are utilized in
subsequent missions. In this way, DT ensures swarm evolution without interfering
with the autonomy of task execution. The UAV interface module is responsible
for data buffering, access to the latest strategies, partial scenario simulation in
fallback modes, and asynchronous updates whenever the situation allows it. Its
presence in the system architecture is essential, as it provides autonomous
interaction with the ground center’s DT and local support without violating the
decentralized principle of swarm control.
DT APPLICATION DIRECTIONS IN TASKS OF AUTONOMOUS UAV SWARM
NAVIGATION
Considering the specific requirements of developing a cognitive AI-platform for
decentralized control of UAV swarms under EW activity conditions, DTs applica-
tion seems justifiable at multiple stages of the system’s life cycle. Within our pro-
ject, the most relevant mathematical models for implementing DTs for the ground
center are as follows.
1. UAV swarm model (system level). The objective of this model is to sim-
ulate and verify swarm behavior of drones within a virtual environment, taking
into account dynamics, communication losses, external disturbances, and changes
in swarm lineup. This model enables:
test decentralized control strategies (including BT) prior to its deployment;
analyze the stability of UAV swarm interaction under various agent loss
scenarios;
debug DDS/RTPS-based communication [12] between agents;
train reconfiguration algorithms without risk to physical drones.
2. Individual drone model (agent level). For each type of UAV, a corre-
sponding model is created that includes aerodynamics, navigation sensors, deci-
sion-making modules, and an interface with the autopilot. Its use allows to:
precisely test software–hardware interaction;
simulate sensor degradation, Global Navigation Satellite System (GNSS)
disruptions, and the impact of EW effects;
predict potential failures and transitions to fail-safe modes;
adapt controller (e.g., PID or MPC) parameters to mission-specific condi-
tions.
3. Environment model. The creation of a virtual 3D environment, which in-
corporates models of obstacles, threats, magnetic anomalies, and signal loss zones
enables to:
generate scenarios for training and testing swarm adaptation capabilities [13];
verify functionality of local planners (e.g., SLAM, obstacle avoidance
system [14]);
develop maps for pre-flight mission simulation and risk analysis.
4. Model of mission carry out. This involves the computer-aided design
and simulation of specific scenarios (e.g., patrol, evacuation, object detection),
which allows to:
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 24
optimize the initial mission BT (BT definition) according to the context;
identify critical nodes and failure points, prepare fallback behavior
branches;
automatically evaluate the effectiveness based on key performance indi-
cators (KPIs).
5. Telemetry model. Simulation of real-time data exchange with the ground
center enables to:
verify telemetry quality;
configure WebUI and ROSBridge protocol;
detect potential delays, data losses, or transmission errors.
Table 2 presents a comparison of simulation environments for the ground
center’s DT.
T a b l e 2 . Comparison of simulation environments for ground center DT
№ Environment Advantages Disadvantages
1 Gazebo + ROS 2 ROS 2 support, realistic physics,
open-source
Higher configuration
complexity
2 Ignition Gazebo Enhanced graphics, DT support Relatively new, limited
plugin ecosystem
3
AirSim
(Microsoft)
Realistic aerodynamics, integration
with Unreal Engine/Unity
High system
requirements
4 Unity + BT.CPP Full flexibility,
BT visualization support
Requires custom
infrastructure
Thus, the deployment of DT is recommended in environments Gazebo +
ROS 2 [15], Ignition Gazebo, AirSim, or Unity/Unreal Engine based emulators
integrated with BehaviorTree.CPP (see Table 2). It is especially appropriate to
implement the “Learn–Simulate–Deploy–Adapt” cycle, which integrates simula-
tion-based learning with the gradual transfer of behavior logic to the real swarm.
In this way, DT models become a key component not only in the R&D phase, but
also in training, testing, certification, and operational support of the system during
mission.
TASK ALLOCATION BETWEEN GROUND CENTER DT AND UAV ONBOARD
SYSTEMS
The overall logic of task distribution between the ground infrastructure and the
onboard UAV systems is as follows:
the ground center is responsible for simulation, training, strategic plan-
ning, and post-mission analysis;
the UAV onboard system provides a secure wireless interface for com-
munication between UAVs, also is it responsible for autonomous diagnostics,
real-time adaptation and navigation of each drone independently of the ground
center.
