arXiv CSopen access

Collaborative Air-Ground Sensing, Communication, Computing, Storage, and Intelligence for Low-Altitude Economy

arXiv:2605.18503v1 [cs.NI] 18 May 2026

Collaborative Air-Ground Sensing, Communication, Computing, Storage, and Intelligence for Low-Altitude Economy YIQIN DENG, Lingnan University, China JUNHUI GAO, City University of Hong Kong, China ZIHAN FANG, City University of Hong Kong, China YANAN MA, City University of Hong Kong, China XIANHAO CHEN, The University of Hong Kong, China YUGUANG FANG, City University of Hong Kong, China Low-altitude economy (LAE) is transforming low-altitude airspace into a new cyber-physical infrastructure. Although air-ground communications have been widely studied, LAE is fundamentally different in the sense that it is mission-centric with diverse requirements, such as stringent safety and compliance constraints not be effectively addressed with a communication-centric design alone, which makes air-ground collaboration indispensable: Only through effectively coordinating air-ground infrastructure and resources can LAE missions be fulfilled. Consequently, LAE calls for task-driven, closed-loop, multi-resource orchestration of Sensing, Communication, Computing, Storage, and Intelligence (SCCSI), where key decisions must be co-designed under mobility and uncertainty. In this paper, we first present a novel framework that connects (i) LAE scenarios and a requirement–resource coupling matrix, (ii) an air–ground collaborative architecture, and (iii) methodological toolboxes for SCCSI co-optimization and online decision-making. We then systematically review enabling technologies for collaborative SCCSI resources and capabilities, emphasizing their coupling and end-to-end tradeoffs. Finally, we summarize testbeds, datasets, and evaluation metrics, and provide representative use cases to illustrate how the proposed framework translates application requirements into practical task-driven optimization designs, together with open challenges and a roadmap toward scalable and trustworthy LAE deployment. Additional Key Words and Phrases: Low-altitude economy, air–ground collaboration, sensing–communication–computing–storage– intelligence (SCCSI), multi-resource orchestration, unmanned aerial vehicle (UAV)

1 Introduction Low-altitude economy (LAE) refers to economic activities in the airspace under 1, 000 meters above ground (with extensions to 3, 000 meters for specialized applications), enabled by advanced air mobility platforms such as unmanned aerial vehicles (UAVs1 ), electric vertical take-off and landing (eVTOL) aircraft, and air taxis. These technologies are catalyzing transformative applications in urban logistics, precision agriculture, emergency medical supply delivery, intelligent transportation management and control, infrastructure inspection, and tourism [1, 45]. 1

In this paper, the terms “UAV” and “drone” are used interchangeably, referring to the same type of unmanned aerial vehicle.

Yiqin Deng is with Lingnan University, Hong Kong, China (email: [email protected]); Junhui Gao, Zihan Fang, Yanan Ma, and Yuguang Fang are with City University of Hong Kong, Hong Kong, China (emails: [email protected], {zihanfang3-c, yananma8-c}@my.cityu.edu.hk, [email protected]); Xianhao Chen is with the University of Hong Kong, Hong Kong, China (email: [email protected]).

2026. Manuscript submitted to ACM Manuscript submitted to ACM

1

2

Y. Deng et al. Aerial Node - UAV Multi-view Imagery Capture Lightweight Feature Extraction GNSS Blockage

Task-oriented Feature Compression

Communication (Feature Delivery & Uplink Constraints)

Losing Locations in GPSdenied City Canyons

Ground Node – Edge Server Computation-Intensive Inference and Refinement Mapping & Localization Refinement Closed-loop Navigation

Compressed Features (e.g., safe navigation) Data Acquisition

Sensing (Multi-view Capture)

Computing (Edge Inference)

Communication (Feature Delivery)

Storage (Map Caching)

Intelligence (Taskoriented Representation)

Precise Localization

Task-oriented Orchestration Algorithm

Fig. 1. A motivating example of a mission-driven LAE task: high-precision UAV navigation in GNSS-denied urban canyons via air– ground collaborative Sensing, Communication, Computing, Storage, and Intelligence (SCCSI) orchestration. The UAV (air side) performs lightweight multi-view sensing and task-oriented feature compression, transmits compact representations over air–ground links, and a ground edge server (ground side) executes computation-intensive inference and map/model-assisted refinement. The resulting navigation state is fed back to the UAV for closed-loop control, illustrating task-driven co-design across Sensing, Communication, Computing, Storage, and Intelligence over the air-and-ground dimensions.

1.1

From application-driven requirements to air–ground collaboration framework

A distinctive feature of LAE is its application-driven heterogeneity. Different LAE missions impose vastly different and often stringent requirements on latency, reliability, safety/compliance, coverage, energy efficiency, and situational awareness. Meanwhile, the LAE operating environment is highly dynamic due to fast-varying 3D mobility, frequent topology changes, and massive spatiotemporal data generation. Similar heterogeneity has been observed in UAV-assisted IoT systems, where diverse application classes exhibit distinct QoS requirement patterns (e.g., latency, reliability, and coverage), thereby motivating task-aware system design [4]. These factors collectively motivate a fundamental paradigm shift: LAE systems cannot be effectively engineered by optimizing a single resource or a single layer in isolation. Instead, they must be designed as mission-centric cyber-physical systems where sensing, communication, computing, storage, and intelligence are jointly orchestrated in a closed loop. Crucially, the above application-driven requirements necessitate air–ground collaboration. Aerial platforms provide unique OTA advantages, rapid deployment, flexible 3D mobility, line-of-sight (LoS) opportunities, and bird’s-eye-view (BEV) sensing, which are invaluable for coverage extension and wide-area perception. However, UAVs are inherently constrained by limited battery energy, payload, and onboard computing/storage constraints; simply “pushing more resources onboard” (e.g., adding GPUs) increases weight and power consumption and is often not practical. In contrast, ground infrastructures (e.g., 5G/6G base stations, cloud/edge computing, and intelligent transportation systems) offer a stable power supply, stronger computing/storage capabilities, and persistent connectivity, yet they often face severe line-of-sight (LoS) obstructions and multipath interference within urban environments, resulting in restricted aerial coverage and awareness. Therefore, LAE service provisioning naturally becomes a two-dimensional (air-and-ground) system design problem: the effective solution is not more resources in one domain, but better orchestration of resources across both domains. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

3

1.2 Motivating example: task-driven air–ground resource synergy To better motivate the aforementioned thinking, we present a motivating example for high-precision and reliable navigation in GNSS-denied urban canyons, where GNSS signals can be severely impaired by blockage and multipath. More broadly, high-performance positioning, navigation, and timing (PNT) is a cornerstone for safe low-altitude operations, and recent studies have discussed space–air–ground collaborative PNT architectures and enabling technologies for LAE [59]. As shown in Fig. 1, in urban canyons, GNSS can fail, while standalone onboard solutions, such as IMU and visual odometry, typically suffer from drift accumulation and substantial real-time computation demands. Although multi-view imagery can improve spatial awareness and localization accuracy, lightweight UAVs often cannot afford high-dimensional visual processing due to computing and energy constraints. To address this challenge, our prior work [29] demonstrated an edge–aerial collaboration paradigm, where a UAV performs lightweight sensing and taskoriented feature extraction/compression, and the ground edge server executes computation-intensive inference and map/model-assisted refinement. This closed-loop workflow enables reliable navigation under stringent uplink and latency constraints. This task-driven design illustrates that a single LAE mission, such as localization and navigation, may inherently require coordinated sensing for multi-view capture, communication for feature delivery, computing for edge inference, storage for localized map caching and data buffering, and intelligence for task-oriented representation and decision making. It also reflects a broader trend: as LAE scales to dense operations, such as logistics swarms, emergency response, and airspace monitoring, SCCSI resources must be jointly managed across air and ground under safety and reliability constraints. This motivates task-driven orchestration algorithms rather than siloed optimization of individual resources. 1.3 Motivations of AG-SCCSI: a task-driven, closed-loop orchestration problem Motivated by the above observations, we focus on Air–Ground integrated Sensing, Communication, Computing, Storage, and Intelligence (AG-SCCSI) as a unified paradigm for LAE. Applications such as urban air mobility, emergency rescue, and large-scale drone logistics require ultra-low latency, ultra-high reliability, and pervasive availability of SCCSI resources across vast, dynamic three-dimensional (3D) airspace [99]. Moreover, safety is a first-order requirement for dense low-altitude operations. Systematic evidence on UAV operational safety assessment further underscores the need for safety-aware orchestration and verifiable decision pipelines [8]. However, neither aerial platforms nor terrestrial systems alone can sustainably meet these mission-level QoS and safety requirements. UAVs excel at flexible sensing and agile connectivity but remain energy- and payload-constrained with volatile links under mobility [27]. Terrestrial systems provide stronger and steadier compute/storage and more stable connectivity, but face limited aerial coverage and reduced responsiveness to aerial dynamics [45]. This complementarity suggests a fundamental design principle: Application-driven LAE requirements necessitate air–ground collaboration, which in turn requires task-driven, closed-loop orchestration of SCCSI resources across the air-and-ground dimensions. Unfortunately, this design also poses several system-level challenges:

• Resource heterogeneity and abstraction: Air and ground nodes differ in mobility, energy models, compute/storage budgets, sensing modalities, and communication interfaces. Effective collaboration requires unified abstractions and interoperable orchestration interfaces. Manuscript submitted to ACM

4

Y. Deng et al.

• Mobility-induced volatility: UAV mobility causes rapid topology changes and fluctuating resource availability, demanding online association, adaptive service migration, and resilient task offloading. • Coupled trade-offs across SCCSI : Offloading reduces onboard energy consumption but incurs communication overhead and handover costs; caching reduces latency but competes with onboard storage; sensing/reporting improves situational awareness but increases communications and computing workloads. These coupling effects require joint, task-driven optimization. • Security, safety, and compliance: Hybrid air–ground operation introduces threats such as spoofing, jamming, and data poisoning, and mandates trustworthy coordination mechanisms and auditable data/decision pipelines. • Scalability in dense operations: Managing thousands of aerial/ground agents in urban airspace requires scalable, lightweight, and distributed orchestration architectures. Existing air–ground integration often treats UAVs as passive relays or views ground infrastructure as a simple backhaul for communication and offloading. Such designs underutilize the mutual complementarity. In LAE, UAV BEV sensing can easily augment ground perception in occluded environments, while ground edge/cloud can sustain intensive computing and storage (e.g., map/model/service caching and compliance logging). These bidirectional synergies motivate AG-SCCSI as a unified framework, enabling both vertical collaboration (e.g., air–ground task offloading, ground-assisted navigation, and OTA advantages)) and horizontal collaboration (e.g., UAV swarm sensing, edge federation among ground nodes), ultimately providing adaptive and mission-tailored service provisioning for LAE [20, 21, 55]. 1.4

Related Works and Our Contributions

Before presenting our AG-SCCSI framework, we compare our work with existing surveys and vision papers. Recent papers have advocated low-altitude aerial intelligent networks (LAIN) to support large-scale LAE services. However, how to translate heterogeneous mission requirements into task-driven, closed-loop orchestration across SCCSI resources remains underexplored [42].

