Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем
This paper presents a concept for a cognitive AI platform that enables autonomous navigation of distributed multi-agent systems, exemplified by UAV swarms. The proposed architecture integrates a ground control center with cognitive services and a multi-layered onboard subsystem, supporting a continu...
Збережено в:
| Дата: | 2025 |
|---|---|
| Автори: | , , , , , , , |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2025
|
| Теми: | |
| Онлайн доступ: | https://journal.iasa.kpi.ua/article/view/342981 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Репозитарії
System research and information technologies| _version_ | 1866303044106321920 |
|---|---|
| author | Zgurovsky, Michael Kasyanov, Pavlo Pankratova, Nataliya Zaychenko, Yuriy Savchenko, Illia Shovkoplyas, Tetyana Paliichuk, Liliia Tytarenko, Andrii |
| author_facet | Zgurovsky, Michael Kasyanov, Pavlo Pankratova, Nataliya Zaychenko, Yuriy Savchenko, Illia Shovkoplyas, Tetyana Paliichuk, Liliia Tytarenko, Andrii |
| author_sort | Zgurovsky, Michael |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2025-11-09T00:01:30Z |
| description | This paper presents a concept for a cognitive AI platform that enables autonomous navigation of distributed multi-agent systems, exemplified by UAV swarms. The proposed architecture integrates a ground control center with cognitive services and a multi-layered onboard subsystem, supporting a continuous loop of learning, adaptation, execution, and behavioral model updates. Several core mission scenarios are introduced, such as reconnaissance, search and rescue, target neutralization, and deception, showcasing the swarm’s ability to operate autonomously and in a decentralized manner, even under adversarial conditions. An example of a search and rescue mission implementation plan using a cognitive platform that includes adaptive planning, SLAM navigation, swarm coordination, and deep object recognition is presented. 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.2308-8893.2025.3.01 |
| first_indexed | 2025-11-09T02:11:02Z |
| format | Article |
| fulltext |
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko, I.O. Savchenko, T.V. Shovkoplyas,
L.S. Paliichuk, A.M. Tytarenko, 2025
Системні дослідження та інформаційні технології, 2025, № 3 7
TIДC
ТЕОРЕТИЧНІ ТА ПРИКЛАДНІ ПРОБЛЕМИ І
МЕТОДИ СИСТЕМНОГО АНАЛІЗУ
UDC 004.896:629.735.33
DOI: 10.20535/SRIT.2308-8893.2025.3.01
COGNITIVE AI PLATFORM FOR AUTONOMOUS NAVIGATION
OF DISTRIBUTED MULTI-AGENT SYSTEMS
M.Z. ZGUROVSKY, P.O. KASYANOV, N.D. PANKRATOVA, Yu.P. ZAYCHENKO,
I.O. SAVCHENKO, T.V. SHOVKOPLYAS, L.S. PALIICHUK, A.M. TYTARENKO
Abstract. This paper presents a concept for a cognitive AI platform that enables
autonomous navigation of distributed multi-agent systems, exemplified by UAV
swarms. The proposed architecture integrates a ground control center with cognitive
services and a multi-layered onboard subsystem, supporting a continuous loop of
learning, adaptation, execution, and behavioral model updates. Several core mission
scenarios are introduced, such as reconnaissance, search and rescue, target neutrali-
zation, and deception, showcasing the swarm’s ability to operate autonomously and
in a decentralized manner, even under adversarial conditions. An example of a
search and rescue mission implementation plan using a cognitive platform that in-
cludes adaptive planning, SLAM navigation, swarm coordination, and deep object
recognition is presented. The results were partially supported by the National Re-
search 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”.
Keywords: artificial intelligence, UAV swarm, autonomous navigation, cognitive
platform, multi-agent systems, behavior trees, digital twin, SLAM.
INTRODUCTION
In modern conditions of increasingly complex combat environment, active elec-
tronic warfare, and loss of reliable satellite connection network, a critical need
arises for creating autonomous, decentralized control framework for distributed
systems, particularly swarms of unmanned aerial vehicles (UAV). In this context
the development of a cognitive AI platform, capable of guaranteeing the coordi-
nated navigation of a multitude of agents prohibited from interaction with a cen-
tralized control point or external infrastructure, becomes especially important [1–3].
This kind of environment requires not only sufficient autonomy level of individ-
ual agents (drones), but also a wholesome approach to the organization of their
collective behavior implemented through cognitive self-learning, self-organization,
adaptation algorithms, and resilient inter-agent information exchange. The theoretical
and methodological basis for constructing this kind of platform was described in
[4–10], in particular the impossibility of full consistency of agents: swarm agents
cannot have a fully coordinated movement direction on spherical surfaces (as well
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko …
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3
8
as on large single-connected compact manifold surfaces without edges, including
geoids), which compromises at least ant colony algorithms, requiring the selection
of special points as regrouping zones [5, Theorem 1].
The AI platform for autonomous navigation of distributed multi-agent sys-
tems is viewed as an integral architecture that combines two closely intercon-
nected components: the on-board component functioning directly at each of the
autonomous agents, particularly the UAV, and the ground control center that pro-
vides learning, simulation, validation and strategic system control. Both compo-
nents are functionally and logically interconnected, and together they form a cog-
nitive AI platform in a broad sense – as an intellectual, self-learning architecture,
capable of adaptation to the changes in environment, and self-improvement on the
basis of accumulated experience.