This division ensures optimal utilization of computational resources, flexibil-
ity and resilience of the system under conditions of limited connectivity and dy-
namic operational environment.
Digital twins in AI-controlled navigation tasks for autonomous UAV swarm
Системні дослідження та інформаційні технології, 2025, № 3 25
Tasks of DT executed at the ground center
In modern multi-layered architectures of autonomous swarm systems, DT which
is implemented at the ground center, plays a pivotal role in providing effective
modeling, testing, mission planning, and adaptation of UAV swarms to complex
and dynamic environments. Its models offload a significant portion of computa-
tional load from onboard UAV platforms to the ground infrastructure, while pre-
serving strategic coordination, behavioral predictability, and operational flexibil-
ity of the swarm. At the system level, the UAV swarm model enables simulation
of global swarm behaviors in various scenarios, testing decentralized control algo-
rithms and robustness of DDS/RTPS protocols, which is critically important in
environments with intermittent communication or in case of individual agents’
failure. Not only does this allow to identify system’s potential vulnerabilities, but
also enhance the swarm’s resilience to catastrophic events.
The environment model allows to construct complex terrain representations
with natural and artificial obstacles, as well as electromagnetic anomalies, which
is a critical factor for planning operations in areas with active EW interference.
Model generates scenarios that ensure high realism in training AI agents and ef-
fective pre-deployment testing of autonomous navigation algorithms. The mission
area visualization provided by the DT serves as a foundation for tactical decision-
making by operators or command centers. At the mission level, the DT supports
simulation of strategic transitions between scenarios, BT design, and the defini-
tion of KPIs, enables mission evaluation not only in terms of task completion, but
also in terms of the achievement of qualitative objectives.
The telemetry model focuses on simulation and verification of communica-
tion interfaces, delays, and telemetry data losses, as well as post-mission analysis
of swarm and individual drones’ behavior. This is particularly important for opti-
mizing information exchange between agents and the control center, as well as for
developing a knowledge base for future missions. Equally important is the indi-
vidual drone model, operating within the simulator, as it enables detailed configu-
ration of behavior logic at the level of a single AI agent, testing responses to envi-
ronmental changes, training UAV operators, and improving the onboard AI
agents installed on drones. The summary of DT mathematical models’ basic func-
tions, implemented at the ground center, are given below.
1. UAV swarm model (system level):
modeling global swarm behavior under various scenarios;
testing decentralized control algorithms;
testing DDS/RTPS communication protocols;
analyzing the impact of communication losses and agent failures on
swarm integrity;
simulating faults and catastrophic events.
2. Environment model:
creating terrain, obstacle, and magnetic anomaly models;
generating mission scenarios under complex conditions (including EW);
preparing training data for preliminary situations modeling;
visualizing the mission area for tactical planning.
3. Model of mission carry out:
designing and testing the mission tree (Behavior Tree (BT) Definition);
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 26
defining strategic transitions and fallback scenarios;
defining mission success criteria (in the form of KPI);
analyzing probable trajectories and synthetic tasks.
4. Telemetry model:
testing WebUI / ROSBridge interfaces;
simulating delays and data loss during transmission;
Analyzing swarm and individual drone behavior logs (offline mode).
5. Individual drone model. (agent level):
configuring behavior logic at the level of a single agent;
training AI agents or real operators in simulated environments.
Therefore, the deployment of DT’s mathematical models within the ground
center serves not merely as a simulation tool but as a foundational component of
adaptive, safe, and strategically coordinated operation of swarm systems. Func-
tioning as a virtual proving ground [16], these models facilitate iterative testing
and refinement of algorithms, significantly reducing operational risks and re-
source expenditures in real world.