• Space-Air-Ground Integrated Networks (SAGIN): Research in this area, such as Xiao et al. [95], often explores broad architectural paradigms for global connectivity, leveraging satellites, aerial platforms, and terrestrial networks for communication. While SAGIN aims for ubiquitous communication coverage, its primary focus is to address the limitations of single networks by integrating heterogeneous segments (space, air, and ground) that often operate on different communication media and network protocols, leading to a “bucket effect” where overall system capability is limited by the weakest part. This body of work, including discussions on SAGIN integration by Wang et al. [87], provides a foundational vision on integrated networks. However, its scope is typically too macroscopic, emphasizing extensive communication coverage and traffic sharing/offloading across vast geographical scales, often overlooking the unique, fine-grained microscopic demands of agile and resourceconstrained applications, such as urban drone logistics or emergency response in LAE. Thus, SAGIN alone cannot address the challenges in LAE where coordinated AG-SCCSI resources are needed. • Intelligent Network Paradigms and Semantic Communications: Another strand of research focuses on infusing intelligence into communication networks. Examples include integrating advanced AI models (e.g., large language models and diffusion models) for intelligent, task-driven networking in SAGIN [31], the use of generative AI to assist network automation and management [103], and emerging paradigms of semantic communication that transmit semantic contexts instead of raw bits [63]. These works promote task-oriented, AI-driven networking to enhance efficiency and adaptability. Nevertheless, their discussions remain mostly either conceptual or Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

5

high-level, focusing on general automation or computing improvements. They have not explicated how “intelligence” can be tightly coordinated with sensing, communication, and computing resources under the dynamic conditions of LAE. In particular, the unique challenges of LAE (e.g., high mobility, strict resource constraints, cross-domain coordination, (near) real-time missions) are not directly addressed by these works. • Integrated Sensing and Edge AI (ISEA): After collecting data, intelligence at the edge can be extracted immediately. Thus, ISEA was proposed and investigated lately. Notably, Jiang et al. introduced ISEA as a new paradigm for 6G intelligent perception [56]. ISEA deeply integrates the acquisition, processing, and transmission of sensing data with edge AI, featuring a holistic co-design of sensing, communication, and computation for intelligence. This represents a significant step toward functional convergence of sensing, communication, and computing. However, ISEA is geared toward general 6G use-cases and lacks LAE-specific design considerations. It has not fully explored the tight integration of all five SCCSI dimensions, notably missing explicit roles of storage (e.g., caching for computation offload) and distributed intelligence (e.g., federated learning across air/ground nodes), under high mobility and stringent resource constraints of LAE missions. • Low-Altitude Wireless Networks (LAWN) and LAE-Native Network Architectures: Recent works have started to examine LAE from a more native networking perspective, beyond generic UAV communications or broad SAGIN abstractions. In particular, Yuan et al. introduce LAWNs as a reconfigurable three-dimensional layered architecture that tightly integrates connectivity (communications), sensing, control, and computing across aerial and terrestrial nodes. In parallel, Wu et al. provide a comprehensive survey of LAWN systems, covering functional evolution, performance metrics, privacy and security, and airspace architecture. These works substantially enrich the architectural understanding of LAE-native networks. However, they mainly emphasize communication, control, and airspace-level integration, and have not explicitly formulated a mission-centric closed-loop orchestration framework that jointly coordinates SCCSI across air and ground nodes [98]. • LAE Network Architecture, Standards, and Integrated Technologies: Recent studies have also started to examine LAE from the perspective of native network architecture and system realization. In particular, Wang et al. [87] discuss the core architecture, related standards, and integrated technologies of LAE networks, covering communication, sensing, computing, positioning, navigation and surveillance, flight control, and airspace management within a unified framework. Compared with generic UAV networking or broad SAGIN abstractions, this line of work provides a more direct system-level view of how LAE networks can be realized in practice. However, its main focus remains on architecture realization, standards support, and multi-technology integration, rather than on a mission-centric closed-loop orchestration framework that explicitly coordinates SCCSI across air and ground nodes [89]. In summary, while numerous studies address subsets of the SCCSI dimensions (e.g., ISAC for sensing-communication, MEC for computing offload) or propose broad architectures and integrated technology roadmaps (e.g., SAGIN, LAWN, and LAE-native network realization frameworks) for next-generation networks, there is a persistent gap in the literature: no existing work provides a holistic survey or framework that systematically defines and integrates all five SCCSI dimensions in a unified air-ground paradigm tailored for LAE. This gap is critical, since LAE’s distinctive requirements for agility, resource efficiency, timely decision making, and cross-domain coordination can only be met with a mission-critical holistic framework that transcends generic architectures or partial integrations. To our best knowledge, this is the first survey to systematically define and explore the AG-SCCSI framework as a unified paradigm specifically for LAE. To visually underscore the novelty and comprehensiveness of our work, we Manuscript submitted to ACM

6

Y. Deng et al. Table 1. Comparison of Survey and Vision Papers on Air-Ground Integration and SCCSI-Related Technologies. Air-Ground Coordination?

LAESpecific?

Reference

Focus Area

Dimensions Covered

[95]

LLM, Diffusion & DT for SAGIN

Sens., Comm., Comp.

[103]

LLM-Driven SAGIN

Sens., Comm.

[91]

ISCC in 6G

Sens., Comm., Comp.

[105]

AG-MEC for Task Offloading

Comm., Comp.

×

[92]

AI-Driven Deployment Automation

Comm., Comp., Intell.

×

[56]

ISEA in 6G

Sens., Comm., Comp., Intell.

[87]

SAGIN

Sens., Comm., Comp., Intell.

[33]

Semantic Communication for 6G

Sens., Comm., Intell.

[63]

Semantic-Enabled SAGIN

Sens., Comm., Comp., Intell.

[79]

UAV-aided ITS

Sens. (Visual), Comm., Planning

[45]

ISAC for LAE

Sens., Comm.

[98]

Low-altitude wireless networks

[89]

LAE network architecture

(Multi-layer) (Multi-layer) (Implicit)

(Implicit)

× × ×

×

(General NTN) (Implicit)

×

(General NTN) (Sub-domain)

Sens., Comm., Comp.

(+Ctrl./ATM)

Sens., Comm., Comp., Intell.

(+PNT/ATM)

[15]

VaaS for Smart Cities

Sens., Comm., Comp., Storage, Intell.

This Work

AG-SCCSI for LAE

All SCCSI

×

×

provide a detailed comparison with existing survey papers in Table 1. Our main contributions are summarized as follows.

• Holistic AG-SCCSI Framework: We propose a novel AG-SCCSI framework, which cohesively unifies multi-dimensional SCCSI resources in 3D space across diverse air and ground platforms. Unlike prior architectures like SAGIN (focused on communication connectivity scale) or ISEA (focused on sensing–AI fusion), AG-SCCSI is explicitly designed for LAE, enabling seamless and efficient coordination of all five SCCSI resource dimensions. This framework directly addresses the interplay among sensing data, wireless links, computing tasks, cached contents, and intelligent navigation and control in air–ground networks. • Comprehensive Survey of Enabling Technologies: We conduct a thorough survey of key enabling technologies that underpin AG-SCCSI, including Integrated Sensing and Communication (ISAC), Air–Ground Mobile Edge Computing (AG-MEC), Digital Twin (DT) modeling, Federated Learning (FL), and Semantic Communications. Our analysis goes beyond generic overviews by focusing on each technology’s specific role and synergies in air– ground integration for LAE. We highlight how these technologies interconnect within the AG-SCCSI framework, providing insights not covered in existing surveys that consider them mostly in isolation. • Concrete LAE Application Scenarios and Validation: We present multiple application scenarios in the LAE application domains, such as emergency response, intelligent traffic management, and urban drone delivery logistics, Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

7 Table 2. List of Abbreviations

Abbreviation

Description

Abbreviation

Description

ABS AG-MEC

Aerial Base Station Air-Ground Multi-access Edge Computing

ADMM AG-SCCSI

AirComp AoI BCD BS CEN C-SLAM DRL DSA FL GNSS GV HFL IMU ISCC KG LAE LoS MAPPO MCTS mD OTA PLS PS QoS RIS SA SAGIN SDN SLP STAR-RIS UAV VPS

Over-the-Air Computation Age of Information Block Coordinate Descent Base Station Cache-Enabled Network Cooperative Simultaneous Localization and Mapping Deep Reinforcement Learning Dynamic Spectrum Access Federated Learning Global Navigation Satellite System Ground Vehicle Hierarchical Federated Learning Inertial Measurement Unit Integrated Sensing–Communication–Computing Knowledge Graph Low-Altitude Economy Line of Sight Multi-Agent Proximal Policy Optimization Monte Carlo Tree Search Multi-Dimensional (links/metrics) Over-the-Air Physical-Layer Security Parameter Server Quality of Service Reconfigurable Intelligent Surface Semantic Attention Space-Air-Ground Integrated Network Software-Defined Networking Symbol-Level Precoding Simultaneously Transmitting & Reflecting RIS Unmanned Aerial Vehicle Visual Positioning System

AoD AP BEV CDT CLP CPN DT ES FMCW GPS HAP HFRL ISAC ITS KKT LiDAR MAB MARL MEC NLoS OTA-FL PPO QoE RF RSU SAoI SCA SemCom SNR UAS UWB VSP

Alternating Direction Method of Multipliers Air-Ground Sensing, Communication, Computing, Storage & Intelligence Angle of Departure Access Point Bird’s-eye View Cooperative Digital Twin Cooperative Long-Term Average Optimization Computing Power Network Digital Twin Edge Server Frequency-Modulated Continuous Wave Global Positioning System High-Altitude Platform Heterogeneous Federated Reinforcement Learning Integrated Sensing and Communication Intelligent Transportation Systems Karush-Kuhn-Tucker Conditions Light Detection and Ranging Multi-Armed Bandit Multi-Agent Reinforcement Learning Multi-access Edge Computing Non-Line of Sight Over-the-Air Federated Learning Proximal Policy Optimization Quality of Experience Radio Frequency Roadside Unit Semantic-Aware Age of Information Successive Convex Approximation Semantic Communication Signal-to-Noise Ratio Unmanned Aircraft System Ultra-Wideband Virtual Service Provider

and analyze how the AG-SCCSI framework elevates service capabilities and operational efficiency in each case. Through these examples, we illustrate the practical benefits of tightly coordinated SCCSI resources (e.g., faster situational awareness, improved reliability, real-time decision making). Furthermore, we discuss potential validation methodologies (e.g., prototyping, simulation testbeds, and field trials) for these scenarios, bridging the gap between theoretical framework and real-world deployment.

• Future Research Directions: We identify several open challenges and promising future research directions to advance AG-SCCSI in LAE. These include developing semantic communication techniques for efficient missioncritical information exchange, exploring LLM-driven UAV swarm control for enhanced autonomy, and innovating sustainable energy management to prolong aerial operations. By highlighting these pressing issues and opportunities, we aim to guide and inspire future research efforts in building intelligent, robust, and sustainable LAE systems. Manuscript submitted to ACM

8

Y. Deng et al. SCCSI Aircrafts

Multi-hop Transmission ... Swarms Air-ground Storage & Intelligence

Air-ground Sensing & Communications

Charging Station

No-Fly Zone

Terristrial Edge Server

s Di Air-ground Cargo Delivery

on uti trib

r nte Ce

f& -of ke Ta

ing

P

for lat

les hic Ve I S

m C SC

nd La

Air-ground Computing

Fig. 2. Scenario illustration of collaborative air-ground sensing, communication, computing, storage, and intelligence in the lowaltitude economy, featuring no-fly zones, charging stations, and other key elements, along with various applications requiring different air-ground SCCSI resources.

1.5

Organization

The rest of this paper is organized as follows. Section 2 introduces the AG-SCCSI framework and the associated airground resource orchestration mechanisms. Section 3 surveys the enabling technologies that support integrated AGSCCSI in LAE. Section 4 reviews representative datasets, testbeds, digital twins emulation, simulators, and evaluation metrics for AG-SCCSI validation. Section 5 presents representative application scenarios. Section 6 discusses open challenges and future research directions. Section 7 concludes the paper. For convenience, Table 2 summarizes the abbreviations used throughout the paper. 2 AG-SCCSI Framework and Resource Orchestration Mechanisms In this section, we present the AG-SCCSI framework and the main resource orchestration mechanisms for LAE applications. We emphasize the vertical collaborations across air and ground for each SCCSI element and the corresponding orchestration mechanisms that together constitute our AG-SCCSI framework. 2.1

AG-SCCSI Framework

As shown in Fig. 2, the AG-SCCSI framework comprises UAVs, ground edge servers (ESs), base stations (BSs)/access points (APs)/roadside units (RSUs), and SCCSI-capable ground vehicles (GVs). UAVs serving LAE missions (e.g., parcel, medical, and food delivery) form OTA mobile networks, providing OTA sensing and communications, while ground Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

