LLM-Centric Agentic AI for UAV Swarms: Architecture, Enabling Technologies, and Open Problems

Jul 16, 2026

Upload by Hazel

UAV swarms enable search and rescue and environmental monitoring, yet real deployment is constrained by limited situational awareness, intermittent connectivity, and significant cybersecurity risk. Agentic AI reframes standalone LLMs into closed-loop cognitive architectures that connect perception, memory, reasoning/planning, and action for adaptive, goal-directed swarm behavior. The paper proposes LLM-Centric Agentic AI for UAV Swarms (LAUS), reviews enabling technologies spanning onboard/edge computing, 5G/6G links, multimodal intelligence, and security mechanisms, and analyzes Priority Manipulation Attacks (PMA). Open problems include hallucination-resistant reasoning, SWaP-aware onboard LLM deployment, and standardized benchmarks for perception–reasoning attacks.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

I. Introduction

II. Proposed LAUS Architecture, Deployment Challenges, and Applications

III. Enabling Technologies and Cybersecurity

IV. Adversarial Threats and Defense Mechanisms

V. Case Study on Priority Manipulation Attacks (PMA)

VI. Challenges and Open Problems

LLM-Centric Agentic AI for UAV Swarms: Architecture, Enabling Technologies, and Open Problems Yousef Emami, Senior Member, IEEE, Rahim Taheri, Senior Member, IEEE, Mohammadhossein Homaei, Senior Member, IEEE, and Muhammad Atif Ur Rehman, Senior Member, IEEE, and Mohammad Shojafar, Senior Member, IEEE arXiv :2607 .09756v 1 [ cs .RO] 5 Jul 2026 Abstract—Uncrewed Aerial Vehicle (UAV) swarms have significant potential for applications such as Search and Rescue (SAR) and environmental monitoring, but their real-world deployment is limited by a lack of situational awareness, intermittent connectivity, and significant cybersecurity risks. Agentic Artificial Intelligence (AI) represents a shift from standalone Large Language Model (LLM) toward closed-loop cognitive architectures that integrate perception, memory, reasoning/planning, and action to enable adaptive, goal-directed swarm behavior. Within this framework, Agentic AI provides a unifying structure for autonomous and adaptive swarm operations while expanding the system’s attack surface compared to conventional AI systems. This paper proposes LLM-Centric Agentic AI for UAV Swarms (LAUS) and reviews key enabling technologies such as onboard and edge computing, 5G/6G connectivity, multimodal intelligence, and cybersecurity mechanisms, and analyzes threats such as Priority Manipulation Attacks (PMA) that can distort decision-making and degrade network performance. Finally, it identifies open research challenges, including hallucination-resistant reasoning, onboard LLM deployment under SWaP constraints, and standardized security benchmarks for perception-reasoning attacks in agentic UAV systems. Index Terms—Uncrewed Aerial Vehicles, Agentic AI, Large Language Model Agents, UAV swarms, Security I. INTRODUCTION Thanks to their high mobility and ability to establish Line-ofSight (LoS) communications, Uncrewed Aerial Vehicle (UAV) swarms have been widely adopted in Search and Rescue (SAR) missions, environmental monitoring, parcel delivery, precision agriculture, and construction [1] . UAV swarms face several key challenges and limitations, including ethical concerns, poor energy efficiency that restricts operational endurance, difficulties in autonomous control and task allocation within complex environments, communication breakdowns as swarm size grows, insufficient robustness and scalability of algorithms for dynamic settings, synchronization issues caused by environmental disturbances, and security vulnerabilities that can lead to coordination failures [2], [3] . Real-world UAV swarm deployment is fundamentally constrained by restricted situational awareness, intermittent connectivity, and persistent cybersecurity threats, challenges that standard LLMs cannot adequately address. Despite their strong language understanding, LLMs are architecturally misaligned with autonomous swarm operations, lacking temporal memory, native tool interaction, and the capacity to generate verifiable real-time control actions in safety-critical settings. Agentic AI addresses this gap by shifting from passive inference to Copyright (c) 2026 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to [pubs-permissions@ieee.org](pubs-permissions@ieee.org). closed-loop autonomy, in which an Agent Loop continuously perceives, plans, executes, and learns, enabling adaptive and self-improving swarm behavior. Crucially, the focus of Agentic AI is the cognitive architecture within each agent, not merely the coexistence of multiple cooperating UAVs [4] . Sapkota et al. [5] provide a foundational framework that defines and differentiates Agentic UAVs from traditional ones, while mapping their transformative potential across seven key societal domains. Koubaa et al. [6] suggest a five-layer Agentic UAV framework structured around perception, reasoning, action, integration, and learning.

FAQs

Why are standard LLMs insufficient for real-world UAV swarm autonomy?

They lack temporal memory, native tool interaction, and the ability to produce verifiable real-time control actions required in safety-critical swarm operations, while swarms also face restricted situational awareness and intermittent connectivity.

What does LAUS propose for agentic UAV swarm behavior?

LAUS offers a closed-loop, LLM-assisted agentic framework that converts mission objectives into coordinated UAV actions, emphasizing the cognitive architecture inside each agent rather than only multi-UAV cooperation.

What is a Priority Manipulation Attack (PMA) in this context?

PMA corrupts input observations so an agent’s decision-making is distorted, degrading swarm performance and exposing an attack surface at the perception–reasoning interface.

Jul 16, 2026

Upload by Hazel