Tasks of interface module executed onboard UAV (in autonomous navigation
mode)
Within the architecture of an autonomous swarm system, each UAV is equipped
with interface module that plays a critical role in ensuring local adaptation, flight
control, and interaction between agents under conditions of partial or complete
loss of communication with the ground center. These modules are designed to
maintain the UAV’s operability as an autonomous, cognitively capable agent
within a localized segment of the overall system. The central element of the inter-
face module is the local AI agent, responsible for monitoring the UAV’s internal
technical parameters: power supply voltage, temperature conditions, and sensor
integrity. Based on this data, the system implements hardware degradation fore-
casting, generates alerts for transition into a protected (fail-safe) mode, and de-
fines threshold conditions for potential mission withdrawal. This approach allows
each UAV not only to detect critical deviations but also to autonomously assess
its operational readiness for further task execution.
Integral to this functionality is the local environmental reconstruction. Em-
bedded odometry or SLAM-algorithms allow each UAV to generate an up-to-date
local map, identify obstacles, hazardous areas, landscape alterations, and predict
potential collisions. This spatial representation serves as the basis for real-time
reactive route planning, which is essential for survival and successful task execu-
tion in dynamic and often hostile environments. Importantly, such an AI agent
enables the UAV not merely to respond to the current operational context but also
to anticipate its change, making its behavior closely to a smart device rather than
a conventionally algorithm-driven system.
Another critical component of autonomy is the module of mission tree eval-
uation and control. This subsystem manages local execution of behavioral
branches, monitors task completion success, and can adaptively switch between
operational modes in response to environmental changes or variations in the
UAV’s internal parameters. This eliminates the limitations of rigid, pre-
programmed scenarios and facilitates decision-making under uncertainty. Concur-
Digital twins in AI-controlled navigation tasks for autonomous UAV swarm
Системні дослідження та інформаційні технології, 2025, № 3 27
rently, each agent maintains an individual log of critical events and, whenever
possible, transmits it to other swarm members, establishing the foundation for the
system’s collective memory.
The system architecture also incorporates a synchronization agent — a com-
pact communication and analytical add-on responsible for maintaining a locally
consistent representation of the swarm’s operational state, data synchronization
among neighboring agents and, in cases of data loss or corruption, initiates local-
ized reconfiguration of behavioral strategies. Synchronization agent provides
swarm’s decentralized response to the loss of one or more UAVs or to data distor-
tion within specific system segments. Such a design enhances the swarm’s resil-
ience, self-recovery capacity, and mission accomplishment potential, even under
unforeseen disruptive influences. The primary onboard functions of the interface
module, operating in autonomous navigation mode, are summarized as follows.
1. Local AI agent:
continuous monitoring of the UAV’s internal state parameters (power
supply voltage, temperature conditions, and sensor integrity);
prognosis of hardware component degradation and initiation of fail-safe
operational modes;
determination of threshold conditions that necessitate mission cancelation;
self-assessment of operational eligibility for continuing mission execution.
2. Local environment reconstruction (SLAM monitoring):
local spatial map construction (SLAM, obstacles, hazardous zones);
prediction of potential collisions and implementation of reactive path
planning;
detection of environmental changes (such as emergence of new obstacles,
threats, etc.);
3. Behavior evaluation and transitions between mission tree branches (be-
havior monitoring):
mission tree execution and task monitoring;
adaptive switching between behavioral modes;
logging of critical events and, when possible, transmission of this infor-
mation to the swarm.
4. Embedded synchronization agent:
maintenance of a coherent local representation of the swarm’s operational
state;
exchange of situational data with neighboring UAVs;
localized reconfiguration of behavioral strategies in response to UAV loss
or system faults.
Thus, the AI agents embedded within the UAVs not only enhance the func-
tional capabilities of individual drones but also establish the foundations for their
subjectivity, self-reflection, adaptive interaction, and coordinated behavior within
the swarm collective. This transforms each UAV from a mere of rigidly prede-
fined algorithms executor to an active participant in a complex, flexible, and
evolving behavioral system, and is necessary condition for the transition from
strictly programmed to self-learning swarm architectures.