9

nodes (ESs, BSs/APs/RSUs, and GVs) contribute ground sensing, communications, and computing. Together they compose a 3D service network with 3D SCCSI resources for LAE missions and smart-city operations and services [2, 15]. Analogous to ground SCCSI service networks [2, 15], mobile things carry SCCSI resources via application-specific Points of Connection (PoCs)2 . Different from ground deployments [2, 15, 23, 24], airborne platforms face strict payload and energy limitations, and hence PoC customization must make tradeoff between capability and weight/power, affecting airframe mechanics and mission endurance. A second fundamental difference from legacy communication frameworks lies in supported services: AG-SCCSI links are not only for message passing but for ultimate mission fulfillment. For instance, OTA links may be established for OTA sensing to guide UAV navigation during delivery, or to discover remote ESs to complete offloaded computing tasks [21]. Consequently, AG-SCCSI must manage multi-dimensional (mD) links between node pairs under heterogeneous physical and logical constraints, with objectives defined by mission semantics rather than pure data throughput. This 3D task-oriented perspective complements the macroscopic coverage-centric view of SAGIN and is essential in LAE where coordinated SCCSI is the bottleneck rather than mere connectivity [95]. 2.2 AG-SCCSI Network Formation and Management Beyond generic SAGIN formation [95], AG-SCCSI forms mission-specific topologies with mD link metrics. For example, for UAV navigation and perception, candidate links must jointly satisfy perception-quality and communication-quality constraints so that an ego UAV can select neighbors whose sensed views support robust fusion3 [29]. Hypergraph abstractions naturally capture such mD relationships. Given high mobility (e.g., for UAVs and GVs), fading, and opportunistic ground computing, the control plane must proactively learn network/resource states and perform low-latency decision making. We advocate a software-defined network (SDN) design approach with an intent-driven controller that fuses measurements and predictions (e.g., via a digital twin, DT) for online what-if evaluation and look-ahead orchestration. Emerging O-RAN DT studies and industry reports show DTs improve RAN monitoring, policy testing and closed-loop optimization, making them suitable for AGSCCSI state tracking and “plan–do–check–act” control [65, 67]. 2.3 Orchestrating Air–Ground Sensing Sensing in 3D urban environments benefits from the additional degrees of freedom (DoFs) offered by aerial platforms. Cooperative SLAM and air–ground collaboration (AGC) allow UAVs to complement ground robots or vehicles in occluded, congested, or otherwise unreachable areas. As illustrated in Fig. 3, a robust AGC system typically involves parallel processing pipelines, where both aerial and ground nodes perform pre-processing, including time synchronization and calibration, as well as front-end SLAM for keyframe extraction. To align these independent trajectories, the system relies on cross-view data association, such as feature or map matching, and back-end joint pose graph or factor graph optimization. This tightly coupled fusion not only builds a globally consistent map, but also directly supports downstream exploration, path planning, and navigation tasks. Recent multi-robot datasets, such as GRACO, provide synchronized multi-modal aerial/ground data with precise calibration for GPS-available and GPS-denied validation [108]. New air–ground perception benchmarks and methods 2

A PoC is a conceptual device with customized resources or capability for SCCSI services [2, 15]. This is analogous to collaborative vision fusion for vehicles on the ground, but UAV images are often offloaded to an ES on the ground for perception fusion due to onboard computing limitations.

3

Manuscript submitted to ACM

10

Y. Deng et al.

UAV Sensors (Camera/LIDAR/Radar)

Pre-processing (Time sync, calibration)

UAV Front-end SLAM (Keyframe selection)

UAV Local Map & Keyframes

Data Association & Relative Localization (Map/Feature Matching)

Map Fusion (Ground Global Map + UGV Local Map)

UAV

Radar LIDAR Camera

Path Planning Joint Pose Graph/ Factor Graph Optimization

Exploration Navigation

Onboard Sensors (LIDAR/rgb-D Camera)

Pre-processing (Time sync, calibration)

Front-end SLAM (Keyframe selection)

Local Map & Keyframes

Fig. 3. Air-Ground Sensing Coordination: Simultaneous Localization and Mapping

further indicate that fusing aerial BEV observations with ground views can substantially improve detection and tracking under occlusion. However, communication cost and UAV altitude remain critical design factors: intermediate fusion often achieves a favorable accuracy–overhead trade-off, while perception performance may peak at moderate altitudes, e.g., around ∼25 m, due to scale and depth-estimation limits [28]. Large-scale real-world aerial–ground datasets and unified benchmarks are also emerging to quantify these trade-offs more systematically [82]. Beyond passive data collection, recent survey work on intelligent active UAV surveillance and monitoring shows that future UAV systems are evolving toward active, predictive, and autonomous sensing agents. This trend is highly consistent with the AG-SCCSI view, where air–ground sensing should be embedded into a closed loop with communication, computing, and decision making [5]. With rapidly growing aerial spectrum use, air–ground collaborative spectrum situation awareness also becomes critical for interference mitigation, dynamic spectrum access (DSA), and electromagnetic security. Recent work on 3D radio-environment mapping (REM) leverages sparse Bayesian methods to construct site-specific 3D maps with shadowing information, enabling interferer localization and policy optimization [83]. UAV-borne sensors can further provide angular diversity for 3D anomaly localization, while multi-angle “soft–then–hard” fusion improves detection confidence when single-sensor coverage is limited. Such REM/DT coupling can support angle-dependent access by admitting secondary users in non-interfering angular sectors relative to primary users, thereby improving spectrum utilization without violating protection constraints. 2.4

Orchestrating Air–Ground Communication

UAVs and HAPs can act as aerial BSs (ABSs) 4 or relays for coverage restoration/boosting for some LAE applications. As illustrated in the left panel of Fig. 4, this creates a joint air–ground deployment architecture where ground base stations (GBSs) and ABSs coordinate under limited wireless backhaul constraints. A key practical challenge is channel4

It is debatable whether UAVs can be used as BSs due to the weight addition and potential downlink interference to existing cellular services. Management hassle may prevent this option from practical deployment if cellular systems are considered in LAE. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

11 Robustness & Efficiency Enablers

ABSs

Platform Motion Mitigation Cellular Downlink/ Uplink ABS Service Link Core Network

Multi-Beam Beamforming

Placement Optimization based on Channel Map

Fiber/Wire Backhaul

Security Mechanisms for LAE

Emergency Logistics

UAV Link

High-LoS Eavesdropping Risk

GBS

Users Public Safety (Police Drone)

Robust schemes for angular uncertainty and platform jitter

STAR-RIS (Simultaneously Transmitting And Reflecting Reconfigurable Intelligent Surface)

Beamforming

Optimized QoS Requirements

Multi-User NOMA

Fig. 4. Air-Ground Communication Coordination: Coverage & Security

and constraint-aware 3D placement of multiple ABSs in cities with limited backhaul and no-fly zones. Channel-map (radio-map)–based ABS placement has been shown to guarantee minimum rates to all users by exploiting site-specific propagation instead of distance-only models [73]. Beamforming with multi-antenna UAVs improves spectral/energy efficiency but is sensitive to platform motion; robust schemes that account for angular uncertainty and platform jitter (see Fig. 4, top right) are recommended in practice. Finally, secure LAE communications must consider high-LoS eavesdropping risk. Covert/secret transmissions with cooperative NOMA and optimization of active/passive beamforming, including STAR-RIS aiding UAV links, are promising to increase covert transmission rates under wardens [84, 96]. ABS deployment should jointly optimize horizontal/vertical positions, backhaul feasibility, and interference coupling. As depicted in the bottom-center panel of Fig. 4, radio-map–driven multi-ABS placement in real maps has matured into tractable convex-relaxation solvers with linear scaling in user count [73]. For LAE’s sensitive traffic (e.g., for public safety and healthcare logistics), recent works explore collaborative secret–covert access (multi-user NOMA) and RIS/STAR-RIS–assisted covert UAV links (Fig. 4, bottom right) to suppress detectability while meeting QoS requirements [84, 96]. Beyond link-level security, mission-driven orchestration is essential for sophisticated LAE tasks. In our recent work [61], we propose an interest-aware BEV multicasting for infrastructure-assisted collaborative perception. By accounting for individual maps of interest, we formulate a joint feature selection and multicast grouping problem to maximize network-wide utility under communication constraints. 2.5 Orchestrating Air–Ground Computing Multi-access edge computing (MEC) offloading cuts down latency and energy consumption; however, terrestrial-only deployments often fail to catch up with the spatio-temporal dynamics of LAE. To address this, an air–ground collaborative MEC (AGC-MEC) architecture, as illustrated in the left panel of Fig. 5, unifies aerial servers (e.g., UAVs, airships, balloons) and terrestrial servers (e.g., base stations, APs, RSUs, and ground vehicles) into a cohesive edge computing continuum. Compared with single-domain MEC, AGC-MEC is both more flexible and more powerful for supporting dynamic edge intelligence and task offloading [105]. However, orchestrating such an air–ground synergy requires solving complex resource allocation problems [18]. As depicted in the right panels of Fig. 5, recent works further address joint multi-objective optimization, encompassing Manuscript submitted to ACM

12

Y. Deng et al. Joint Multi-Objective Optimization Module

ABSs

1. Aerial Trajectory Cellular Downlink/ Uplink ABS Service Link

2. Task Offloading Decisions

4. Backhaul Management

3. Computing Resources

RSUs

KPI Improvements

Core Network

After Before

Latency (↓)

GBS Baseline Reliability (↑)

Users Public Safety Emergency (Police Drone) Logistics Mobile Server

Energy Consumption (↓)

Proposed AGC-MEC (↑)

Resource Utilization (↑)

Fig. 5. Air-Ground Computing Coordination: multi-access edge computing continuum

Joint Optimization

Cache-enabled UAVs

3D UAV Deployment LoS Short-cut (Multicast Delivery)

GBS

PopularityAware Caching

User Association

Network Backhaul Load

# of Multicast Channels

User QoE

Throughput Gain

Transmission Reliability

Impact Analysis

Backhaul Fetch Limits

Fig. 6. Air-Ground Storage Coordination: edge caching

UAV trajectory planning, task offloading decisions, computing resource allocation, and backhaul management. Moreover, integrating joint caching–offloading strategies with reliability constraints in UAV-enabled MEC has demonstrated significant reduction in latency and energy for realistic traffic scenarios [73, 76]. 2.6

Orchestrating Air–Ground Storage

Edge caching reduces backbone and backhaul traffic load while minimizing tail latency for recurrent popular contents. UAVs, thanks to their low cost and high mobility, can serve as cache-enabled aerial nodes for temporary hot spots (e.g., football stadiums), significantly improving quality of experience (QoE) via line-of-sight (LoS) short-cuts. As illustrated in the left panel of Fig. 6, orchestrating this air–ground storage involves a strategic split between locally cached content delivery and remote backhaul fetching. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

13 UAV PS Optimization Loop

UAV as a dynamic parameter server (PS)

AirComp • Compressed Uplink Rounds • Lower Latency • Privacy Reservation

SNR Performance

Waveform Superposition

Computing Performance

ISEA Paradigm Sematic-Aware Sensor Selection

Camera LiDAR

Heterogeneous Devices

IoT Sensors

1. UAV Trajectory 2. Device Scheduling 3. Power Consumption

UAV path planning

Mobile Devices

RF

Multi-Agent Sensing

Edge Machine Learning ML Model Training

Fig. 7. Air-Ground Intelligence Coordination: AirComp-based Federated Learning

Large-scale analyses that include interference show that popularity-aware (PA) multicast scheduling in cache-enabled UAV networks increases the number of successfully served users. As depicted in the right panels of Fig. 6, optimizing the numbers of multicast channels and backhaul fetched limits is the key to the design under high signal-to-interference ratio (SIR) thresholds [34]. Furthermore, joint 3D UAV deployment, user association, and caching design collectively form a tightly coupled optimization problem, which further reduces the average access delay in multi-UAV cache-enabled networks [101]. 2.7 Orchestrating Air–Ground Intelligence To navigate over the air safely, the LAE requires on-site learning while preserving privacy. Federated learning (FL) avoids raw data sharing for privacy preservation, but it often suffers from uplink upload overhead and stragglers (links or computing nodes) for model updates. As illustrated in the left panel of Fig. 7, OTA computation (AirComp) leverages waveform superposition to enable efficient uplink aggregation, while semi-/asynchronous AirComp-FL further mitigates stragglers. Recent results show lower latency and improved round efficiency compared to orthogonal schemes [94, 107]. Furthermore, UAVs can serve as agile parameter servers. As depicted in the top-right panel of Fig. 7, these aerial servers can be dynamically repositioned to balance communication and computing performance, thereby improving fairness and convergence. Energy-efficient UAV-FL jointly optimizes the UAV trajectory, device scheduling, and power consumption [30]. Beyond pure FL, the integrated sensing–edge-AI (ISEA) paradigm tightens the loop between sensing and edge ML (see Fig. 7, bottom right). Several existing surveys and papers have already identified AirComp-based feature fusion and semantic-relevance-aware sensor selection as key enablers for scalable multi-agent perception [56– 58]. 3 Enabling Technologies for Integrated AG-SCCSI While Section 2 establishes the AG-SCCSI framework and identifies the main resource orchestration mechanisms across SCCSI resources, this section focuses on the enabling technologies that make such orchestration practically Manuscript submitted to ACM

14

Y. Deng et al. Trust & Security

Decentralized Intelligence

Computing Coordination

Security & Distributed Intelligence Trust & Blockchain

Secure RIS Reflection Surface

Dynamic Topology & Mobility

Network & Topology Management Networking Layer

Agile UAV Mobility Trajectories

Adaptive Wireless Links

Data & Resource Heterogeneity

Over-the-Air Computation

AG-SCCSI Core

Application Layer

Federated Learning

Edge Servers

UAVs

Mathematical Optimization

Scalability in Large Networks

Hierarchical Topology

Multi-agent Swarm

Energy Efficiency

Physical & Resources Constraints Foundation Layer

Heterogeneous Nodes

Non-IID Data Distribution

Battery

Energy

Optimized UAV Flight Paths

Fig. 8. A multi-layered taxonomy of enabling technologies for the AG-SCCSI framework, illustrating the synergistic evolution from foundational resource management and dynamic networking to secure, decentralized intelligence in LAE.