The on-board component of the AI platform provides the completely auton-
omous functioning of its agents. It implements the capability for independent nav-
igation without the GPS (Global Positioning System) satellite signals, making
decisions in real time, decentralized swarm coordination, and adaptation in case
of losing individual agents, or changes in the environment. Its functioning is
based on the on-board AI modules, sensor systems, stygmergy algorithms, decen-
tralized planning, reinforcement learning methods, self-learning and self-
organization, SLAM (Simultaneous Localization and Mapping) methods, and
other modern approaches [11–13]. This component in particular implements the
cognitive behavior during missions: each drone is able to orient itself, perform the
assigned tasks, and interact with other swarm agents without centralized control.
The ground control center performs the role of the strategic brain center of
the system. It provides both primary, and cyclical training of the neural networks,
modeling mission scenarios in the simulation environment using digital twins
[14–17], testing and validation of the models, as well as the generation of the be-
havioral politics for on-board implementation. The ground center aggregates in-
formation from OSINT/ESINT sources, adapts the models to the operational con-
text using analytics, supports visualization, monitoring and strategic correction.
Through secure human-machine interface the operator obtains access to pa-
rametrization of missions, system state management, and updates to the AI mod-
ules software.
The interaction between the on-board and ground systems is organized as a
closed cognitive loop. In the pre-missionary phase, the ground control center im-
plements the training of models, mission modeling; creates the digital twins for
drones, and uploads the updated algorithms to the on-board systems. This process
involves analytical modules that aggregate OSINT (Open-Source Intelligence) for
adaptation to the current context. During missions, the drones operate autono-
mously, performing swarm coordination, and in case the secure connection is
available, transmit telemetry to the center which conducts monitoring and pro-
vides corrections if necessary. After the mission, the collected data is analyzed,
log files are checked for anomalies, the models are tweaked, and the new cycle of
training is started. Thus, the system is capable of continuous cognitive evolution –
it learns on its own experience, gradually increasing the efficiency and resilience
to new challenges of modern combat environment.
The cognitive AI platform is the only intellectual architecture system that in-
cludes ground and on-board components that jointly form the adaptive and viable
complex for coordinated autonomous navigation of a UAV swarm. This complex
functions within a continuous loop of adaptation and improvement, encompassing
Cognitive AI platform for autonomous navigation of distributed multi-agent systems
Системні дослідження та інформаційні технології, 2025, № 3
9
pre-mission preparation, autonomous task completion, post-mission analysis, and
further additional training. This loop implements the concept of a cognitive core
as a system capable of forming, updating and generalizing knowledge based on its
own experience, react to the variable conditions, support collective behavior of
agents, and retain efficiency in a complex, dynamic, and hostile environment.
The purpose of this research is to create architecture and principles of the
system operation where each UAV behaves as an autonomous cognitive agent,
capable of navigating without GPS, make decisions based on the local informa-
tion, exchange signals with its neighbors using stygmergy or a mesh network,
while acting within a single coordinated environment (the swarm). The construc-
tion of a new generation cognitive AI platform that combines adaptivity, resil-
ience and scalability, is envisioned, enabling the UAV swarm to operate inde-
pendently of external control, and efficiently complete the assigned tasks
(missions) even under critical circumstances. This research is aimed at imple-
menting the swarm intelligence in defense and rescue technologies, and forms the
theoretical and engineering base for the next generation of double purpose
autonomous systems.
THE GROUND CONTROL CENTER FOR THE AI PLATFORM WITH
COGNITIVE SERVICES
The ground control center for neural network training is a critical architecture el-
ement of the general AI platform for cognitive control of the autonomous drone
swarm. It performs the functions of development, testing, adaptation, security
check, and preparation of the behavior strategies and cognitive models that will be
uploaded to each of the drones before the actual mission assignment. The struc-
ture of this center is modular, logically decentralized, but centralized by computa-
tional power. It includes the following main functional blocks (Fig. 1):
Model training module. This block is responsible for the primary and recur-
rent training of the neural networks that will be applied in drone systems. The
technologies involved include Reinforcement Learning models, self-learning
models, perception models for detection and tracking of objects, as well as graph
GROUND CONTROL CENTER
OF THE AI PLATFORM
MODEL TRAINING
MODULE
SIMULATION
LAB WITH
DIGITAL TWIN
SWARM
CONTROL AND
MISSION
PREPARATION
MODULE
SECURITY AND
VALIDATION
MODULE
ANALYTICAL
BLOCK
MODULE
Fig. 1. The architecture of the ground center of the AI platform
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko …
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3
10
neural networks (GNN) for optimization of behavior in swarm configurations.
The training is performed both on the historical data, and the data obtained during
previous missions.
Simulation lab with digital twin. The digital twin of the ground center is a
critical element of the general AI platform architecture that allows to test the neu-
ral network behavior in complex and variable scenarios. Here both the standard
situations are simulated, and the stress scenarios, including the loss of the swarm
elements, navigation under interference, electronic warfare conditions. This stage
provides adaptivity and resilience of the trained behavior even before the real op-
eration.
Analytical block. This module conducts the analysis of open-source data
(OSINT). Analytical insight regarding the potential risks, typical tactics of the
enemy, or features of the mission territory can be promptly integrated in the proc-
ess of preparation for the real mission, increasing the relevance of the drone be-
havior. This may include, in particular, the location of the notable objects, rele-
vant mission details, maps etc.
