LLM-Enabled UAV Dispatch: A System-Level Survey and Taxonomy

Xiao Han; Aoyang Quan; Xiangyu Zhao; Xiangjie Kong; Guojiang Shen

cs.AIOct 7, 2026

LLM-Enabled UAV Dispatch: A System-Level Survey and Taxonomy

Authors: Xiao Han, Aoyang Quan, Xiangyu Zhao, Xiangjie Kong, Guojiang Shen

Organizations: School of Computer Science, Zhejiang University of Technology, China

Abstract

Unmanned aerial vehicle (UAV) dispatch is beginning to move beyond isolated path planning and optimization-driven resource allocation toward system-level coordination supported by semantic reasoning and LLM-based interfaces. This survey provides a unified characterization of LLM-enabled UAV dispatch systems that bridges semantic intent, symbolic decision-making, and physical UAV execution. Rather than treating LLMs as standalone add-ons, we conceptualize them as a cross-layer semantic orchestration layer connecting human instructions, external solvers, and distributed control modules. We organize the literature into four representative dispatch paradigms: pipeline dispatch, global assignment dispatch, decentralized agentic dispatch, and divide-and-conquer dispatch. For each paradigm, we analyze its decision logic, system structure, control flow, representative methods, and potential LLM roles. We further examine how LLMs support semantic parsing, retrieval-grounded planning, solver orchestration, local agent reasoning, multi-agent coordination, safety assessment, and human-facing explanation. We discuss the implications of these paradigms for scalability, robustness, coordination burden, and verification requirements, and identify open challenges including latency-aware reasoning, grounding reliability, physical feasibility guarantees, edge deployment, privacy protection, and distributed consistency. This survey provides a system-level taxonomy and design perspective for integrating LLMs into safety-critical UAV dispatch systems.

Figures & tables

Fig. 1: Evolution and scope of LLM-enabled UAV dispatch.

Surveys	Focus & Scope	Key Findings / Application Categories
[ 1 ]	Reviews low-altitude airspace management for uncrewed aircraft and advanced air mobility.	Summarizes airspace concepts, operational requirements, BVLOS integration, regulatory constraints, and traffic-management challenges.
[ 31 ]	Surveys trajectory-prediction techniques for UAVs across modeling, data-driven learning, and motion-pattern analysis.	Classifies trajectory-prediction methods and discusses their role in collision avoidance, motion forecasting, and safe UAV operation.
[ 32 ]	Reviews UAV path-planning techniques with emphasis on planning algorithms, environment representation, and motion constraints.	Categorizes path-planning approaches by optimization strategy, search mechanism, and constraint handling, with limited attention to end-to-end dispatch logic.
[ 33 ]	Surveys computational-intelligence algorithms for UAV swarm networking and collaboration.	Reviews communication, cooperative control, formation, search, learning-based collaboration, and future swarm-networking directions.
[ 34 ]	Reviews swarm-intelligence algorithms for multi-UAV collaboration.	Summarizes distributed cooperation, formation, task coordination, search, and collective decision-making mechanisms for UAV swarms.
[ 35 ]	Reviews privacy and confidentiality issues in drone operations.	Identifies privacy leakage, data exposure, trust, authentication, and secure communication challenges in UAV deployment.

TABLE I: Comparison with representative surveys related to UAV dispatch and LLM-enabled aerial intelligence.

Fig. 2: Pipeline dispatch decomposes a mission into ordered processing stages, where an LLM layer can parse intent, call tools, and trigger verification.

Pipeline Transition	Classical Stage Role	LLM-Era Orchestration Role	Representative Methods
$f_{1}:\mathcal{M}\!\rightarrow\!\mathcal{T}$	Spatial or task decomposition	Semantic scene parsing and RAG-grounded task graph construction	Li et al. [ 42 ] , Xiong et al. [ 43 ]
	Hierarchical mission decomposition	Natural-language mission breakdown and role prompting	Luan et al. [ 44 ] , Cheng et al. [ 45 ] , Guo et al. [ 46 ] , Li et al. [ 42 ] , Jünger et al. [ 47 ]
	Rolling or adaptive clustering	Context-aware reclustering from execution feedback	Yanmaz et al. [ 48 ] , Ren et al. [ 5 ]
$f_{2}:\mathcal{T}\!\rightarrow\!\mathbf{X}$	Task or role assignment	Tool-call handoff from task graph to assignment solver	Yanmaz et al. [ 48 ] , Guo et al. [ 46 ] , Zhao et al. [ 49 ] , Li et al. [ 42 ] , Gong et al. [ 6 ]
	Hierarchical task allocation	Constraint generation and priority explanation	Luan et al. [ 44 ] , Cheng et al. [ 45 ] , Guo et al. [ 46 ]
	Battery or inspection-task matching	Resource-state summarization and repair hints	Alyassi et al. [ 50 ] , Ren et al. [ 5 ]