realizable in LAE. Our emphasis is not on isolated techniques, but on cross-layer and cross-domain enablers that jointly shape the dynamic behavior of SCCSI resources over air and ground nodes. As illustrated by the multi-layered taxonomy in Fig. 8 and the six-pillar summary in Table 3, several recurring technology families play central roles in AG-SCCSI, including integrated sensing and communication (ISAC), radiomap/digital-twin (DT) driven orchestration, air-ground mobile edge computing (AG-MEC), federated/OTA learning (FL/AirComp), semantic communication (SemCom), and reconfigurable intelligent surfaces (RIS/STAR-RIS). These pillars provide a complementary technology taxonomy. However, because practical LAE missions should be accomplished under multiple intertwined system constraints, the discussion below is organized along seven cross-cutting challenges: heterogeneity, topology dynamics, energy-performance tradeoffs, trust and security, scalability, decentralized intelligence, and large-scale computing coordination. In practice, each challenge is typically addressed by multiple technology pillars operating jointly in a closed loop. 3.1

Addressing Data and Resource Heterogeneity through Cross-Layer Integration

Heterogeneity is one of the most fundamental obstacles in AG-SCCSI. Air and ground nodes differ significantly in sensing modality, computing capability, energy budget, mobility pattern, communication quality, and data distribution. In LAE, such heterogeneity is further amplified by mission diversity and topology volatility, which makes collaboration substantially more difficult than in conventional terrestrial systems. As a result, heterogeneity cannot be handled at a single layer alone; instead, it requires cross-layer coordination across sensing, communication, computing, and learning. 3.1.1

Mitigating Sensing-Induced Data Heterogeneity in FL. In UAV-assisted federated learning, dynamic flight paths

and non-uniform sensing conditions lead to highly uneven data acquisition, thereby aggravating non-IID data distributions across clients. Tang et al. [78] propose an ISAC-enabled framework in which UAV sensing outcomes directly influence the quality and diversity of collected training data. By introducing a probabilistic sensing model that links Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

15

Table 3. Taxonomy of enabling technologies for mission-centric AG-SCCSI in low-altitude economy. Enabling Pillar

Main SCCSI Dimensions Affected

Integrated Sensing and Communication (ISAC)

Primary Control / Design Variables

Typical Timescale

Representative LAE Tasks

Main Coupling With Other Pillars

Main Limitations / Open Issues

Representative Refs.

Sensing, Com- Waveform design, munication beamforming, bandwidth split, sensing– communication resource sharing

Fast to medium

Obstacle awareness, navigation, spectrum sensing, environment perception

Strongly coupled with RIS, DT, SemCom, and FL

Sensing– communication tradeoff, mobility robustness, hardware constraints, waveform complexity

[2], [51], [54], [56]

Radio-Map / Digital-Twin (DT) Driven Orchestration

Sensing, Com- Radio-map munication, construction, state Intelligence prediction, what-if evaluation, topology planning, intent-driven control

Medium to long

ABS deployment, interference-aware access, mission planning, predictive orchestration

Coupled with ISAC, AG-MEC, RIS, and applicationlevel scheduling

DT fidelity, model update cost, domain gap, online synchronization, scalability

[28], [29], [34], [35]

Air-Ground Mobile Edge Computing (AG-MEC)

Computing, Communication, Storage

Task offloading, server selection, trajectory planning, compute allocation, backhaul management

Medium

Edge inference, low-latency task execution, collaborative navigation, emergency processing

Coupled with caching, DT, SemCom, and FL

Joint optimization [23], [38], complexity, volatile [39] connectivity, heterogeneous compute resources

Federated / Over-the-Air Learning (FL / AirComp)

Intelligence, Communication, Computing

Client selection, Medium to aggregation long scheduling, trajectory adaptation, OTA aggregation, model placement

Distributed perception, on-site learning, collaborative model updating, privacy-preserving intelligence

Coupled with ISAC, AG-MEC, SemCom, and trust mechanisms

Non-IID data, stragglers, communication overhead, trust/privacy risks

[42], [43], [44], [48], [49], [52], [53]

Semantic Communication (SemCom)

Communication, Semantic extraction, Intelligence, relevance-aware Storage compression, task-oriented source coding, feature transmission

Visual navigation, mission reporting, feature delivery under tight uplink constraints

Coupled with AG-MEC, FL, and ISAC

Task dependence, semantic distortion measurement, cross-model compatibility, benchmark scarcity

[13], [14], [31], [32], [45]–[47]

RIS / STAR-RIS

Communication, Phase-shift control, Fast to Sensing, environment shaping, medium Security passive beamforming, propagation reconfiguration

Coverage enhancement, covert transmission, mobility-resilient links, secure ISAC

Coupled with ISAC, DT, and communication planning

CSI acquisition, hardware control overhead, deployment practicality in LAE

[37], [54], [63]–[66]

Fast to medium

UAV deployment to data collection success, they show that sensing-side randomness can significantly affect FL convergence and that such effects can be mitigated through joint optimization of UAV deployment and bandwidth allocation. This result is important for AG-SCCSI because it shows that data heterogeneity in LAE is not only a statistical learning issue, but also a physical-layer and mobility-induced problem. More broadly, the challenges of statistical, system, and communication heterogeneity in aerial federated learning are surveyed in [69], which discusses how limited battery, unstable links, and unequal sensing opportunities further worsen imbalance across participating UAVs. The broader lesson is that FL design in LAE should not treat data heterogeneity as a fixed input; rather, it should actively reduce heterogeneity through sensing-aware trajectory design, adaptive communication scheduling, and resource-aware participation control. Manuscript submitted to ACM

16

Y. Deng et al.

3.1.2

Unified Waveform Design for Balanced ISAC Performance. Heterogeneity also appears at the functional level:

some nodes may be communication-capable but sensing-weak, while others suffer from limited waveform flexibility or hardware capability. Traditional communication-centric signals such as OFDM are often poorly matched to sensing requirements because of strong sidelobes and imbalanced sensing-communication performance. Li et al. [51] address this issue through symbol-level precoding (SLP) for MIMO-OFDM ISAC waveform design, suppressing range-Doppler sidelobes while preserving communication QoS. Their work shows that waveform-level co-design can reduce functional disparity across heterogeneous nodes and enable even lightweight aerial platforms to contribute meaningfully to both sensing and communication tasks. 3.1.3

Hierarchical Aggregation to Alleviate Non-IID Effects. When node capability and data quality vary substantially,

architectural designs are often needed in addition to physical-layer or scheduling optimizations. Li et al. [52] propose a hierarchical FL framework in which local groups first aggregate models to reduce intra-group non-IID effects, after which a higher-level aggregator performs global fusion. Such a two-tier design improves robustness against weak or poorly performing clients and reduces the sensitivity of global learning to node-level disparities. In AG-SCCSI, this type of hierarchical architecture is particularly useful because it aligns naturally with air-ground multi-tier resource structures, where UAV clusters, roadside infrastructure, and edge servers may play different aggregation roles. Overall, heterogeneity in AG-SCCSI should be treated as a coupled sensing-communication-computing-learning problem. Physical-layer design, client scheduling, and hierarchical learning architectures are therefore complementary rather than competing. 3.2

Managing Dynamic Topologies via Adaptive Sensing and Mobility Control

Unlike static terrestrial systems, AG-SCCSI operates over highly dynamic topologies driven by UAV mobility, intermittent links, and rapidly changing mission geometry. As a result, topology management in LAE is inseparable from sensing, mobility control, and communication adaptation. This requirement is closely related to UAV path-planning research, where a recent survey paper has reviewed sampling-based, potential-field, bio-inspired, and AI-based planning methods and analyzed their applicability under different environmental and operational constraints [48]. However, AG-SCCSI goes beyond standalone path planning because mobility decisions must be jointly coordinated with sensing quality, link stability, computing availability, and mission-level service requirements. The enabling technologies in this subsection, therefore, focus on how to maintain robust collaboration under time-varying air-ground connectivity. 3.2.1

Trajectory-Aware Data Acquisition for Reliable FL. Fu et al. [30] study UAV-assisted federated learning under

dynamic air-ground links by jointly optimizing UAV trajectory, device scheduling, and communication-computing resources. Their formulation derives scheduling constraints from FL convergence analysis so that only sufficiently reliable devices are selected to participate in each round. This highlights a key design principle for AG-SCCSI: in dynamic topologies, learning reliability depends not only on the FL algorithm itself, but also on mobility-aware and/or reliability-aware participation selection. 3.2.2

Integrated Sensing for Environmental Awareness. ISAC further supports topology management by enabling UAVs

to perceive their environment while communicating. By exploiting reflected communication signals, UAVs can perform real-time obstacle detection, terrain awareness, or environmental monitoring, and then adapt their trajectories accordingly [45]. This creates a closed sensing-mobility loop in which communication infrastructure is not merely a data pipe but also part of the situational-awareness control mechanism. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

17

3.2.3 RIS-Enhanced Channel Stability. Dynamic topology also creates unstable channels due to blockage, angular fluctuation, and UAV motion jitter. RIS provides an additional degree of freedom by reconfiguring the propagation environment in response to mobility-induced changes [60]. Chen et al. [14] jointly optimize waveform and RIS phase shifts to improve link robustness under time-varying conditions. In AG-SCCSI, this makes RIS not only a coverage enhancer but also a mobility-resilience enabler that stabilizes communication for sensing feedback, FL model updates, and latency-sensitive coordination. Taken together, these works show that topology management in LAE should be formulated as a joint sensingmobility-communication adaptation problem rather than a pure routing or association problem. 3.3 Balancing Energy Efficiency and System Performance Energy is a stringent constraint in LAE, as airborne platforms must balance propulsion power with the energy required for sensing, communication, and computing tasks. Consequently, energy optimization in AG-SCCSI transcends simple communication power minimization, necessitating a multi-dimensional trade-off among mobility, sensing fidelity, and mission-level task execution. 3.3.1 Energy-Efficient Federated Learning via Joint Optimization. Fu et al. [30] minimize the total energy consumption of UAV-assisted FL systems by jointly optimizing UAV trajectory, device scheduling, transmission power, computing power, and local convergence speed and inference accuracy. Their results show that significant energy savings can be achieved without sacrificing model quality when mobility and learning are optimized together. This is particularly relevant to AG-SCCSI, where energy-efficient orchestration must account for both service-layer objectives and physical flight constraints. 3.3.2 Computation-Aware Compression. Peng et al. [68] incorporate compression ratio into a joint ISCC design problem and optimize trajectory, resource allocation, and data compression using Multi-Agent Proximal Policy Optimization (MAPPO). Their study shows that compression is not merely a joint source-channel coding issue, but a crossdomain control variable that directly affects communication load, computing effort, and energy expenditure. For AGSCCSI, this provides an important insight: semantic or task-oriented data reduction can be an effective way to improve energy efficiency without simply reducing mission completion quality. 3.3.3 Resource-Efficient ISAC under Sparse Occupancy. Mura et al. [64] investigate OFDM-based ISAC waveform design under partial spectrum occupancy and show that high sensing accuracy can still be maintained with limited spectral resources. Their use of Schatten-𝑝 quasi-norm matrix completion improves sensing efficiency while reducing bandwidth burden. This is especially attractive for LAE systems, where aerial nodes often operate under simultaneous energy and spectrum constraints. Overall, energy-efficient AG-SCCSI should be understood as a multi-dimensional tradeoff among mobility, sensing quality, communication cost, computing load, and mission timeliness. Single-layer energy optimization is therefore insufficient. 3.4 Ensuring Trust and Security in Collaborative Intelligence No system can become useful and practical without sound security protection, and thus, in this section, we focus on security issues on AG-SCCSI. Recent survey papers on UAV swarm security show that aerial networks face spoofing, jamming, eavesdropping, hijacking, and coordinated cyber-physical attacks, motivating joint protection across Manuscript submitted to ACM

18

Y. Deng et al.

communication, control, and learning layers [88]. Because AG-SCCSI tightly couples cyber and physical operations, failures in trust and/or security can directly lead to mission degradation or even safety-critical incidents. In collaborative air-ground intelligence, secure aggregation, trustworthy participation, incentive compatibility, and physical-layer protection are all essential. This security perspective is also consistent with significant research efforts on integrated sensing, communication, and computing in emerging cyber-physical systems. However, AG-SCCSI in LAE further requires explicit air-ground orchestration under mobility, uncertainty, and safety-critical mission constraints, which makes its trust and security design fundamentally more coupled and mission-dependent, and hence more challenging [85]. 3.4.1

Trust-Aware Participation and Secure Execution. Trust is the fundamental aspect of security to be addressed.