Security and validation module. Following the primary training, all models
are tested to ensure they meet resilience, security, and durability requirements. In
particular, this check includes a model’s capability of detecting anomalies, resto-
ration after errors, resilience to attacks at the data level, connection channels, and
model integrity. Validation is the obligatory stage before the mission implementation.
Swarm control and mission preparation module. This block represents the
control interface that aggregates the results from all other blocks and prepares the
behavior model for uploading to the drones; forms the detailed missions; distrib-
utes the tasks among agents; plans the route networks; defines the zonal priorities.
This module is used to upload the prepared cognitive software to the drones be-
fore their assignment to the real or test mission. The center also performs the
functions of the swarm state monitoring, interactive control, and strategy adapta-
tion in real time.
As the Fig. 1 shows, the interaction between the sub-systems of the ground
control center is organized as a closed cognitive loop that guarantees the whole-
some functioning of the drone swarm control system. In this loop the models
formed in the training module are automatically transferred to the simulation lab,
where they are subject to testing under the circumstances as close as possible to
the real environment. The simulation results are analyzed by the validation mod-
ule that makes the decision regarding the fitness of the models for combat use.
The OSINT module works in parallel, generating the contextual scenarios using
open-source intelligence data; these scenarios are integrated into the training pro-
cesses, increasing adaptivity and relevance of the trained models.
When the neural networks complete all verification stages, the swarm con-
trol center uploads them on-board of the drones, initiating missions, and perform-
ing their accompaniment, monitoring and correction in real time. Thus, the
ground center acts as a “cognitive foundry” of the system – the environment
where the artificial intelligence is not only created but also evolves under the in-
fluence of the new data, combat experience, and strategical analysis. Here the in-
tellectual potential of the swarm is formed, allowing the drones to act as intelli-
gent autonomous agents with high adaptation abilities, mutual understanding, and
collective behavior in the complex and hostile environment.
Cognitive AI platform for autonomous navigation of distributed multi-agent systems
Системні дослідження та інформаційні технології, 2025, № 3
11
BASIC SCENARIOS (MISSIONS) FOR THE AUTONOMOUS NAVIGATION OF
THE DISTRIBUTED MULTI-AGENT SYSTEMS
In modern combat and rescue conditions the scenarios for the drone swarm con-
stitute the basis for the cognitive behavior of the autonomous agents that function
within the integral AI platform. These scenarios are not just simple instructions –
they represent the structured, multi-component algorithmic descriptions, prelimi-
narily modeled in the ground control center. Due to the involvement of digital
simulation environments (such as Gazebo or AirSim), analytical modeling, mis-
sion planning tools (such as QGroundControl), and machine learning methods,
the scenarios achieve high adaptivity to the complex and dynamic environment.
After modeling they are saved in JSON, XML, TensorRT, ONNX [18] etc. for-
mats, and are uploaded to the computational blocks of each drone through a se-
cure channel before the mission starts.
The content of these scenarios includes several critically important func-
tional blocks: mission planning, autonomous navigation, recognition, decision
making, and swarm coordination. The planner contains the vectorized task de-
scription, temporal parameters, action sequences, and defined objectives. The
autonomous navigation modules provide route planning in real time using SLAM,
localization and obstacle avoidance algorithms. The recognition components are
responsible for the processing of sensor data from cameras, thermal imagers and
radars, allowing them to detect objectives, obstacles and threats. The decision
making is implemented through cognitive models capable of situational analysis,
and producing reactions based on environment assessment. Finally, the swarm
coordination provides the dynamic distribution of roles between agents, syncing
of the trajectories, and coordinated behavior within the swarm [19].
The unique nature of these scenarios lies in their ability to activate the on-
board drone cognitive modules that provide adaptive behavior even in case of the
absent connection to the control center, external interference, or the shifting envi-
ronment. In other words, the drones not only implement the previously assigned
actions, but also learn from the current situation, predict risks, and react collec-
tively. This is made possible by the integration of reinforcement learning meth-
ods, graph neural networks, and large language models that enable flexible, situ-
ational cognition at the swarm level [20].
Let us provide a list of basic scenarios:
Scenario 1: Enemy territory reconnaissance. The drone swarm distributes
the reconnaissance area (e.g., 10×10 km), with each sub-area assigned to an indi-
vidual drone. The results are obtained as a shared locality map. The scenario is
performed by 6–12 drones that cover up to 100 km2 in 15–40 minutes.
Scenario 2: Targeted strike with autonomous guidance. Several drones at-
tack the target from different directions, overcoming air defenses by dispersed
planning. Up to 7 drones attack the target’s coordinates in 3–10 minutes after its
detection.
Scenario 3: Communication relay. The drone swarm creates a temporary
mesh network, providing connection under electronic warfare. For example, 5–15
drones create a 5–10 km long linear network, providing communication for 20–60
minutes.
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko …
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3
12
Scenario 4: Search and rescue. The swarm autonomously scouts the de-
struction zone, detecting people and animals by performing scanning with distri-
bution of routes. Up to 20 drones are used, with coverage area 10–40 km2 for one
hour.
Scenario 5: “Death ring” swarm attack. The drones fly round the target
from all directions, forming a ring, and strike it simultaneously. In this scenario
5–10 drones are used, with 100–500 m attack radius during 5–15 minutes.