TABLE II: Taxonomy of representative pipeline dispatch methods and their LLM-era orchestration roles aligned with Eq. ( 2 ).

Fig. 3: Global assignment dispatch uses a central dispatcher to jointly determine tasks, routes, schedules, and resources, with an LLM layer supporting intent interpretation and solver selection.

Central Dispatch Dimension	Coupled Decision Role	LLM-Era Assignment Role	Representative Methods
Mobility–communication	Trajectory and link scheduling	Global-state summarization and link-aware prompt grounding	Liu and Zheng [ 68 ] , He et al. [ 69 ] , Hu et al. [ 70 ] , Jing et al. [ 52 ] , Zhang et al. [ 53 ] , Wu et al. [ 71 ]
	Data collection, offloading, and MEC	Service-intent parsing and resource constraint extraction	Nguyen et al. [ 72 ] , Zhu et al. [ 73 ] , Liu and Zheng [ 68 ] , He et al. [ 69 ] , Hu et al. [ 70 ] , Shi et al. [ 74 ]
	Association, wireless power, and ISAC	Multi-objective tradeoff explanation and search guidance	Jing et al. [ 52 ] , Zhang et al. [ 53 ] , Wu et al. [ 71 ] , Li et al. [ 28 ]
Routing–scheduling	TSP/VRP-style UAV routing	Natural-language service request to routing constraints	Montemanni and Dell’Amico [ 75 ] , Joo et al. [ 76 ]
	Truck, vehicle, or mothership coordination	Cross-platform synchronization explanation	Joo et al. [ 76 ]
	Delivery, monitoring, inspection, and rescue	Priority-aware dispatch briefing and schedule repair	Yang et al. [ 77 ] , Rigas et al. [ 78 ] , Zhao et al. [ 79 ] , Mao et al. [ 80 ] , Pei et al. [ 81 ] , Betti Sorbelli et al. [ 82 ]

TABLE III: Taxonomy of representative global assignment dispatch methods and their LLM-era assignment roles.

Fig. 4: Decentralized agentic dispatch lets local UAV agents select tasks from observations, utilities, short-term memory, and limited peer messages.

Agentic Layer	Local Dispatch Role	LLM-Agent Role	Representative Methods
Local decision layer	Reinforcement-learning trajectory control	Perception–memory–action loop for local intent interpretation	Ebrahimi et al. [ 108 ] , Nguyen et al. [ 72 ] , Zhu et al. [ 73 , 86 ]
	Active perception under uncertainty	Multimodal observation summarizer and next-view planner	Zhu and Chen [ 60 ] , Zhu et al. [ 61 ] , Kirillov et al. [ 63 ]
	Tracking, avoidance, and covert motion	Local safety critic and intent-aware motion proposal	Xu et al. [ 109 ] , Zhang et al. [ 110 ] , Huang et al. [ 111 ]
Swarm coordination layer	Decentralized coverage and exploration	Peer-agent message compression and local consensus prompting	blueZhao et al. [ 112 ] , Huang et al. [ 113 ]
	Relative-state UAV interaction	Relative-observation grounding for semantic communication	Hasan et al. [ 114 ]
	MARL and distributed reinforcement learning	Role-aware multi-agent reasoning with bounded messages	Venturini et al. [ 115 ] , Qin and Pournaras [ 90 ] , Hu et al. [ 87 ] , Alkahtani et al. [ 116 ] , Luo et al. [ 29 ]

TABLE IV: Taxonomy of representative decentralized agentic UAV dispatch methods and their LLM-agent roles.

Fig. 5: Divide-and-conquer dispatch partitions a large mission into local subproblems, solves them separately, and reconciles them through a coordinator.