Xu et al. [97] propose a Bayesian trust assessment mechanism that evaluates a participant’s trust through local model quality and peer recommendations. Such trust-aware selection helps filter out malicious or low-quality contributors before aggregation, thereby improving model integrity. For LAE, this is important because participating aerial and ground devices may differ not only in resource capability, but also in trustworthiness under dynamic and challenged environments. Recent studies further show that trust assessment should explicitly incorporate delay, energy, and data imbalance. For example, Zheng et al. [106] develop a lightweight trust metric for tiny-data FL in UAV/IoT systems, while Alaya et al. [7] review broader trust mechanisms including reputation, blockchain logging, and secure aggregation. These works suggest that trust in AG-SCCSI should be treated as a dynamic orchestration variable rather than a static admission rule. Beyond participant selection, trustworthy AG-SCCSI also requires protection of the execution environment and the broader UAV cyber-physical attack surface. Trusted execution environments have recently been surveyed as a promising hardware-assisted mechanism for protecting UAV confidentiality and integrity, which is particularly relevant when AG-SCCSI pipelines involve safety-critical sensing, learning, and control decisions [25]. Recent work on Internet-of-Drones cybersecurity further develops taxonomies of UAV assets, attacks, countermeasures, and proactive threat-modeling approaches, reinforcing the need for AG-SCCSI security designs that jointly protect communication, sensing, control, and intelligence assets [75]. 3.4.2

Incentive Mechanisms for Active Participation. Our AG-SCCSI framework heavily relies on node/user participa-

tion. Collaborative intelligence also depends on sustained and trustworthy participation. In LAE, however, participating devices may incur various privacy leakage, energy expenditure, and opportunity costs, so incentive design becomes essential. Raja et al. [71] develop a Stackelberg-game-based scheme in which leaders provide rewards to participating devices and optimize the tradeoff between fairness and performance. Hu et al. [40] further show that in drone collaborative training, participation drops significantly without appropriate reward allocation and auditing. Hewa [46] reviews incentive-based FL more broadly and emphasizes the importance of accounting for data quality, privacy cost, and device heterogeneity. For AG-SCCSI, these studies indicate that incentive compatibility is not auxiliary; it is central to maintaining scalable and trustworthy collaboration. 3.4.3

RIS as a Security Enabler. RIS can also enhance security by actively shaping the propagation environment. Be-

yond improving link quality, RIS may suppress eavesdroppers, create favorable alternative paths, or support joint secure sensing and communication. Chen et al. [14] show how RIS can be integrated with ISAC, while Li et al. [50] study Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

19

secure transmission through multiple intelligent reflecting surfaces. Ahmed et al. [6] review RIS-enhanced physicallayer security in UAV-assisted networks, and Li et al. [53] discuss RIS-assisted PLS in ISAC systems. Cai et al. [12] further examine spoofing, jamming, and eavesdropping threats in LAE networking. Collectively, these works show that RIS in AG-SCCSI should be viewed not only as a communication enhancer, but also as a security-oriented environment control mechanism. In fact, UAVs can be leveraged to create more alternative secure paths tp proactively circumvent adversaries. In summary, secure AG-SCCSI requires coordinated mechanisms across trust management, participation incentive design, secure aggregation, and physical-layer security protection. No single mechanism alone is sufficient in AGSCCSI framework. 3.5 Achieving Scalability in Large-Scale Air-Ground Networks Scalability becomes a major concern once LAE moves from isolated UAV missions to dense swarms, large urban deployments, and highly coupled multi-agent, or multi-vendor systems. In such settings, centralized optimization quickly becomes computationally expensive and communication-heavy. The following techniques illustrate how scalable coordination can still be achieved. 3.5.1 Hierarchical FL for Communication Efficiency. Hierarchical FL is a natural scalability mechanism because it reduces long-range communication and organizes participants into aggregation layers. Li et al. [52] show that local aggregation within UAV groups can significantly reduce uplink cost and improve robustness against node failure. In large AG-SCCSI systems, such hierarchical structures also map well onto the natural air-ground resource hierarchy, where local UAV clusters and roadside units can perform intermediate coordination before forwarding compressed updates to edge servers. 3.5.2 MAPPO-Based Decentralized Coordination. MAPPO provides a promising approach to scalable multi-agent coordination by allowing individual UAVs to optimize actions using local observations while still learning cooperative behaviors. Peng et al. [68] use MAPPO for integrated sensing, communication, and computing optimization, while Wei et al. [90] demonstrate its effectiveness in dynamic target search. These studies indicate that decentralized reinforcement learning can reduce the need for full centralized control in large UAV swarms, though its training stability and interpretability remain important open issues. More broadly, a recent survey paper on reinforcement learning for UAV systems provides a systematic taxonomy of RL-based UAV control, planning, and coordination methods, further supporting the use of decentralized learning for scalable AG-SCCSI orchestration [13]. 3.5.3 Model-Driven Deep Learning for Real-Time ISAC. Scalability is not only about the number of nodes, but also about real-time scalability. Model-driven deep learning addresses this issue by embedding optimization structure into learnable architectures. Jiang et al. [44] show that optimization for joint RIS-aided waveform and beamforming design can be unfolded into a model-driven learning architecture, achieving near-optimal performance with much lower online complexity. Complementarily, Cho et al. [17] investigate an aerial RIS-assisted ISAC system and use deep reinforcement learning to jointly adapt transmit beamforming, RIS phase shifts, and aerial trajectory under dynamic conditions. For AG-SCCSI, such approaches are attractive because they preserve part of the interpretability of modelbased design while improving real-time responsiveness in large dynamic networks. Overall, scalable AG-SCCSI will likely require hierarchical architectures, decentralized control, and learning-accelerated optimization to coexist. Manuscript submitted to ACM

20

Y. Deng et al.

3.6

Enabling Decentralized Intelligence with Privacy and Heterogeneity

A defining feature of AG-SCCSI is that intelligence is inheritedly distributed across heterogeneous aerial and ground nodes rather than concentrated in a single cloud or edge server. This decentralization is necessary because LAE involves high mobility, intermittent connectivity, privacy-sensitive data, and mission-driven local decision making. At the same time, it introduces new design challenges related to heterogeneity, trust, and cross-domain coordination. This view is consistent with prior survey paper on multi-access edge computing and machine learning in the Internet of UAV (IoUAV), which highlights that learning, offloading, and mobility management are deeply intertwined in UAVcentric systems [66]. It is also aligned with a broader survey paper on machine learning in the Internet of Drones (IoD), which shows that UAV intelligence increasingly relies on the tight interplay among sensing, communication, edge computing, and autonomous decision making [38]. Compared with these broader IoUAV and the IoD perspectives, AG-SCCSI further stresses explicit air-ground collaboration and end-to-end resource orchestration across all SCCSI dimensions, with explicit roles for storage, semantic communication, and mission-driven closed-loop control. 3.6.1

Federated Learning under Device and Data Heterogeneity. In air-ground systems, clients differ in sensing modal-

ity, computational capability, battery power, and mobility pattern. These differences induce both statistical and system heterogeneity, which can severely affect FL convergence and inference accuracy. Recent studies show that clustered, hierarchical, and personalized FL architectures are particularly relevant in such settings [52]. Zhang and Han [100] demonstrate how multi-UAV networks can benefit from federated coordination, while Cheriguene et al. [16] propose participant-selection mechanisms that balance model accuracy, UAV energy consumption, and data heterogeneity. The takeaway here is that FL in AG-SCCSI must be resource-aware and heterogeneity-aware from the outset. 3.6.2

Trust, Privacy, and Incentive Mechanisms in Decentralized Intelligence. Decentralized intelligence also raises pri-

vacy and trust concerns because model updates may leak information and adversarial participants may poison training. Kapoor and Kumar [47] review attacks and defenses in federated urban sensing systems, while Wang et al. [81] discuss energy-efficient UAV-assisted FL under uncertainty. In parallel, blockchain-assisted clustered FL and trust-based participation mechanisms have been proposed to improve both scalability and robustness. For AG-SCCSI, these results suggest that privacy, trust, and incentives should be designed jointly rather than treated as separate add-ons to FL. 3.6.3

OTA Aggregation and Mobility-Aware Scheduling. In delay-sensitive air-ground environments, AirComp can sub-

stantially reduce FL uplink latency by aggregating simultaneous transmissions directly over the air during the transmission process. Mobility-aware scheduling is then needed to determine when, where, and from which clients model updates should be collected. Recent studies show that OTA aggregation, asynchronous FL, and client scheduling based on link quality and resource budget are all promising for high-mobility systems [49, 71, 81, 104]. For LAE, this means that learning rounds must be jointly designed with trajectory, communication, and service timing. 3.6.4

Edge Intelligence Federation and Cross-Domain Orchestration. Beyond isolated FL rounds, AG-SCCSI requires

cross-domain intelligence federation across UAVs, vehicles, roadside infrastructure, and edge servers. This implies that SCCSI decisions jointly shape how intelligence is generated and propagated. Quan et al. [70] highlight this broader cyber-physical perspective, while [11] points to the need for collaborative learning across heterogeneous aerial and ground perception agents. In LAE, cross-domain orchestration is particularly important because aerial and ground nodes often have asymmetric latency, energy, and reliability profiles. Thus, decentralized intelligence should be treated as a multi-domain resource orchestration problem, not merely as distributed model training. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

21

In summary, decentralized intelligence in AG-SCCSI requires joint attention to heterogeneity, trust, mobility, and cross-domain orchestration. These aspects are tightly coupled and should be co-designed. 3.7 Scaling Computing Coordination in Large, Uncertain Networks A prior survey paper on UAV-enabled edge computing has systematically reviewed computation offloading, trajectory design, and resource allocation from a resource-management perspective, providing a useful baseline for understanding aerial computing coordination [93]. In AG-SCCSI, these issues become broader because computing coordination must be jointly considered with sensing, communication, storage, and mission-level control rather than treated as an isolated MEC computing resource allocation problem. As AG-SCCSI scales up, the accessible computing resource pool is not fixed; it depends on UAV position, altitude, link quality, and demand distribution, which makes conventional centralized optimization increasingly ineffective in large uncertain environments. Recent work along this line has begun to interpret this problem from a computing-power-network (CPN) perspective. In particular, Sun et al. [77] connect CPNs with LAE intelligent communications and argue that LAE should be viewed as a broader air-ground orchestration problem in which aerial mobility actively reshapes the accessible computing landscape. This viewpoint is highly relevant to AG-SCCSI because it extends the notion of offloading from simple server selection to mobility-dependent computation-network co-design. 3.7.1 Evolutionary Algorithms for Scalable Optimization. Han et al. [37] use a swarm-intelligence-based Fireworks Algorithm to jointly optimize user association, 3D deployment, and flight trajectories in large-scale UAV-assisted MEC systems. Evolutionary algorithms are attractive here because they handle multi-objective, non-convex search spaces without requiring strong analytical structure. Their value in AG-SCCSI lies in offering scalable approximate solutions for complex air-ground coordination problems where exact convex reformulation is hard. 3.7.2 Stochastic Modeling for Dynamic Environments. When large-scale systems are too complex to model for deterministic optimization alone, stochastic modeling provides a tractable way to capture uncertainty and spatial randomness. Hui et al. [43] model user locations via stochastic geometry and define successful edge computing probability as a reliability-oriented performance metric. Their results show that UAV altitude and offloading decisions should be jointly optimized because service coverage and reliability are both spatially uncertain. Such models are particularly useful in LAE because they reveal structural tradeoffs before one commits to detailed deployment design. 3.7.3 Altitude-Dependent Computing Power in MEC Systems. Deng et al. [18–21] show that in relay-based MEC systems, UAV altitude adaptations not only improve link quality but also reach more ground computing resources. Higher altitudes do improve the accessibility to ground computing nodes, but may also degrade communication quality, thereby creating a non-trivial tradeoff between computation availability and transmission performance. By combining stochastic geometry and queueing-theoretic reasoning, these works highlight that computing coordination in LAE is inherently mobility-shaped and should be analyzed at the end-to-end task level rather than only at the link level. In [? ], we apply the information bottleneck theory to develop a task-oriented feature compression scheme to maximize the end-to-end visual inference quality under bandwidth constraints, which provides a use case of this design philosophy. 3.7.4 Multi-Objective Optimization for Large-Scale UAV Networks. Large uncertain systems also require balancing multiple objectives such as delay, energy, throughput, and task success. Baktayan et al. [10] investigate multi-agent task Manuscript submitted to ACM