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 28
AI agents and DTs within the architecture of autonomous swarm systems
have both civilian and military applications, which significantly enhances the
flexibility and universality of their deployment [17], [18]. In the military domain,
such systems allow to organize autonomous combat patrols, convoy escort opera-
tions, and the evacuation of wounded personnel from active combat zones, mini-
mizing risks to human operators. In the civilian sector, their functional capabili-
ties can be repurposed for search-and-rescue missions under challenging
conditions (e.g., post-natural disaster scenarios), wildfire monitoring, and the in-
spection of critical infrastructure such as bridges, gas pipelines, and power trans-
mission lines. Such dual-use ensures the maximization of technological potential
in both peacetime and wartime.
A particular focus is ensuring the cyber resilience of AI agents and DTs, as
they operate within environments with potentially high risks of external interfer-
ence [19]. To address these challenges, the system architecture incorporates ro-
bust protective mechanisms, including end-to-end communication channels en-
cryption according to DDS/RTPS protocols, guaranteeing the confidentiality of
transmitted data. To protect against data tampering, spoofing, or cyberattacks,
data authenticity verification is implemented using digital signatures. Addition-
ally, threat detection algorithms based on AI are employed that work through real-
time identification of anomalies and atypical behaviors. In case of communication
loss or corruption, fallback modes are activated, enabling the system to maintain
functionality and complete its mission despite partial isolation of individual ele-
ments. Collectively, these measures establish a reliable foundation for deploying
DT in complex informational, technological, and combat environments.
EXAMPLE OF PRACTICAL APPLICATION OF A DT IN UAV SWARM
NAVIGATION
As a part of demonstration scenario of critical infrastructure patrol under commu-
nication jamming (EW activity), a computer simulation of the DT for mathemati-
cal models of swarm, the environment, and individual drones was conducted.
During the mission preparation phase, the ground-based DT modelled a 3D map
of the operational area, which included magnetic anomalies, physical obstacles,
and signal loss zones. This model was used to generate the route traverse scenar-
ios for swarm groups, that take into account the limited availability of GNSS sig-
nals.
After uploading the BT and waypoints into the onboard systems, the UAVs
were executing the mission autonomously. During the experiment, Inertial Meas-
urement Unit (IMU) sensor failure for one of the agents was purposely simulated.
The UAV’s AI agent detected the corresponding anomaly, initiated a fallback sta-
bilization mode, and excluded the affected agent from coordinated interaction,
notifying the other UAVs via the synchronization agent.
Quantitative mission parameters: during the simulation scenario, a virtual
swarm consisting of 3 UAVs was patrolling 100 × 100 m area, as illustrated in
Fig. 2. Within the operational zone, the following conditions were modeled:
3 obstacles (representing buildings or infrastructure objects);
one EW zone with a diameter of 40 m, centered at coordinates (60, 60) point;
5 route waypoints shared by all UAVs;
Digital twins in AI-controlled navigation tasks for autonomous UAV swarm
Системні дослідження та інформаційні технології, 2025, № 3 29
mission duration: 9 minutes;
average UAV velocity: 4.2 m/s;
maximum positioning error within the EW zone: up to 3.6 m;
telemetry transmission delay (simulated): up to 2.5 s;
IMU sensor failure detection time (drone 1): 0.8 s;
transition time to fail-safe mode: 1.1 s from the moment of anomaly detection;
communication packet loss rate (in EW zone): up to 18%.
Therefore, the UAV AI agent identified a degradation via sensor temperature
prediction, isolated drone 3 from the interaction zone, and broadcasted a status
update to the remaining drones. Subsequently, the DT in the ground center up-
dated the failure prediction model using the recorded logs. Upon mission comple-
tion, all telemetry buffers’ content was transmitted to the ground center, where the
DT performed trajectories visualization, calculated KPI, evaluated decision-
making effectiveness, and retrained the fault detection model. Obtained data were
utilized to update strategic adaptation modules for further missions.
Thus, this case exemplifies how DT can not only be a training and prepara-
tion tool, but also actively participate in the autonomous control process, enhanc-
ing swarm safety and adaptability in real time. The technical capabilities of the
DT were demonstrated, as well as DT role in ensuring fault tolerance, post-
mission learning, and deployment of self-adaptive swarm architectures, which is
critical for the development of next-generation dual-use AI systems.