Scenario 6: Scattering false targets/misinformation. The swarm scatters imi-
tation objects to mislead the enemy or mask the actual swarm’s goals, by per-
forming a coordinated placement of false targets (vehicle imitations), or modeling
the behavior of a real vehicle column. During 10–30 minutes 5–10 drones place
signal imitators along the 20 km route, using GPS and waypoint navigation (a
drone moves from one waypoint to another in a predetermined sequence).
The compiled scenario (mission) parameters are given in Table 1.
T a b l e 1 . The compiled scenario (mission) parameters for UAV swarms
Scenario Drone
quantity
Surface/length
of coverage Duration Communication/
protocol
Scenario 1.
Enemy territory re-
connaissance
6–12 Up to 100 km² 15–40 minutes
DDS or ROS
topics + sensors
(LIDAR/camera)
Scenario 2.
Targeted strike with
autonomous guidance
Up to 7
Depends
on target
(up to 10 km)
3–10 minutes MAVLink/mesh
connection
Scenario 3.
Communication relay 5–15 5–10 km 20–60 minutes DDS+RTPS with
real-time QoS profile
Scenario 4.
Search and rescue Up to 20 10–40 km² Up to 1 hour ROS topics +
thermal imager
Scenario 5.
Death ring 5–10 Attack radius up
to 500 m 5–15 minutes ROS2 + DDS
Scenario 6.
Scattering
false targets
5–10
depending
on the route
Up to 20 km
route 10–30 minutes MAVLink with
waypoint navigation
So, the scenarios for the drone swarm become the key element for the cogni-
tive AI platform, combining high precision planning, realistic simulation, analyti-
cal adaptation, and self-learning. Their exploitation not only increases mission
efficiency, but also provides resilience to the uncertainty factors, which is critical
in the environment where each second and each decision is significant.
ON-BOARD COMPONENT OF THE AI PLATFORM WITH COGNITIVE
SERVICES
The on-board component is a key functional environment where the autonomous
intelligence of each drone in the swarm is implemented. This is the place where
the integration of cognitive models, sensory perception, swarm interaction, flight
control, and adaptive decision making in real time is performed. The architecture
of this component (Fig. 2) is multi-layered and includes a number of modules that
jointly ensure the independence of the drone from external control, its self-
learning capacity, and flexible reaction to a dynamic environment.
Cognitive AI platform for autonomous navigation of distributed multi-agent systems
Системні дослідження та інформаційні технології, 2025, № 3
13
Let us consider each module of the architecture presented in Fig. 2, review-
ing its functions, and their mutual interaction.
The center of the architecture is comprised of the cognitive core that acts as
a drone’s “brain”, and is responsible for situational analysis, adaptation and deci-
sion making. Its fundament is the Swarm coordination module, implemented us-
ing the hybrid approach where the swarm AI methods are applied using a hybrid
scheme: the Behavioral trees (BTs), and Global swarm optimization (Global Best
PSO) that can reconfigure in real time depending on the changes in the environ-
ment [21–22]. This allows each agent to form the sequence of actions, independ-
ently react to the loss of communication, emergence of new threats, or changes in
objectives.
Combined with the Adaptive behavior planning module that analyzes risks,
priorities and current context, the system acquires the ability for conscious deci-
sion making even having incomplete information. It performs the incremental on-
board learning (given the appropriate resources), bufferization of the field data,
and the backhaul retraining loop implementation – the transmission of the col-
lected data to the ground control center, with the subsequent updates in the mod-
els. This mechanism forms the basis of the system evolution, as it allows to take
previous experience into account in the future missions. This approach allows to
coordinate local trajectories, synchronize agent sub-groups, and sustain the over-
all mission goals at the lower autonomy level.
To enable these cognitive processes, the drone requires a constant flow of in-
formation about the environment. This task is achieved by the Machine vision and
data processing module that aggregates the data from cameras, ultra-sonic sensors
etc., forming the local space maps using SLAM algorithms [23–24]. An important
feature of this layer is its capability for semantic classification of the objects (e.g.,
enemy units, civilians, allied units), and detection of the situational patterns that
allow to construct not only a spatial, but also a behavioral model of the environ-
ment.
For coordinated interaction among the swarm elements, the platform con-
tains the communication module, based on low latency DDS/RTPS protocols. It
provides the interchange of statuses between agents by behavioral subtree broad-
cast, and allows to maintain the swarm coordination without the centralized con-
trol [25–27]. Even in case of losses or disruption in network channels the module
Cognitive core
Swarm
optimization
Adaptive
behavior
planning
Machine vision
and data proc-
essing module
Security, threat
and anomaly
detection
module
DDS/RTPS
communication
module, swarm
exchange
Navigation
and flight
module
Digital twin
(modeling,
simulation)
Energy
management
and hardware
security module
Fig. 2. The structural scheme of the on-board component of the AI platform
with cognitive services
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko …
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3
14
remains operational due to the QoS control network that allows to duplicate criti-
cal data, and adapt priorities.
The physical implementation of the cognitive core is done by the Navigation
and flight module that is the interface to the autopilots like PX4 or ArduPilot. It
performs maneuvers, passing route points and avoiding obstacles, while relying
on the visual odometry and SLAM data to ensure collision safety.