22

Y. Deng et al. Table 4. Representative datasets and benchmarks for AG-SCCSI research in low-altitude economy.

Name

Type

GRACO

Real-world multimodal dataset

Main Task

Main Notes / Limitations

Cooperative localiza- Provides synchronized LiDAR, stereo images, IMU, and tion and mapping GPS from aerial and ground perspectives, with millisecondlevel synchronization and centimetre-level RTK ground truth. It is highly suitable for cooperative SLAM and map fusion, but mainly focuses on localization/mapping rather than the full AG-SCCSI stack. Griffin Synthetic Aerial-ground coop- A large-scale aerial-ground cooperative perception benchbenchmark erative detection and mark with over 250 dynamic scenes, varied UAV altitudes, dataset tracking and unified protocols for communication efficiency, latency tolerance, and altitude adaptability. It is strong for controlled evaluation, but is still simulation-based. AGC-Drive Real-world per- Aerial-ground cooper- A large-scale real-world UAV–vehicle collaborative percepception dataset ative 3D perception tion dataset with annotated 3D bounding boxes and public benchmark tools. It is valuable for realistic cross-view perception evaluation, especially under time delays and pose errors, but currently emphasizes perception more than networking or computing. OpenUAV / UAV-Need- Platform + Realistic UAV naviga- Provides a realistic UAV simulation platform and an Help benchmark tion assistant-guided object-search benchmark, with about 12k trajectories for UAV vision-language navigation. It is useful for task-driven UAV autonomy, but less directly oriented to communication/computing benchmarking. AERPAW RF Sensor Real-world RF 3D UAV localization / Contains TDOA localization estimates, ground-truth UAV Measurements with dataset RF sensing positions, and LOS/NLOS labels under varying altitudes UAV and bandwidths. It is useful for localization, spectrum, and RF sensing studies, but is communication-centric. AERPAW wireless Dataset collec- Aerial wireless prop- Consolidates public AERPAW datasets covering I/Q samdataset collection tion / survey agation and network- ples, spectrum measurements, LoRa, 5G-NSA, RF localizaing tion, and propagation scenarios. It is valuable for reproducible aerial-network research, but does not by itself provide mission-level control annotations.

Refs. [108]

[82]

[39]

[86]

[22]

[72]

offloading and resource allocation for UAV swarms using multi-objective optimization. Such approaches are important for AG-SCCSI because they explicitly recognize that no single metric is sufficient in large mission-driven systems. Instead, computing coordination must be designed under coupled service, mobility, and uncertainty constraints. Overall, scaling computing coordination in AG-SCCSI requires a combination of structural modeling, scalable heuristics, and mobility-aware communication-computing-buffering co-design. This is one of the clearest areas where LAE departs from conventional terrestrial MEC. This insight is also relevant beyond generic task offloading. For UAV visual navigation in GNSS-denied environments, ground computing resources can serve as an expandable computing and inference substrate, enabling lightweight aerial nodes to offload computation-intensive localization and map-assisted refinement while maintaining closed-loop responsiveness [29]. 4 Datasets, Testbeds, and Evaluation Methodology 4.1

Datasets and Benchmarks

To provide an objective assessment of the AG-SCCSI framework and its corresponding design mechanisms, it is crucial to offer rich datasets, effective testbeds, and a viable methodology. In this section, we review a representative set of datasets and benchmarks. Table 4 groups representative AG-SCCSI data resources into real-world cooperative SLAM/perception datasets, task-oriented UAV navigation benchmarks, and aerial wireless channel measurements. Existing public resources already support important subproblems such as cooperative localization and mapping, aerialground 3D perception, realistic UAV navigation, and RF localization or propagation studies. Representative examples Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

23

include GRACO for cooperative SLAM, Griffin and AGC-Drive for aerial-ground perception, OpenUAV/UAV-NeedHelp for realistic UAV navigation, and the emerging AERPAW wireless datasets for aerial connectivity and localization evaluation. From an AG-SCCSI perspective, however, the main challenge is not only data scarcity but also benchmark fragmentation. Most public resources still emphasize sensing, perception, navigation, or wireless measurements in isolation, mostly for specific purposes, whereas unified benchmarks that jointly expose synchronized sensing, communication, computing, storage, control actions, and mission executions remain rare. This makes it difficult to compare task-driven orchestration methods under consistent assumptions, especially when latency, reliability, mobility, and safety must be evaluated. 4.2 Testbeds, Digital Twins, and Simulators Table 5 summarizes representative experimental platforms for AG-SCCSI research, including open field testbeds, digitaltwin frameworks, and modular co-simulation engines. AERPAW is particularly valuable for real-flight wireless experimentation, while the O-RAN DT-RAN provides a structured reference for digital-twin-enabled planning, testing, and energy optimization. At the simulator level, CARLA and AirSim, together with CARLA–AirSim co-simulation and ns3-based networking modules, offer a practical path toward repeatable evaluation of urban mobility, aerial sensing, UAV control, and wireless channel dynamics. To further bridge the reality gap in complex environments, recent efforts such as [36] incorporate physics-informed digital twins to achieve high-fidelity modeling for RIS-assisted wireless systems, ensuring that simulated resource orchestration aligns with physical electromagnetic constraints. Despite this progress, no single open platform yet provides a mature full-stack LAE evaluation loop, spanning flight physics, air-ground sensing, wireless networking, edge computing and learning, storage or caching, and multitimescale resource orchestration. A promising direction is therefore interoperable co-simulation, in which scenario generation, network emulation, service execution, and policy control share synchronized states and standardized logs for reproducible mission-level benchmarking. Existing platforms have already provided partial support for sensing, communication, control, or mobility evaluation, but few expose synchronized observability across all SCCSI dimensions. As a result, full-stack validation of mission-driven air-ground orchestration remains an open challenge. Recent O-RAN-enabled LAE studies further show that open interfaces, RAN intelligent controllers, and xApp/rApp-based closed-loop control can provide a practical software architecture for mission-critical LAE orchestration [3]. 4.3 Evaluation Metrics for Mission-Centric AG-SCCSI Tables 6 and 7 summarize representative evaluation metrics for AG-SCCSI. Unlike conventional communication-centric systems, AG-SCCSI should not be evaluated solely by throughput or isolated task completion performance. Since LAE services are mission-driven and closed-loop by nature, future evaluations should jointly report mission fulfillment, latency, reliability, energy, safety, and orchestration overhead, together with dimension-specific performance indicators for SCCSI. In addition to reporting conventional sensing or communication indicators, future AG-SCCSI evaluations should specify whether the benchmark provides synchronized control logs, edge execution traces, wireless context, and safety annotations. Without such cross-layer observability, it remains difficult to determine whether a method truly improves mission-level performance or only optimizes an isolated subsystem. Manuscript submitted to ACM

24

Y. Deng et al. Table 5. Representative testbeds, digital twins, and simulators for AG-SCCSI research in low-altitude economy.

Platform

Type

AERPAW

Open testbed

O-RAN DT-RAN

CARLA

AirSim

CARLA–AirSim simulation

ns-3

Main Scenario

Main Notes / Capability

Refs.

field

UAV + advanced wire- An open aerial wireless experimentation platform built less experimentation around the convergence of 5G and autonomous drones. It is valuable for real-flight validation, RF measurements, and reproducible aerial networking studies, but does not natively provide a full AG-SCCSI software stack. Digital-twin Planning, testing, en- Provides a structured DT-RAN reference covering AI/ML framework ergy, and site opti- training and testing, network testing automation, network mization planning, network energy saving, and site-specific optimization. It is not LAE-specific, but is highly relevant as a DT template for AG-SCCSI orchestration. Urban simula- Traffic, sensing, and An open-source urban simulator with flexible sensor suites, tor control actor control, multi-client architecture, traffic scenario generation, and ROS integration. It is strong for ground mobility and urban scene generation, but does not natively model detailed UAV flight dynamics. UAV / robotics Drone dynamics and An open-source drone simulator that supports physically simulator autonomy and visually realistic simulation with PX4 HITL/SITL integration. It is strong for UAV control and perception prototyping, but usually needs to be paired with other tools for realistic urban traffic and network-layer evaluation. co- Co-simulation Aerial-ground per- Demonstrated in recent aerial-ground cooperative percepstack ception benchmark tion benchmarking, this stack combines CARLA urban trafgeneration fic scenes with AirSim-based drone dynamics. It is useful for synchronized aerial-ground scenario generation, but remains primarily perception-oriented unless combined with network simulators. Network simu- Wireless / protocol A mature open-source discrete-event network simulator for lator / propagation evalua- Internet and wireless systems. Recent releases continue to tion improve 3GPP propagation support, making ns-3 a strong networking layer for AG-SCCSI co-simulation, though it must be integrated with mobility and task-level simulators for full mission evaluation.

[62]

[67]

[26]

[74]

[26, 74, 82]

[80]

Table 6. Mission- and system-level metrics for AG-SCCSI in low-altitude economy. Metric

Level

What It Measures

Typical Use

Notes

Task Completion Probability / Ratio

Mission-level Mission / system

Offloading, collaborative navigation, edge-aerial task execution Navigation, emergency response, logistics, edge inference

Requires an explicit task model and deadline definition

End-to-End Latency

Probability or ratio that a task is completed within mission constraints Delay from sensing or request generation to final action or decision

Reliability / Deadline Satisfac- Mission-level tion Probability Energy Consumption / En- System-level ergy per Mission Safety Violation Probability

Mission-level

Scalability Overhead

System-level

/

Orchestration

Manuscript submitted to ACM

Should ideally be decomposed into sensing, transmission, computing, and control latency Probability that service meets Real-time mission assurance Reliability targets should be deadline or QoS requirement and delay-sensitive tasks defined according to scenario and mission type Total or normalized energy UAV planning, FL, offloading, It is preferable to separate cost for mission execution hot-spot caching propulsion energy from SCCSI service energy Probability of violating safety Dense LAE operations, navi- Requires an explicit safety constraints, no-fly zones, or gation, airspace management model and scenario-specific mission-critical limits definition Growth of signaling, control, Swarm coordination, DT- Should include both controlor computation cost with net- assisted control, hierarchical plane and data-plane costs work size FL

Collaborative Air-Ground SCCSI for LAE

25

Table 7. Dimension-specific metrics for AG-SCCSI for low-altitude economy. Metric

SCCSI sion

Dimen- What It Measures

Localization / Mapping Accu- Sensing racy

Position error, pose estimation quality, and map consistency Detection / Tracking Perfor- Sensing / Intelli- mAP, tracking accuracy, ID mance gence switches, and occlusion robustness Semantic Utility / Task- Communication / Usefulness of transmitted seRelevance Score Intelligence mantic features for downstream tasks Throughput / Achievable Rate Communication Raw transmission capability Coverage Probability / Outage Probability Age of Information (AoI) / Semantic-AoI Cache Hit Ratio / Backhaul Load Reduction

Communication Communication / Intelligence Storage

Learning Accuracy / Conver- Intelligence gence Speed Privacy / Security Overhead

Intelligence / Communication

Typical Use

Notes

C-SLAM, GNSS-denied navi- Should be interpreted togation, inspection gether with latency and computing cost Traffic monitoring, surveil- Should be paired with comlance, cooperative perception munication overhead SemCom, edge-aerial feature delivery

More task-aligned than raw bitrate, but harder to standardize Backhaul, feature delivery, Still necessary as a supportABS coverage ing metric, but not sufficient alone Likelihood of service cover- Air-ground connectivity, ABS Needs scenario-dependent inage or link failure deployment, relay planning terpretation Freshness of updates deliv- Monitoring, control loops, Freshness alone does not meaered to controllers or learners sensing updates sure content usefulness How much demand is served Aerial caching, hot-spot deliv- Does not fully capture QoE or locally and how much back- ery fairness by itself haul is saved Model quality and training ef- FL, OTA-FL, distributed intel- Should be interpreted toficiency ligence gether with communication and energy costs Cost of secure aggregation, Secure FL, trusted collabora- Often omitted in current simencryption, authentication, or tion, mission-critical commu- ulation studies privacy mechanisms nications