2
1
3
4
1
2
3
4
Mission scenario with UAV digital twin
X coordinate
Y
c
oo
rd
in
at
e
Fig. 2. Scenarios of patrolling missions by UAV swarm with DT
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 30
CONCLUSION
1. The ground center’s DT is a critical element of the cognitive AI-platform
for an autonomous UAV swarm, providing a closed cycle of adaptive training,
simulation, deployment and improvement of the swarm’s behavior in the condi-
tions of a real combat environment and EW systems operation. Its functionality
allows to effectively combine strategic planning and local autonomy. The ground
center’s DT performs modeling, training and verification, while the onboard sys-
tem implements adaptation, self-control and reconfiguration in real time. The AI
agent interface module on the UAV board provides asynchronous mission sup-
port, autonomous data buffering, partial environment reconstruction and deter-
mines behavior strategies without dependence on stable communication.
2. The classification of mathematical models for implementing the ground
center’s DT was proposed — models of the swarm, individual agents, environ-
ment, mission, and telemetry — which provides comprehensive simulation cover-
age of all aspects of swarm navigation. This contributes to the reliability, fault
tolerance, and adaptability of the system. The ground center’s DT functions as a
virtual proving ground, where autonomous navigation algorithms, BT, and swarm
coordination mechanisms are tested, verified, and refined, and SLAM algorithm
parameters are configured. AI agents on board of each drone provide UAV auton-
omy, enabling it to independently assess its state, predict malfunctions, adapt be-
havior, and interact with other agents even under critical conditions.
3. The demonstration scenario of patrolling under EW systems activity con-
firmed the effectiveness of the DT in failure detection, anomaly adaptation, and
swarm coordination restoration, thereby proving its role in providing self-learning
and fault-tolerant architectures. The implementation of the DT-based “Learn–
Simulate–Deploy–Adapt” cycle is strategically important for transforming auton-
omous swarms into evolving, intelligent dual-use systems.
4. During mission execution, the ground control and training station (de-
pending on the presence and intensity of EW interference) can operate in several
modes: as a DT (when connection with the UAV swarm is fully available); as a
digital shadow (in the case of limited connection with swarm elements); or in
combat task simulation mode (if connection with the UAVs is completely un-
available). To enhance mission performance, UAV AI agents may be equipped
with a data relay function to support the communication with the ground center.
In this mode, one drones are assigned to relay communication, and the other focus
on mission execution, thereby increasing the overall efficiency of the system.
REFERENCES
1. F. Tao, H. Zhang, A. Liu, A.Y. Nee, “Digital twin in industry: State-of-the-art,”
IEEE Transactions on Industrial Informatics, vol. 15, no. 4, pp. 2405–2415, 2018.
doi: 10.1109/TII.2018.2873186
2. Lang Dai, “Intelligent manufacturing digital twin creation based on BIM and rein-
forcement learning,” Proc. SPIE 13691, Third International Conference on Ad-
vanced Materials and Equipment Manufacturing (AMEM 2024), 1369134, 8 July
2025. doi: https://doi.org/10.1117/12.3070521
3. W. Kritzinger, M. Karner, G. Traar, J. Henjes, W. Sihn, “Digital Twin in manufac-
turing: A categorical literature review and classification,” Ifac-PapersOnline, vol.
51, no. 11, pp. 1016–1022, 2018. doi: 10.1016/j.ifacol.2018.08.474
Digital twins in AI-controlled navigation tasks for autonomous UAV swarm
Системні дослідження та інформаційні технології, 2025, № 3 31
4. M.W. Grieves, “Product lifecycle management: the new paradigm for enterprises,”
International Journal of Product Development, vol. 2, no. 12, pp. 71–84, 2005. doi:
10.1504/IJPD.2005.006669
5. “The Industrial Internet Reference Architecture,” An Industry IoT Consortium Foun-
dational Document, 2022. Available: https://www.iiconsortium.org/wp-
content/uploads/sites/2/2022/11/IIRA-v1.10.pdf
6. E. Kayacan, M.A. Khanesar, O. Kaynak, Fuzzy Neural Networks for Real Time Con-
trol Applications. Butterworth-Heinemann, 2015.
7. S.K. Zhou, D. Rueckert, G. Fichtinger, Machine Learning for Medical Image Analy-
sis: Advances in Models and Methods. Academic Press, 2019.