At the same time, the Security, threat and anomaly detection module is re-
sponsible for self-observation: temperature monitoring, CPU/GPU load, system
degradation detection, and activates fail-safe scenarios, or dynamically resched-
ules the swarm tasks in case of losses of individual agents. Detecting anomalies in
time series of sensory indicators allows the system to automatically react to po-
tential threats, detect compromised swarm participants, analyzing the irregular
patterns in input data. This approach is more flexible than the traditional heuristic
rules in robotized systems [4].
A strategically important link is the Digital twin module – a limited repre-
sentation of a fully functional digital twin deployed in the ground control center.
On-board this module is responsible for maintaining the relevant strategies, simu-
lation of the partial actions, and asynchronous renewal of the behavioral models
[14–17]. It guarantees the autonomous behavior even in case of a complete con-
nection loss, synchronizing data later.
Finally, the stability and security of the system is sustained by the Energy
management and hardware security module that includes communication encryp-
tion, agent authentication, multi-layered service backup, and power management.
This module allows the system to adapt to power supply limitations, lowering the
sensor operation intensity, or switching to the energy-saving mode in critical
moments. The whole multi-layered system provides the autonomous, adaptive and
resilient UAV swarm operation even in hostile or unpredictable environment, im-
plementing the modern approaches to the on-board cognitive management.
SCENARIO 4 (SEARCH AND RESCUE) IMPLEMENTATION PLAN EXAMPLE
The operational situation: after a large-scale earthquake in some region several
settlements were ruined. There is a risk of further collapses, and the access for the
ground rescue groups is limited. An autonomous scanning of territory with a total
area of nearly 30 km2 is required to find the victims, designate safe evacuation
zones, and transmit the coordinates to the ground forces.
The employment of the AI platform. To implement the scenario, a swarm sys-
tem of 16 autonomous quadcopters will be deployed. The drones will be equipped
with thermal imagers, RGB cameras, and laser rangefinders (LiDARs). The com-
putational platform of each drone allows local image processing, map charting,
and decision making. SLAM navigation, along with visual odometry and obstacle
avoidance module, will be used to form local maps, and dynamically plan routes
in real time. The behavioral coordination in the swarm will be implemented on
the base of combined Behavior Trees and Graph Neural Networks that will allow
adaptively distribute the tasks between agents, avoid duplication of the search
zones, and optimize the area coverage.
The platform will ensure:
Cognitive AI platform for autonomous navigation of distributed multi-agent systems
Системні дослідження та інформаційні технології, 2025, № 3
15
distribution of the swarm into sub-groups of 4 drones with partial (~10%)
overlap of the areas for increased probability of object detection;
detection of heat anomalies using a pre-trained neural network;
suppressing background noise (e.g., heat from transport or infrastructure);
exchanging scanned area tags, and analysis results between participants.
For synchronization of the swarm behavior the implementation of the sub-
tree broadcast protocol is planned that will transmit the minimal context every
few seconds. Communication between agents is planned to be achieved through
the ROS Topics + DDS with QoS parameters stack, providing reliable data ex-
change.
The expected data to be utilized includes:
previous mission simulation models, formed on the base of satellite image
data, topographical data and OSINT;
fallback behavior scenarios for cases of connection loss or situation change.
The transmission to the ground center is conducted through relay drones that
hover at up to 120 m height and form the mesh network. They transmit:
local maps;
visual confirmations;
coordinates of detected objects and safe areas;
GPS/SLAM log files.
The expected results include:
detection of the potentially alive targets using thermal signatures;
coverage map charting, and marking the risk areas;
designation of safe routes for evacuation;
transmission of the structured coordinates and statuses to the operational
headquarters.
CONCLUSIONS
1. The developed AI platform for the autonomous navigation of UAV
swarms presents a fundamentally new approach to handling the distributed multi-
agent systems under conditions of a complex, dynamic, and hostile environment.
Its architecture combines the ground control center, and the autonomous on-board
subsystem, providing a continuous loop of adaptation, learning and evolution for
artificial intelligence during each of the mission stages, from pre-mission model-
ing, to post-mission analysis. The ground control center performs the functions of
simulation, training, validation and strategic coordination, while each drone, due
to its cognitive core, sensory stack and communication modules, implements au-
tonomous navigation, recognition, and decision making without centralized control.
2. A number of basic scenarios (missions) is formed that cover a broad spec-
trum of combat and humanitarian tasks. These scenarios include both classic ob-
jectives (reconnaissance, targeted strikes, communication relay), and specialized
missions (search and rescue, misinformation, “death ring” strike), proving the
platform’s scalable and universal nature in dynamic environments. Formalization
and typification of such scenarios allow to not just quickly adapt the swarm to
new conditions but also form a repository for behavioral patterns that will be im-
proved using the principles of cognitive learning over time.
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko …
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3
16
3. An on-board component of the AI platform with cognitive services was
developed by combining a cognitive core, a sensor and analytical layer, naviga-
tion, communication, and security modules. Each drone in the system can act in-
dependently, adapt to the changes in environment, make critical decisions in real
time, and interact with other agents without centralized control. The hybrid appli-
cation of the AI swarm intelligence methods “Behavior Trees” and “Global
swarm optimization”, and SLAM methods provides situational prediction and
flexible reaction. The availability of power management, self-observation, and
local knowledge updates additionally fortifies the system’s survivability, and the
digital twin module provides the asynchronous swarm evolution even after con-
nection loss. All these functional capabilities prove that the on-board component
is not just a computational node, but an accomplished cognitive agent, able to
conduct missions within the decentralized new generation architecture.