5 Emerging Application Scenarios Low-altitude UAV deployment enables a broad range of LAE applications, including traffic management, collaborative logistics, emergency response, and GNSS-denied navigation. In this section, we discuss representative application scenarios and highlight how AG-SCCSI enabling technologies can be orchestrated in closed-loop mission workflows. To make the discussion more concrete, Table 8 summarizes these scenarios from a mission-centric AG-SCCSI perspective, covering their mission objectives, dominant SCCSI dimensions, and representative enabling technologies. This structured view complements the narrative discussion below and illustrates how different LAE services map to different orchestration priorities. These case studies are also intentionally selected to reflect the bidirectional nature of air–ground collaboration in LAE. While some scenarios highlight how aerial platforms enhance ground-side sensing and service delivery, others emphasize how ground infrastructures actively support airborne perception, navigation, and closed-loop autonomy through communication, computing, storage, and model assistance. 5.1 UAV-assisted Ground Traffic Control UAV-assisted AG-SCCSI can provide a new paradigm for real-time traffic management by combining wide-area aerial perception, ground-side microscopic sensing, edge intelligence, and closed-loop control. In such systems, UAVs offer a macroscopic view of ground traffic networks by conducting wide-area monitoring, while ground vehicles, including Connected Autonomous Vehicles (CAVs), provide microscopic data on local vehicle interactions. For instance, UAVs equipped with high-resolution cameras and LiDAR can capture real-time traffic density, lane occupancy rates, and abnormal events, such as reckless driving, accidents, or illegal parking, across areas Manuscript submitted to ACM

26

Y. Deng et al. Table 8. Representative mission-centric AG-SCCSI case studies: scenario overview.

Scenario Real-Time Air-Ground Traffic Control Intelligent Collaborative Logistics GNSS-Denied Air-Ground Navigation

Mission Objective

Dominant sions

SCCSI

Dimen- Representative Enabling Technologies

Wide-area traffic monitoring, Sensing, Communication, Intel- ISAC, SemCom, AG-MEC, DT-assisted orchestration, congestion mitigation, and in- ligence, Computing FL/MARL, RIS/STAR-RIS cident response Multi-modal parcel delivery, Communication, Computing, DT/radio-map-driven orchestration, AG-MEC, caching, hub coordination, and time- Storage, Intelligence, Sensing MARL/FL, ISAC-based obstacle avoidance sensitive logistics under nofly and mobility constraints High-precision and reliable Sensing, Communication, Com- Task-oriented feature compression, edge–aerial collabUAV localization/navigation puting, Intelligence, Storage oration, ground CPN-assisted computing, map/model in urban canyons under strict caching and refinement, DT-assisted localization uplink, latency, and energy constraints

spanning 5–10 km in radius [9]. Instead of overwhelming the uplink with raw video streams, UAVs can employ Semantic Communication (SemCom) to extract and transmit only task-relevant features. These aerial insights complement the detailed data collected by ground CAV sensors regarding vehicle acceleration patterns and driver behaviors. Thus, this hybrid sensing architecture facilitates the identification of potential “hot spots”, improves bottleneck diagnosis efficiency, and refines traffic alerting systems [102]. The feedback control loop in such systems can be characterized as follows:

• Sensing Layer: UAVs and CAVs cooperatively detect traffic oscillations using Integrated Sensing and Communication (ISAC), which allows the simultaneous tracking of vehicle velocities and data transmission without extra spectrum overhead. • Communication/Computing Layer: 5G/6G networks deliver semantic data to edge nodes, with UAVs acting as supplementary airborne edge servers via the AG-MEC continuum to process latency-sensitive tasks locally. • Decision Layer: Digital Twin (DT)-assisted orchestration constructs a real-time virtual replica of the traffic network, allowing Multi-Agent Reinforcement Learning (MARL) or Federated Learning (FL) models to generate collaborative control strategies, such as variable speed limits, while preserving data privacy across different vehicle fleets. • Execution Layer: Controllable vehicles, such as CAVs and buses, act as “mobile regulators” to smooth traffic waves, while UAVs broadcast advisories via V2X communications, potentially assisted by RIS/STAR-RIS to guarantee reliable downlink coverage in urban canyons. Compared with conventional traffic control mechanisms, this AG-SCCSI-enabled system can enhance congestion mitigation, incident response, and transportation throughput through real-time information processing and multimodal coordination. 5.2

Ground CPN-Assisted OTA Perception for GNSS-Denied Navigation

Ground support is equally important in AG-SCCSI, especially when aerial nodes face stringent payload, energy, and onboard computing constraints. A representative example is UAV navigation in GNSS-denied urban canyons, where satellite positioning can be severely degraded by blockage and multipath, while purely onboard visual-inertial solutions often suffer from drift accumulation and substantial real-time computation burden. In this scenario, the ground side Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

27

should not be viewed merely as a communication backhaul, but as a proactive provider of computing, storage, and machine learning model assistance for aerial localization and navigation. Following the edge–aerial collaboration paradigm, in our prior work [29], the UAV performs multi-view image acquisition together with lightweight task-oriented feature extraction and compression, while the ground-side computing power network (CPN) offers an expandable pool of edge resources for computation-intensive inference, map/modelassisted localization refinement, and closed-loop navigation feedback. Such a design transforms onboard perception from an isolated sensing process into a mission-oriented air–ground collaborative service pipeline. More specifically, the operational workflow can be summarized as follows:

• Sensing and Representation Layer: The UAV captures multi-view imagery in complex urban canyons and extracts compact task-relevant visual features instead of transmitting raw video streams. This reduces the uplink burden while preserving navigation-critical information. • Communication and Access Layer: Based on uplink quality, latency budget, and resource availability, the UAV selects an appropriate ground point of connection and delivers compressed features to the accessible ground CPN resources through air–ground links. • Ground Computing and Storage Layer: Ground edge servers execute computation-intensive inference and refinement, including feature matching, map-assisted localization, model-based state estimation, and retrieval of cached maps or navigation priors. In this way, the ground side provides not only stronger computing capability but also persistent storage support that lightweight UAVs cannot easily carry onboard. • Closed-Loop Control Layer: The refined localization and navigation state is fed back to the UAV for trajectory correction, path replanning, and safe motion control, thereby forming a closed sensing–communication– computing–intelligence loop. Compared with purely onboard autonomy, this ground-assisted design can substantially improve localization robustness and navigation reliability under strict energy and onboard computing constraints. More broadly, it demonstrates a complementary direction of AG-SCCSI: not only can aerial platforms enhance ground-side perception and service coverage, but ground infrastructures can also actively empower OTA perception, airborne decision making, and mission execution through computing, storage, and machine learning support.

5.3 Intelligent Collaborative Logistics The integration of UAVs with ground transportation systems forms an efficient multi-modal logistics network capable of addressing diverse service demands [32]. In this cooperative framework, UAVs can support suburban or urban-edge delivery tasks, while ground vehicles, couriers, and logistics hubs provide complementary capacity, endurance, and accessibility. Such a dual-layer system can mitigate traffic-induced delays, especially for time-sensitive deliveries such as blood products, emergency pharmaceuticals, and diagnostic specimens. Recent cooperative delivery designs further explore UAV–ground-vehicle coordination, where aerial mobility and ground transportation resources are jointly scheduled to improve delivery timeliness and service flexibility. For example, TriDeliver models cooperative air–ground instant delivery with UAVs, couriers, and crowdsourced ground vehicles, illustrating how heterogeneous delivery agents can be jointly orchestrated in urban logistics [32]. The key operational workflow of UAV–ground cooperative logistics systems can be summarized as follows: Manuscript submitted to ACM

28

Y. Deng et al.

• Intelligent Order Parsing: Upon receiving user orders via mobile or web platforms, the system extracts multidimensional parameters, including cargo type, weight, and urgency level. These parameters are tagged with priority labels using NLP-based semantic analysis at AG-MEC servers. • Cross-Modal Task Allocation: Leveraging Multi-Agent Reinforcement Learning (MARL) and Trust-Aware Federated Learning (FL), orders are dynamically assigned to UAVs, couriers, and ground vehicles. This decentralized intelligence supports resource utilization while avoiding unnecessary exposure of proprietary delivery data among different logistics operators. • Air-Ground Collaborative Path Optimization: Based on 3D urban models, the system generates initial paths and uses real-time environmental data to trigger local obstacle avoidance via ISAC. Meanwhile, trajectory-aware computing coordination synchronizes UAV–ground vehicle rendezvous at logistics hubs, ensuring seamless physical and data handovers. 6 Open Challenges and Future Directions Although recent studies have advanced individual dimensions of SCCSI for LAE, a number of fundamental challenges remain open before integrated AG-SCCSI can be deployed at scale. In particular, existing solutions are still largely fragmented across layers, resources, and control timescales, whereas practical LAE services require mission-oriented, closed-loop, and safety-aware orchestration over highly dynamic air-ground systems. In what follows, we highlight several promising research directions that, in our view, are particularly important for the next stage of AG-SCCSI development. 6.1

Mission-Oriented Closed-Loop Orchestration

A central open issue is how to transition from resource-level optimization to mission-oriented closed-loop orchestration. Most existing studies optimize one or two dimensions, such as communication throughput, task offloading latency, sensing quality, or inference accuracy, in isolation. However, LAE missions are inherently end-to-end in nature. For example, a delivery, inspection, or emergency-response task is successful only when sensing, communication, computing, storage, and decision making jointly satisfy latency, reliability, safety, and energy constraints. Future research should therefore develop mission-level orchestration frameworks that directly optimize task completion quality rather than surrogate single-layer metrics. This requires unified abstractions that map heterogeneous mission requirements into multi-resource control variables across air and ground nodes. It also calls for hierarchical control architectures in which fast-timescale adaptation handles link fluctuations, mobility, and sensing uncertainty, while slower-timescale planning governs deployment, caching, model placement, and service migration. Another important direction is the design of online and predictive orchestration mechanisms that combine digital twins, mobility prediction, and uncertainty-aware optimization to support look-ahead decision making in dynamic 3D airspace. 6.2

Energy-Sustainable AG-SCCSI Systems

Energy remains one of the most stringent bottlenecks in LAE, especially for UAVs carrying sensing, communication, and computing payloads. Existing work often treats propulsion energy, communication energy, and computation energy separately, which is insufficient for AG-SCCSI systems where these components are tightly coupled. For instance, a higher altitude or longer hovering duration may improve sensing coverage or access to more ground resources, but it also increases propulsion cost and may degrade communication quality. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

29

A key future direction is to establish mission-aware energy models that jointly characterize mechanical flight energy and SCCSI service energy under realistic mobility, payload, and environment conditions. Based on such models, future systems should support energy-aware task partitioning, adaptive sensing fidelity control, dynamic communicationcomputing tradeoff optimization, and charging-aware trajectory planning. Another important topic is the integration of energy replenishment mechanisms, such as battery swapping, opportunity charging, and vehicle-assisted or stationassisted energy support, into the orchestration loop. Rather than viewing energy merely as a hard constraint, future AG-SCCSI designs should treat it as the most important system state that directly shapes service continuity and mission feasibility [35]. 6.3 Interference-Resilient and Spectrum-Aware Operation As LAE scales toward dense and heterogeneous operations, interference management will become substantially more challenging than in conventional terrestrial systems. Air-ground links are affected not only by co-channel interference from neighboring UAVs and ground infrastructure, but also by strong line-of-sight exposure, dynamic topology changes, and spectrum contention across multiple services. In addition, future LAE systems are expected to support integrated sensing and communication, which further couples spectrum usage with sensing quality and situational awareness. These challenges indicate that interference resilience cannot be addressed by communication-layer adaptation alone. Recent LAE-oriented studies on cellular-connected UAVs therefore emphasize the need for joint communication-control co-design to handle coverage holes, interference uncertainty, and spectrum scarcity in dense aerial deployments [54]. Future research should move beyond static interference avoidance and develop interference-resilient AG-SCCSI architectures that combine spectrum awareness, environment perception, and adaptive control. Promising directions include radio-map-assisted spectrum access, digital-twin-enabled interference prediction, trajectory-aware interference coordination, and RIS-assisted or beamforming-assisted propagation shaping. It is also important to investigate how interference can be incorporated into task-level decision making, rather than being treated purely as a physical-layer impairment. For example, future systems may jointly optimize sensing fidelity, communication reliability, and endto-end quality-aware UAV-assisted computation offloading under interference uncertainty, thereby enabling robust mission execution in crowded low-altitude environments. 6.4 Scalable Cooperative Intelligence under Heterogeneity Distributed intelligence is expected to play a foundational role in AG-SCCSI, but current approaches still face major challenges in scalability, robustness, and various heterogeneities. In practical LAE scenarios, aerial and ground nodes differ significantly in sensing modality, data quality, energy budget, mobility pattern, computing capacity, and communication reliability. These factors lead to severe statistical and system heterogeneities, which can significantly degrade the effectiveness of federated learning, distributed inference, and multi-agent coordination. A major future direction is to build heterogeneity-aware cooperative intelligence frameworks for air-ground systems. This includes hierarchical and clustered learning architectures, asynchronous model aggregation, client selection under mobility and resource constraints, and personalized or task-adaptive model updates for heterogeneous agents. Another promising direction is the joint design of learning and networking, where communication scheduling, trajectory control, sensing decisions, and model aggregation are optimized together. For large-scale swarms and multi-domain systems, scalable coordination also requires decentralized decision mechanisms, such as multi-agent reinforcement learning and game-theoretic control, with explicit consideration of convergence stability, communication overhead, Manuscript submitted to ACM