8. Y. LeCun, Y. Bengio, G. Hinton, “Deep learning,” Nature, vol. 521, pp. 436–444,
2015. doi: 10.1038/nature14539
9. S.M. Vovk, V.V. Hnatushenko, M.V. Bondarenko, Image processing methods and
computer vision. Dnipropetrovsk: Lira, 2016, 148 p.
10. I. Nevlyudov, S. Novoselov, K. Sukhachov, “Method of simultaneous localization
and mapping for construction of 2.5D maps of the environment using ROS,” Innova-
tive Technologies and Scientific Solutions for Industries, no. 2 (24), рp. 145–160,
2023. doi: https://doi.org/10.30837/ITSSI.2023.24.145
11. Y. Bengio, A. Courville, P. Vincent, “Representation learning: A review and new
perspectives,” IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 35, no. 8, pp. 1798–1828, 2013. doi: 10.1109/TPAMI.2013.50
12. Eclipse Foundation, “DDS for eXtremely Resource-constrained Environments
(XRCE-DDS),” OMG Specification, 2021. [Online]. Available: https://www.omg.org/
spec/DDS-XRCE/
13. Z. Hou, I. Fantoni, “Distributed Leader-Follower Formation Control for Multiple
Quadrotors with Weighted Topology,” 10th System of Systems Engineering Confer-
ence (SoSE), 2015. doi: https://doi.org/10.1109/SYSOSE.2015.7151924
14. M. Norbelt, X. Luo, J. Sun, U. Claude, “UAV Localization in Urban Area Mobility
Environment Based on Monocular VSLAM with Deep Learning,” Drones, 9(3), 171,
2025. doi: https://doi.org/10.3390/drones9030171
15. M. Quigley et al., “ROS: an open-source Robot Operating System,” ICRA Workshop
on Open Source Software, vol. 3, no. 3.2, 2009.
16. E.H. Glaessgen, D. Stargel, “The digital twin paradigm for future NASA and US Air
Force vehicles,” in 53rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynam-
ics and Materials Conference, 2012.
17. J. Tang, G. Liu, Q. Pan, “A review on representative swarm intelligence algorithms
for solving optimization problems: Applications and trends,” IEEE/CAA Journal of
Automatica Sinica, vol. 8, no. 10, pp. 1627–1643, 2021. doi: https://doi.org/10.1109/
JAS.2021.1004129
18. T.I. Zohdi, “A machine-learning enabled digital-twin framework for tactical drone-
swarm design,” Computer Methods in Applied Mechanics and Engineering, vol. 442,
117999, 2025. doi: https://doi.org/10.1016/j.cma.2025.11799919
19. X. Wang et al., “A Survey on Security of UAV Swarm Networks: Attacks and
Countermeasures,” ACM Computing Surveys, vol. 57, no. 3, 2025. doi: https://
doi.org/10.1145/3703625
Received 08.08.2025
INFORMATION ON THE ARTICLE
Michael Z. Zgurovsky, ORCID: 0000-0001-5896-7466, Educational and Scientific Com-
plex “Institute for Applied System Analysis” of the National Technical University of
Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: zgu-
rovsm@hotmail.com
M.Z. Zgurovsky, N.D. Pankratova, I.M. Golinko, K.D. Grishyn
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 32
Nataliya D. Pankratova, ORCID: 0000-0002-6372-5813, Educational and Scientific
Complex “Institute for Applied System Analysis” of the National Technical University of
Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: na-
talidmp@gmail.com
Igor M. Golinko, ORCID: 0000-0002-7640-4760, Educational and Scientific Institute of
Nuclear and Thermal Energy of the National Technical University of Ukraine “Igor Sikor-
sky Kyiv Polytechnic Institute”, Ukraine, e-mail: golinko.igor@lll.kpi.ua
Kostiantyn D. Grishyn, ORCID: 0009-0006-5950-3739, Educational and Research
Institute for Applied System Analysis of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, student, e-mail: constantine1223h
@gmail.com
ЦИФРОВІ ДВІЙНИКИ В ЗАДАЧАХ АВТОНОМНОЇ РОЙОВОЇ НАВІГАЦІЇ
ДРОНІВ ПІД УПРАВЛІННЯМ ШТУЧНОГО ІНТЕЛЕКТУ / М.З. Згуровський,
Н.Д. Панкратова, І.М. Голінко, К.Д. Грішин
Анотація. Розглянуто концепцію та запропоновано архітектуру цифрових
двійників у задачах автономної ройової навігації безпілотних літальних апара-
тів (БПЛА), керованих штучним інтелектом. Показано, що ефективне функці-
онування рою дронів в умовах відсутності стабільного зв’язку з наземним