4. A model search and rescue scenario of people after a catastrophe is pro-
posed, where a drone swarm autonomously scans the investigated area, detects
heat anomalies, identifies casualties, and transmits the coordinates for evacuation.
In preparing this manuscript, we used ChatGPT 4.0 to improve the style.
REFERENCES
1. S.-J. Chung, A.A. Paranjape, P. Dames, S. Shen, V. Kumar, “A survey on aerial
swarm robotics,” IEEE Transactions on Robotics, 34(4), pp. 837–855, 2018. doi:
https://doi.org/10.1109/TRO.2018.2857475
2. M. Brambilla, E. Ferrante, M. Birattari, M. Dorigo, “Swarm robotics: A review from
the swarm engineering perspective,” Swarm Intelligence, 7(1), pp. 1–41, 2013. doi:
https://doi.org/10.1007/s11721-012-0075-2
3. Z. Xia, J. Du, C. Jiang, J. Wang, Y. Ren, G. Li, “Multi-UAV cooperative target
tracking based on swarm intelligence,” in ICC 2021 – IEEE International Con-
ference on Communications, pp. 1–6. doi: https://doi.org/10.1109/ICC42927.
2021.9500771
4. A. Tytarenko, “Detecting unsafe behavior in neural network imitation policies for
caregiving robotics,” System Research and Information Technologies, no. 4, pp. 86–96,
2024. doi: https://doi.org/10.20535/SRIT.2308-8893.2024.4.07
5. P.O. Kasyanov, L.S. Paliichuk, “Acute angle lemma for noncompact image sets,”
Journal of Fixed Point Theory and Applications, 27(3), Article 67, 2025. doi:
https://doi.org/10.1007/s11784-025-01220-4
6. M.Z. Zgurovsky, P.O. Kasyanov, L.B. Levenchuk, “Formalization of Methods for
the Development of Autonomous Artificial Intelligence Systems,” Cybern. Syst.
Anal., 59, pp. 763–771, 2023. doi: https://doi.org/10.1007/s10559-023-00612-z
7. M.Z. Zgurovsky, Y.P. Zaychenko, Big Data: Conceptual Analysis and Applications.
Springer, 2020. doi: https://doi.org/10.1007/978-3-030-14298-8
8. M.Z. Zgurovsky, Y.P. Zaychenko, The Fundamentals of Computational Intelligence:
System Approach (Studies in Computational Intelligence, Vol. 652). Springer, 2017.
doi: https://doi.org/10.1007/978-3-319-35162-9
9. N.D. Pankratova, K.D. Grishyn, V.E. Barilko, “Digital twins: Stages of concept de-
velopment, areas of use, prospects,” System Research and Information Technologies,
no. 2, pp. 7–21, 2023. doi: https://doi.org/10.20535/SRIT.2308-8893.2023.2.01
10. N.D. Pankratova, I.M. Golinko, “Development of digital twins to support the func-
tioning of cyber-physical systems,” Computer Science Journal of Moldova,
31(3(93)), pp. 299–320, 2023. doi: https://doi.org/10.56415/csjm.v31.15
11. Y. Alqudsi, M. Makaracı, “UAV swarms: Research, challenges, and future direc-
tions,” Journal of Engineering and Applied Science, 72, Article 12, 2025. doi:
https://doi.org/10.1186/s44147-025-00582-3
Cognitive AI platform for autonomous navigation of distributed multi-agent systems
Системні дослідження та інформаційні технології, 2025, № 3
17
12. R. Arranz, D. Carramiñana, G. de Miguel, J.A. Besada, A.M. Bernardos, “Applica-
tion of deep reinforcement learning to UAV swarming for ground surveillance,” Sen-
sors, 23(21), 8766, 2025. doi: https://doi.org/10.3390/s23218766
13. L. Tan et al., “Multi-UAV-enabled collaborative edge computing: Deployment, off-
loading and resource optimization,” IEEE Transactions on Intelligent Transportation
Systems, 25(11), pp. 13741–13754, 2022. doi: https://doi.org/10.1109/TITS.