Decomposition Principle	Dispatch Role	LLM-Era Orchestration Role	Representative Methods
Decision and objective decomposition	Trajectory, energy, and link tradeoff	Objective decomposer and Pareto-search guide	Javed et al. [ 125 ] , Meng et al. [ 126 ] , Jing et al. [ 52 ]
	Secure trajectory, power, and robustness	Security-rule retrieval and feasibility critic	Kang et al. [ 91 ] , Zhang et al. [ 92 ] , Dang-Ngoc et al. [ 93 ]
	Temporal, connectivity, and ISAC objectives	Communication–sensing tradeoff explanation	Meng et al. [ 126 ] , Jing et al. [ 52 ] , Zhang et al. [ 53 ] , Li et al. [ 28 ]
Spatial and regional decomposition	Three-dimensional and multi-region coverage	Semantic map and region partitioning	Xiong et al. [ 43 ] , Li et al. [ 42 ]
	Emergency waypoints and rescue paths	Disaster-scene grounding and priority partitioning	Liu et al. [ 127 ] , Liu et al. [ 30 ] , Lin et al. [ 24 , 56 ] , Xiong et al. [ 43 ]
	Inspection and agriculture scene modeling	Multimodal region labeling and task extraction	Liu et al. [ 127 ] , Liu et al. [ 30 ] , Zhang et al. [ 9 ] , Zhu and Chen [ 60 ] , Kirillov et al. [ 63 ]

TABLE V: Taxonomy of representative divide-and-conquer dispatch methods and their LLM-era orchestration roles.

Hotspot	Dispatch Focus	Key Research Issues
RAG-grounded dispatch knowledge bases	Retrieve maps, regulations, weather, mission logs, and communication context for planning	Source freshness, retrieval faithfulness, traceable constraint generation
LLM-tool-solver co-design	Connect LLM reasoning with optimizers, simulators, planners, and safety monitors	Stable interfaces, feasibility checking, repairable plan representations
Decentralized agentic dispatch	Support local semantic interpretation, task claiming, peer negotiation, and replanning	Latency, compact protocols, prompt safety, distributed verification
Semantic divide-and-conquer	Partition missions by regions, roles, objectives, knowledge sources, and model placement	Boundary conflicts, load balancing, cross-agent consistency
Multimodal aerial reasoning	Fuse aerial images, videos, maps, text instructions, and sensor summaries	Scene grounding, hallucination control, sim-to-real transfer
Edge multi-LLM orchestration	Assign reasoning tasks across onboard, edge, and cloud models	Model selection, fallback policies, energy and bandwidth limits

TABLE VI: Future research directions for LLM-enabled UAV dispatch.

Explore similar work

May 5, 2026cs.AI

Say the Mission, Execute the Swarm: Agent-Enhanced LLM Reasoning in the Web-of-Drones

Large Language Models (LLMs) are increasingly explored as high-level reasoning engines for cyber-physical systems, yet their application to real-time UAV swarm management remains challenging due to heterogeneous interfaces, limited grounding, and the need for long-running closed-loop execution. This paper presents a mission-agnostic, agent-enhanced LLM framework for UAV swarm control, where users express mission objectives in natural language and the system autonomously executes them through grounded, real-time interactions. The proposed architecture combines an LLM-based Agent Core with a Model Context Protocol (MCP) gateway and a Web-of-Drones abstraction based on W3C Web of Things (WoT) standards. By exposing drones, sensors, and services as standardized WoT Things, the framework enables structured tool-based interaction, continuous state observation, and safe actuation without relying on code generation. We evaluate the framework using ArduPilot-based simulation across four swarm missions and six state-of-the-art LLMs. Results show that, despite strong reasoning abilities, current general-purpose LLMs still struggle to achieve reliable execution - even for simple swarm tasks - when operating without explicit grounding and execution support. Task-specific planning tools and runtime guardrails substantially improve robustness, while token consumption alone is not indicative of execution quality or reliability.