центром можливе завдяки розподілу функцій цифрового двійника наземної
станції і ШІ-агентами бортового рівня. Математичні моделі наземного ЦД за-
безпечують стратегічне моделювання, навчання, симуляцію місій і аналіз ре-
зультатів, тоді як бортові ШІ-агенти зосереджені на локальній адаптації, діаг-
ностиці, реконструкції середовища й когнітивному управлінні поведінкою
дронів. Особливу увагу приділено інтерфейсному модулю ШІ-агента БПЛА,
що забезпечує асинхронну взаємодію з наземною інфраструктурою. Запропо-
новано функціональний поділ математичних моделей ЦД на моделі рою, сере-
довища, місії, телеметрії та окремого ШІ-агента. Обґрунтовано доцільність ви-
користання циклу «Learn–Simulate–Deploy–Adapt» для безперервного
вдосконалення ройових систем в умовах дії РЕБ і динамічного бойового сере-
довища. Результати частково підтримано Національним фондом досліджень
України, грант № 2025.06/0022 «Платформа штучного інтелекту з когнітивни-
ми сервісами для скоординованої автономної навігації розподілених систем,
що складаються з великої кількості об’єктів».
Ключові слова: цифровий двійник, ройовий інтелект, автономна навігація,
безпілотні літальні апарати, платформа штучного інтелекту, децентралізоване
управління, імітаційне моделювання, метод локалізації та картографування
(SLAM), поведінкові дерева, РЕБ.
|
| id | journaliasakpiua-article-342991 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-11-09T02:11:02Z |
| publishDate | 2025 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/c1/1f128dcbfd87164cfb1b69a8269999c1.pdf |
| spelling | journaliasakpiua-article-3429912025-11-09T00:01:30Z Digital twins in AI-controlled navigation tasks for autonomous UAV swarm Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту Zgurovsky, Michael Pankratova, Nataliya Golinko, Igor Grishyn, Kostiantyn цифровий двійник ройовий інтелект автономна навігація безпілотні літальні апарати платформа штучного інтелекту децентралізоване управління імітаційне моделювання метод локалізації та картографування (SLAM) поведінкові дерева РЕБ digital twin swarm intelligence autonomous navigation unmanned aerial vehicles (UAVs) cognitive artificial intelligence platform decentralized control simulation modeling simultaneous localization and mapping (SLAM) behavior trees (BT) electronic warfare The article presents the concept and architecture of digital twins (DT) in the tasks of autonomous swarm navigation for unmanned aerial vehicles (UAVs) controlled by artificial intelligence. Study demonstrated that the effective operation of a drone swarm under conditions of disrupted or absent communication with the ground center is enabled by the functional distribution of DT components between the ground center and onboard levels of AI agents. Mathematical models of ground center’s DT provide strategic modeling, training, mission simulation, and post-mission analysis, while onboard AI agents focus on local adaptation, diagnostics, environmental reconstruction, and cognitive behavior control. Special attention is paid to the interface module of the DT, which provides asynchronous interaction with the ground infrastructure. A functional division on the swarm-level, environment, mission, telemetry, and agent-level DTs is proposed. The effectiveness of the “Learn–Simulate–Deploy–Adapt” cycle for continuous improvement of swarm systems in the context of electronic warfare (EW) and dynamic operational environments was justified. The results were partially supported by the National Research Foundation of Ukraine, grant No. 2025.06/0022 “AI platform with cognitive services for coordinated autonomous navigation of distributed systems consisting of a large number of objects”. Розглянуто концепцію та запропоновано архітектуру цифрових двійників у задачах автономної ройової навігації безпілотних літальних апаратів (БПЛА), керованих штучним інтелектом. Показано, що ефективне функціонування рою дронів в умовах відсутності стабільного зв’язку з наземним центром можливе