2024.3432818
14. Z. Li, L. Lei, G. Shen, X. Liu, X. Liu, “Digital twin-enabled deep reinforcement
learning for safety-guaranteed flocking motion of UAV swarm,” Transactions on
Emerging Telecommunications Technologies, 35(11), e70011, 2024. doi:
https://doi.org/10.1002/ett.70011
15. L.R. Salinas, G. Tzoumas, L. Pitonakova, S. Hauert, “Digital twin technology for
wildfire monitoring using UAV swarms,” in 2023 International Conference on Un-
manned Aircraft Systems (ICUAS), IEEE, 2023, pp. 586–593. doi: https://doi.org/
10.1109/ICUAS57906.2023.10155819
16. B.R. Barricelli, E. Casiraghi, D. Fogli, “A survey on digital twin: Definitions, char-
acteristics, applications, and design implications,” IEEE Access, 7, pp. 167653–167671,
2019. doi: https://doi.org/10.1109/ACCESS.2019.2953499
17. T. Li et al., “Digital twin-based task-driven resource management in intelligent UAV
swarms,” IEEE Transactions on Intelligent Transportation Systems, 26(4), pp. 3905–3918,
2025. doi: https://doi.org/10.1109/TITS.2025.3531120
18. T. Bray, J. Paoli, C.M. Sperberg-McQueen, E. Maler, F. Yergeau, “Extensible
Markup Language (XML) 1.0,” W3C, 2008. Avaialble: https://www.w3.org/TR/xml/
19. H.R. Ahmed, J.I. Glasgow, Swarm intelligence: Concepts, models and applications;
Technical Report 2012-585. Queen’s University, School of Computing Technical
Reports, 2012. doi: https://doi.org/10.13140/2.1.1320.2568
20. D. Vernon, “Cognitive system,” in K. Ikeuchi (Ed.), Computer Vision. Springer,
Cham, 2021. doi: https://doi.org/10.1007/978-3-030-63416-2_82
21. H. Huang, J. Qiu, K. Riedl, “On the global convergence of particle swarm optimiza-
tion methods,” Applied Mathematics & Optimization, 88, 30, 2023. doi:
https://doi.org/10.1007/s00245-023-09983-3
22. M. Iovino, E. Scukins, J. Styrud, P. Ögren, C. Smith, “A survey of behavior trees in
robotics and AI,” Robotics and Autonomous Systems, 154, 104096, 2022. doi:
https://doi.org/10.1016/j.robot.2022.104096
23. C. Theodorou, V. Velisavljevic, V. Dyo, F. Nonyelu, “Visual SLAM algorithms and
their application for AR, mapping, localization and wayfinding,” Array, 15, 100222,
2022. doi: https://doi.org/10.1016/j.array.2022.100222
24. P. Su, S. Luo, X. Huang, “Real-time dynamic SLAM algorithm based on deep learn-
ing,” IEEE Access, 10, pp. 87264–87276, 2022. doi: https://doi.org/10.1109/
ACCESS.2022.3199350
25. “Data Distribution Service (DDS) Specification Version 1.4,” Object Management
Group, 2015. doi: https://www.omg.org/spec/DDS/
26. B. Tekinerdogan, Ö. Köksal, T. Çelik, “Data distribution service-based architecture
design for the Internet of Things systems,” in Z. Mahmood (Ed.), Connected Envi-
ronments for the Internet of Things, pp. 201–220. Springer, Cham, 2017. doi:
https://doi.org/10.1007/978-3-319-70102-8_13
27. V. Bode, C. Trinitis, M. Schulz, D. Buettner, T. Preclik, “DDS implementations as
real-time middleware – A systematic evaluation,” in 2023 IEEE 29th International
Conference on Embedded and Real-Time Computing Systems and Applications
(RTCSA), pp. 39–46. doi: https://doi.org/10.1109/RTCSA58653.2023.00030
Received 14.08.2025
M.Z. Zgurovsky, P.O. Kasyanov, N.D. Pankratova, Yu.P. Zaychenko …
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3
18
INFORMATION ON THE ARTICLE
Michael Z. Zgurovsky, ORCID: 0000-0001-5896-7466, Educational and Scientific Complex
“Institute for Applied System Analysis” of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: zgurovsm@hotmail.com
Pavlo O. Kasyanov, ORCID: 0000-0002-6662-0160, Educational and Scientific Complex
“Institute for Applied System Analysis” of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: p.o.kasyanov@gmail.com
Nataliya D. Pankratova, ORCID: 0000-0002-6372-5813, Educational and Scientific
Complex “Institute for Applied System Analysis” of the National Technical Univer-
sity of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail:
natalidmp@gmail.com
Yuriy P. Zaychenko, ORCID: 0000-0001-9662-3269, Educational and Research Institute
for Applied System Analysis of the National Technical University of Ukraine “Igor Sikor-
sky Kyiv Polytechnic Institute”, Ukraine, e-mail: zaychenkoyuri@ukr.net
Illia O. Savchenko, ORCID: 0000-0002-0921-5425, Educational and Research Institute
for Applied System Analysis of the National Technical University of Ukraine “Igor Sikor-
sky Kyiv Polytechnic Institute”, Ukraine, e-mail: i.savchenko@kpi.ua
Tetyana V. Shovkoplyas, ORCID: 0009-0004-8991-0285, Taras Shevchenko Na-
tional University of Kyiv, Ukraine, e-mail: from_tatyana@ukr.net
Liliia S. Paliichuk, ORCID: 0000-0003-1654-4371, Educational and Scientific Complex
“Institute for Applied System Analysis” of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: lili262808@gmail.com
Andrii M. Tytarenko, ORCID: 0000-0002-8265-642X, Educational and Research Insti-
tute for Applied System Analysis of the National Technical University of Ukraine “Igor
Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: titarenkoan@gmail.com
КОГНІТИВНА AI-ПЛАТФОРМА ДЛЯ АВТОНОМНОЇ НАВІГАЦІЇ
РОЗПОДІЛЕНИХ БАГАТОАГЕНТНИХ СИСТЕМ / М.З. Згуровський,
П.О. Касьянов, Н.Д. Панкратова, Ю.П. Зайченко, І.О. Савченко, Т.В. Шовкопляс,
Л.С. Палійчук, А.М. Титаренко
Анотація. Подано концепцію когнітивної AI-платформи для автономної наві-
гації розподілених багатоагентних систем на прикладі рою безпілотних літа-
льних апаратів. Запропоновано архітектуру, яка поєднує наземний центр із ко-
гнітивними сервісами та багаторівневу бортову підсистему, що забезпечують
безперервний цикл навчання, адаптації, виконання та оновлення поведінкових
моделей. Сформульовано базові сценарії місій, зокрема розвідка, пошук і ря-
тування, ураження цілей, дезінформація, які демонструють можливості рою до
автономної, децентралізованої взаємодії навіть у ворожому середовищі. Пред-
ставлено приклад плану реалізації місії пошуку і рятування із використанням
когнітивної платформи, що включає адаптивне планування, SLAM-навігацію,
ройову координацію та глибоке розпізнавання об’єктів. Результати частково
підтримано Національним фондом досліджень України, грант № 2025.06/0022
«AI-платформа з когнітивними сервісами для координованої автономної наві-
гації розподілених систем, що складаються з великої кількості об’єктів».