Andrea Iannoli, Lorenzo Gigli, Luca Sciullo +2

Department of Computer Science and Engineering University of Bologna Bologna, Italy

Unmanned Aerial Vehicle Swarms Swarms Autonomous Drones LLM Reasoning Strategies Model Context Protocol

Ask PDF

Jun 30, 2026cs.AI

MultiUAV-Plat: An LLM-Oriented Platform, Benchmark and Framework for Multi-UAV Collaborative Task Planning

Large language models (LLMs) provide a promising interface for high-level robotic task planning, but their use in multi-UAV collaboration remains difficult to evaluate systematically. Existing UAV simulators mainly emphasize dynamics, perception, or low-level control, while existing LLM-agent benchmarks rarely capture aerial-robotics constraints such as partial observability, spatial coverage, UAV assignment, and multi-vehicle coordination. To bridge this gap, we present MultiUAV-Plat, a lightweight, easy-to-use, LLM-agent-oriented simulation platform for multi-UAV collaborative task planning. The platform exposes concise RESTful APIs, agent-facing observations, role-based information access, hidden validation logic, and optional 2D/3D visualization, allowing agents to solve missions through realistic tool interaction rather than privileged simulator access. Built on this platform, the MultiUAV-Plat Benchmark contains 75 mission sessions, 1500 natural-language tasks, and 9396 validation checks across target assignment, area search, and area assignment and patrol scenarios. We further propose Agent4Drone, a task-specific LLM agent framework that structures multi-UAV behavior into memory, observation, task understanding, planning, execution, and verification. In a full paired benchmark comparison, Agent4Drone achieves a 57.9% task pass rate, a 74.6% average task check pass rate, and a 72.0% global check pass rate, substantially outperforming a ReAct baseline at 30.6%, 47.9%, and 43.1%, respectively. Agent4Drone also reduces the total failed task rate from 32.4% to 12.9%. These results demonstrate that MultiUAV-Plat and MultiUAV-Plat Benchmark provide a reproducible foundation for studying LLM-driven multi-UAV autonomy under realistic information and execution constraints.

Sheng Zhang, Qinglin Li, Yuechao Zang +3

National Key Laboratory of Information Systems Engineering, National University of Defense Technology, Deya Road, Changsha, 410073, Hunan, China

Multi-Uav Autonomous Drones Large Language Model Agents Planning

Ask PDF

Aug 12, 2026cs.CV

Advancing MLLM-based UAV Image Understanding and Reasoning: A Benchmark and a Training-Free Multi-Agent System

Multimodal Large Language Model (MLLM)-based UAV aerial image understanding and reasoning is essential for aerial intelligence yet poses distinct challenges arising from extreme scale variation, arbitrary camera orientations, and high object density. Despite growing interest, existing evaluations remain fragmented across individual datasets and narrow tasks, leaving a critical gap in unified assessment of UAV understanding and reasoning capabilities. To fill this gap, we construct UAVQA-Bench, a benchmark of 1,500 human-annotated QA pairs drawn from 13 public UAV datasets, covering 6 capability dimensions and 16 tasks in both multiple-choice and visual grounding formats. Systematic evaluation of a broad range of open-source and closed-source MLLMs as well as agent-based systems on UAVQA-Bench identifies three key failure modes: domain-toolset mismatch, unchecked error propagation, and static reasoning. Motivated by these findings, we propose UAV-MAS, a training-free multi-agent system for MLLM-based UAV aerial image understanding and reasoning, comprising a Domain-Specific Perception Engine (DSPE) that routes queries to task-appropriate visual tools, a Context-Aware Iterative Refinement module (CAIR) that validates intermediate reasoning to curb error accumulation, and a Difficulty-Aware Adaptive Search mechanism (DAAS) that adjusts search depth to question difficulty. UAV-MAS with a 32B open-source MLLM achieves 77.0% overall accuracy on UAVQA-Bench, surpassing Gemini 3 Pro by 4.0%, while the 8B variant improves 8.7% over its base model.

Haoyu Zhang, Shuoxun Zhang, Peng Ye +5

Embedded Deep Learning and Visual Analysis Laboratory, College of Future Information Technology, Fudan University, Shanghai 200433, China · Embedded Deep Learning and Visual Analysis Laboratory, College of Future Information Technology, Fudan University, Shanghai 200433, China, and also with Shanghai Artificial Intelligence Laboratory, Shanghai 200232, China, and also with The Chinese University of Hong Kong, Hong Kong · College of Future Information Technology, Fudan University, Shanghai 200433, China +1

Aerial Dataset Unmanned Aerial Vehicles Mixture-Of-Agents

Ask PDF