завдяки розподілу функцій цифрового двійника наземної станції і ШІ-агентами бортового рівня. Математичні моделі наземного ЦД забезпечують стратегічне моделювання, навчання, симуляцію місій і аналіз результатів, тоді як бортові ШІ-агенти зосереджені на локальній адаптації, діагностиці, реконструкції середовища й когнітивному управлінні поведінкою дронів. Особливу увагу приділено інтерфейсному модулю ШІ-агента БПЛА, що забезпечує асинхронну взаємодію з наземною інфраструктурою. Запропоновано функціональний поділ математичних моделей ЦД на моделі рою, середовища, місії, телеметрії та окремого ШІ-агента. Обґрунтовано доцільність використання циклу "Learn–Simulate–Deploy–Adapt" для безперервного вдосконалення ройових систем в умовах дії РЕБ і динамічного бойового середовища. Результати частково підтримано Національним фондом досліджень України, грант № 2025.06/0022 "Платформа штучного інтелекту з когнітивними сервісами для скоординованої автономної навігації розподілених систем, що складаються з великої кількості об’єктів". The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2025-09-29 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/342991 10.20535/SRIT.2308-8893.2025.3.02 System research and information technologies; No. 3 (2025); 19-32 Системные исследования и информационные технологии; № 3 (2025); 19-32 Системні дослідження та інформаційні технології; № 3 (2025); 19-32 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/342991/330925 |
| spellingShingle | цифровий двійник ройовий інтелект автономна навігація безпілотні літальні апарати платформа штучного інтелекту децентралізоване управління імітаційне моделювання метод локалізації та картографування (SLAM) поведінкові дерева РЕБ Zgurovsky, Michael Pankratova, Nataliya Golinko, Igor Grishyn, Kostiantyn Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| title | Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| title_alt | Digital twins in AI-controlled navigation tasks for autonomous UAV swarm |
| title_full | Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| title_fullStr | Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| title_full_unstemmed | Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| title_short | Цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| title_sort | цифрові двійники в задачах автономної ройової навігації дронів під управлінням штучного інтелекту |
| topic | цифровий двійник ройовий інтелект автономна навігація безпілотні літальні апарати платформа штучного інтелекту децентралізоване управління імітаційне моделювання метод локалізації та картографування (SLAM) поведінкові дерева РЕБ |
| topic_facet | цифровий двійник ройовий інтелект автономна навігація безпілотні літальні апарати платформа штучного інтелекту децентралізоване управління імітаційне моделювання метод локалізації та картографування (SLAM) поведінкові дерева РЕБ digital twin swarm intelligence autonomous navigation unmanned aerial vehicles (UAVs) cognitive artificial intelligence platform decentralized control simulation modeling simultaneous localization and mapping (SLAM) behavior trees (BT) electronic warfare |
| url | https://journal.iasa.kpi.ua/article/view/342991 |
| work_keys_str_mv | AT zgurovskymichael digitaltwinsinaicontrollednavigationtasksforautonomousuavswarm AT pankratovanataliya digitaltwinsinaicontrollednavigationtasksforautonomousuavswarm AT golinkoigor digitaltwinsinaicontrollednavigationtasksforautonomousuavswarm AT grishynkostiantyn digitaltwinsinaicontrollednavigationtasksforautonomousuavswarm AT zgurovskymichael cifrovídvíjnikivzadačahavtonomnoírojovoínavígacíídronívpídupravlínnâmštučnogoíntelektu AT pankratovanataliya cifrovídvíjnikivzadačahavtonomnoírojovoínavígacíídronívpídupravlínnâmštučnogoíntelektu AT golinkoigor cifrovídvíjnikivzadačahavtonomnoírojovoínavígacíídronívpídupravlínnâmštučnogoíntelektu AT grishynkostiantyn cifrovídvíjnikivzadačahavtonomnoírojovoínavígacíídronívpídupravlínnâmštučnogoíntelektu |