Ключові слова: штучний інтелект, рій дронів, автономна навігація, когнітив-
на платформа, мультиагентні системи, поведінкові дерева, цифровий двійник,
SLAM.
|
| id | journaliasakpiua-article-342981 |
| 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/18/cc461b01cfcc2265a516b98c88563518.pdf |
| spelling | journaliasakpiua-article-3429812025-11-09T00:01:30Z Cognitive AI platform for autonomous navigation of distributed multi-agent systems Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем Zgurovsky, Michael Kasyanov, Pavlo Pankratova, Nataliya Zaychenko, Yuriy Savchenko, Illia Shovkoplyas, Tetyana Paliichuk, Liliia Tytarenko, Andrii artificial intelligence UAV swarm autonomous navigation cognitive platform multi-agent systems behavior trees digital twin SLAM штучний інтелект рій дронів автономна навігація когнітивна платформа мультиагентні системи поведінкові дерева цифровий двійник SLAM This paper presents a concept for a cognitive AI platform that enables autonomous navigation of distributed multi-agent systems, exemplified by UAV swarms. The proposed architecture integrates a ground control center with cognitive services and a multi-layered onboard subsystem, supporting a continuous loop of learning, adaptation, execution, and behavioral model updates. Several core mission scenarios are introduced, such as reconnaissance, search and rescue, target neutralization, and deception, showcasing the swarm’s ability to operate autonomously and in a decentralized manner, even under adversarial conditions. An example of a search and rescue mission implementation plan using a cognitive platform that includes adaptive planning, SLAM navigation, swarm coordination, and deep object recognition is presented. 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”. Подано концепцію когнітивної AI-платформи для автономної навігації розподілених багатоагентних систем на прикладі рою безпілотних літальних апаратів. Запропоновано архітектуру, яка поєднує наземний центр із когнітивними сервісами та багаторівневу бортову підсистему, що забезпечують безперервний цикл навчання, адаптації, виконання та оновлення поведінкових моделей. Сформульовано базові сценарії місій, зокрема розвідка, пошук і рятування, ураження цілей, дезінформація, які демонструють можливості рою до автономної, децентралізованої взаємодії навіть у ворожому середовищі. Представлено приклад плану реалізації місії пошуку і рятування із використанням когнітивної платформи, що включає адаптивне планування, SLAM-навігацію, ройову координацію та глибоке розпізнавання об’єктів. Результати частково підтримано Національним фондом досліджень України, грант № 2025.06/0022 "AI-платформа з когнітивними сервісами для координованої 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/342981 10.20535/SRIT.2308-8893.2025.3.01 System research and information technologies; No. 3 (2025); 7-18 Системные исследования и информационные технологии; № 3 (2025); 7-18 Системні дослідження та інформаційні технології; № 3 (2025); 7-18 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/342981/330918 |
| spellingShingle | штучний інтелект рій дронів автономна навігація когнітивна платформа мультиагентні системи поведінкові дерева цифровий двійник SLAM Zgurovsky, Michael Kasyanov, Pavlo Pankratova, Nataliya Zaychenko, Yuriy Savchenko, Illia Shovkoplyas, Tetyana Paliichuk, Liliia Tytarenko, Andrii Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем |
| title | Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем |
| title_alt | Cognitive AI platform for autonomous navigation of distributed multi-agent systems |
| title_full | Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем |
| title_fullStr | Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем |
| title_full_unstemmed | Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем |
| title_short | Когнітивна AI-платформа для автономної навігації розподілених багатоагентних систем |
| title_sort | когнітивна ai-платформа для автономної навігації розподілених багатоагентних систем |
| topic | штучний інтелект рій дронів автономна навігація когнітивна платформа мультиагентні системи поведінкові дерева цифровий двійник SLAM |
| topic_facet | artificial intelligence UAV swarm autonomous navigation cognitive platform multi-agent systems behavior trees digital twin SLAM штучний інтелект рій дронів автономна навігація когнітивна платформа мультиагентні системи поведінкові дерева цифровий двійник SLAM |
| url | https://journal.iasa.kpi.ua/article/view/342981 |
| work_keys_str_mv | AT zgurovskymichael cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT kasyanovpavlo cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT pankratovanataliya cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT zaychenkoyuriy cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT savchenkoillia cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT shovkoplyastetyana cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT paliichukliliia cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT tytarenkoandrii cognitiveaiplatformforautonomousnavigationofdistributedmultiagentsystems AT zgurovskymichael kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT kasyanovpavlo kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT pankratovanataliya kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT zaychenkoyuriy kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT savchenkoillia kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT shovkoplyastetyana kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT paliichukliliia kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem AT tytarenkoandrii kognítivnaaiplatformadlâavtonomnoínavígacíírozpodílenihbagatoagentnihsistem |