30

Y. Deng et al.

and safety constraints. Importantly, future work should not only improve learning accuracy, but also characterize how distributed intelligence affects mission-level utility, timeliness, and reliability. 6.5

Safety, Security, and Compliance by Design

Safety cannot be treated as an afterthought in LAE. Unlike many traditional wireless systems, low-altitude operations directly interact with physical environments, public infrastructure, and human users. As a result, failures in sensing, communication, control, or intelligence may be translated into tangible operational risks. Meanwhile, the integration of air and ground resources also enlarges the attack surface, including spoofing, jamming, model poisoning, malicious data injection, and unauthorized access to mission-critical services. Future AG-SCCSI systems, therefore, need to adopt safety, security, and compliance by design. One important research direction is the development of assurance-aware orchestration mechanisms that explicitly incorporate risk, uncertainty, and constraint satisfaction into decision making. Another is to establish trustworthy pipelines for data collection, model training, inference, and control execution, supported by secure logging, provenance tracking, and auditable coordination. Beyond cyber security, regulatory compliance should also be translated into machine-interpretable operational constraints, such as no-fly zones, privacy rules, airspace restrictions, and emergency response priorities. More broadly, future work should investigate how to combine formal methods, runtime monitoring, and learning-enabled control to provide certifiable and explainable autonomy for AG-SCCSI systems. 6.6

Benchmarks, Digital Twins, and Reproducible Evaluation

Another critical gap lies in evaluation methodology. Existing studies often rely on isolated simulation settings, taskspecific assumptions, and non-unified performance metrics, which makes it difficult to compare different AG-SCCSI designs in a meaningful way. Moreover, many important dimensions in LAE, such as safety margin, mission completion reliability, cross-layer overhead, and adaptability to uncertainty, are still insufficiently benchmarked. Future research should prioritize the development of shared datasets, interoperable digital twins, and reproducible testbeds for air-ground collaborative systems. In particular, digital twins should evolve from passive visualization tools to active experimentation platforms that can support what-if analysis, policy validation, and cross-layer co-simulation of mobility, sensing, networking, computing, and control. Equally important is the establishment of mission-level evaluation metrics, such as task completion ratio, end-to-end timeliness, energy per mission, safety violation probability, and robustness under topology or channel uncertainty. Standardized benchmarks along these dimensions would significantly improve the comparability, credibility, and practical relevance of future AG-SCCSI research. 6.7

Toward a Unified AG-SCCSI Research Agenda

Overall, the next phase of AG-SCCSI research should move toward a unified agenda in which mission requirements, multi-resource orchestration, trustworthy intelligence, and reproducible evaluation are treated in an integrated manner. The long-term challenge is not merely to connect aerial and terrestrial systems, but to enable them to cooperate as a coherent, adaptive, and verifiable service fabric for LAE. This service fabric may also extend beyond air-ground collaboration toward multi-tier aerial infrastructure. For instance, recent work highlights the potential of high-altitude platforms as supervisory and computing-capable infrastructure for LAE, bridging communication, computing, navigation integrity, and regulatory oversight across regional low-altitude operations [41]. Addressing these challenges will require advances in system modeling, algorithm design, digital twinning, distributed intelligence, and standardization. Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

31

We believe that such a shift from fragmented optimization to mission-centric integration will be essential for realizing scalable and trustworthy low-altitude services. 7 Conclusion This survey has presented a mission-centric view of low-altitude economy (LAE) through the unified AG-SCCSI framework, which integrates sensing, communication, computing, storage, and intelligence across air and ground systems. Rather than treating UAV networking, edge computing, caching, or distributed intelligence as isolated design problems, we have argued that practical LAE services must be understood as closed-loop multi-resource orchestration problems shaped by mobility, uncertainty, safety, and heterogeneous mission requirements. From this perspective, we have first introduced the AG-SCCSI framework and the corresponding air-ground resource orchestration mechanisms. We have then reviewed the enabling technologies that support integrated AG-SCCSI, including collaborative sensing and ISAC, air-ground communications, AG-MEC, storage and caching, decentralized intelligence, and large-scale computing coordination. Beyond technologies alone, we have also summarized representative datasets, testbeds, digital twins, simulators, and mission-centric evaluation metrics, and discussed representative application scenarios to illustrate how different LAE services map to different orchestration priorities. A central takeaway of this survey is that the main bottleneck in LAE is no longer mere connectivity. Instead, the core challenge lies in how to translate application-driven mission requirements into coordinated sensing, communication, computing, storage, and intelligence decisions over highly dynamic air-ground systems. This requires moving beyond siloed optimization toward unified, trustworthy, and reproducible air-ground service design. To the best of our knowledge, this is the first survey that systematically frames LAE from the perspective of AG-SCCSI. We hope this work can serve not only as a structured reference on the current literature, but also as a roadmap for future research on mission-oriented orchestration, scalable cooperative intelligence, safety-aware autonomy, and reproducible validation for low-altitude services. Acknowledgments The research work described in this paper was conducted in the JC STEM Lab of Smart City funded by The Hong Kong Jockey Club Charities Trust. This work was also supported in part by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. CityU 11216324) and in part by the Hong Kong SAR Government under the Global STEM Professorship. The work of Y. Deng was also supported in part by the National Natural Science Foundation of China under Grant No. 62301300 and in part by the Shandong Province Science Foundation under Grant No. ZR2023QF053. References [1] 2024. Develop the Low-Altitude Economy. The Chief Executive’s 2024 Policy Address. https://www.policyaddress.gov.hk/2024/en/p81.html Accessed Oct. 2024. [2] 2025. Resources on the Move for Smart City: A Disruptive Perspective on the Grand Convergence of SCCSI. IEEE Communications Magazine (2025). doi:10.1109/MCOM.001.2400084 Early Access. [3] Aly Sabri Abdalla and Vuk Marojevic. 2026. Next Generation Intelligent Low-Altitude Economy Deployments: The O-RAN Perspective. arXiv preprint arXiv:2601.00257 (2026). doi:10.48550/arXiv.2601.00257 [4] Muhammad Adil, Houbing Song, Mian Ahmad Jan, Muhammad Khurram Khan, Xiangjian He, Ahmed Farouk, and Zhanpeng Jin. 2024. UAVAssisted IoT Applications, QoS Requirements and Challenges with Future Research Directions. ACM Computing Surveys 56, 10 (May 2024), 1–35. [5] Tazeem Ahmad, Alicia Esquivel Morel, Nuo Cheng, K. Palaniappan, Prasad Calyam, K. Sun, and Jianli Pan. 2026. Future UAV/Drone Systems for Intelligent Active Surveillance and Monitoring. ACM Computing Surveys 58, 2, Article 35 (2026), 37 pages. doi:10.1145/3760389 [6] Mansoor Ahmed, Wali Ullah Khan, Afrasyab Khan Soofi, and Zhu Han. 2025. A Comprehensive Survey on RIS-Enhanced Physical Layer Security in UAV-Assisted Networks. IEEE Internet of Things Journal (2025). To be published. Manuscript submitted to ACM

32

Y. Deng et al.

[7] Haithem Alaya, Asma Ben Letaifa, and Abderrezak Rachedi. 2025. State of the Art and Taxonomy Survey on Federated Learning and Blockchain Integration in UAV Applications. Journal of Supercomputing 81 (2025), 655. doi:10.1007/s11227-025-07124-x [8] Omid Asghari, Naghmeh Ivaki, and Henrique Madeira. 2025. UAV Operations Safety Assessment: A Systematic Literature Review. ACM Computing Surveys 57, 10 (Oct. 2025), 1–37. [9] Murat Bakirci. 2024. Utilizing YOLOv8 for Enhanced Traffic Monitoring in Intelligent Transportation Systems (ITS) Applications. Digital Signal Processing 152 (Sept. 2024), 104594. [10] Asrar Ahmed Baktayan, Ammar Thabit Zahary, Axel Sikora, and Dominik Welte. 2024. Computational Offloading into UAV Swarm Networks: A Systematic Literature Review. EURASIP Journal on Wireless Communications and Networking 2024, 1 (2024), 69. [11] Yanis Bardes, Hassan Soubra, Zineb Noumir, and Amar Ramdane-Cherif. 2025. Federated Learning for UAV Perception Task: A Survey. TechRxiv (June 2025). doi:10.36227/techrxiv.174906148.89641328/v1 [12] Lingyi Cai, Jiacheng Wang, Ruichen Zhang, Yu Zhang, Tao Jiang, Dusit Niyato, Xianbin Wang, Abbas Jamalipour, and Xuemin Shen. 2025. Secure Physical Layer Communications for Low-Altitude Economy Networking: A Survey. https://arxiv.org/abs/2504.09153 [13] Hengsheng Chen, Yuan-Chyuan Andy Lin, Mingjian Fu, Lina Yao, and Quan Z. Sheng. 2026. A Survey on Reinforcement Learning Methods for UAV Systems. ACM Computing Surveys 58, 4, Article 103 (2026), 37 pages. doi:10.1145/3769426 [14] Jinsong Chen, Kai Wu, Jinping Niu, Yanyan Li, Pengfei Xu, and J. Andrew Zhang. 2025. Spectral and Energy Efficient Waveform Design for RIS-Assisted ISAC. IEEE Transactions on Communications 73, 1 (Jan. 2025), 158–172. [15] Xianhao Chen, Yiqin Deng, Haichuan Ding, Guandong Qu, Haixia Zhang, Pan Li, and Yuguang Fang. 2024. Vehicle as a Service (VaaS): Leverage Vehicles to Build Service Networks and Capabilities for Smart Cities. IEEE Communications Surveys & Tutorials 42, 3 (2024), 2048–2081. [16] Youssra Cheriguene, Wael Jaafar, Halim Yanikomeroglu, and Chaker Abdelaziz Kerrache. 2023. Data-Efficient Energy-Aware Participant Selection for UAV-Enabled Federated Learning. In 2023 IEEE 34th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC). IEEE, 1–7. [17] Hyunsang Cho, Seonghoon Yoo, Bang Chul Jung, and Joonhyuk Kang. 2024. Enhancing Battlefield Awareness: An Aerial RIS-Assisted ISAC System with Deep Reinforcement Learning. In Proceedings of the 2024 IEEE Military Communications Conference (MILCOM). 469–474. [18] Yiqin Deng, Zhengru Fang, Senkang Hu, Yanan Ma, Xiaoyu Guo, Haixia Zhang, and Yuguang Fang. 2026. UAV-Enabled Computing Power Networks: Design and Performance Analysis under Energy Constraints. IEEE Transactions on Mobile Computing (Jan. 2026). doi:10.1109/TMC. 2026.3655118 Early Access. [19] Yiqin Deng, Zhengru Fang, Senkang Hu, Yanan Ma, Haixia Zhang, and Yuguang Fang. 2025. UAV-Enabled Computing Power Networks: Task Completion Probability Analysis. In Proc. IEEE Global Communications Conference (GLOBECOM). Taipei, Taiwan, Dec. 8–12, 2025. [20] Yiqin Deng, Haixia Zhang, Xianhao Chen, and Yuguang Fang. 2024. UAV-Assisted MEC with an Expandable Computing Resource Pool: Rethinking the UAV Deployment. IEEE Wireless Communications 31, 5 (Oct. 2024), 110–116. [21] Yiqin Deng, Haixia Zhang, Xianhao Chen, and Yuguang Fang. 2024. UAV-Assisted Multi-Access Edge Computing With Altitude-Dependent Computing Power. IEEE Transactions on Wireless Communications 23, 8 (Aug. 2024), 9404–9418. doi:10.1109/TWC.2024.3362375 [22] Cole Dickerson, Ozgur Ozdemir, Mihail Sichitiu, and Ismail Guvenc. 2025. AERPAW Keysight N6841A RF Sensor UAV Localization Measurements with LOS/NLOS Information, July 2024. Dryad. doi:10.5061/dryad.vq83bk44h Dataset. [23] Haichuan Ding, Yuguang Fang, Xiaoxia Huang, Miao Pan, Pan Li, and Savo Glisic. 2017. Cognitive Capacity Harvesting Networks: Architectural Evolution Towards Future Cognitive Radio Networks. IEEE Communications Surveys & Tutorials 19, 3 (2017), 1902–1923. [24] Haichuan Ding, Chi Zhang, Ying Cai, and Yuguang Fang. 2018. Smart Cities on Wheels: A Newly Emerging Vehicular Cognitive Capability Harvesting Network for Data Transportation. IEEE Wireless Communications 25, 2 (2018), 160–169. [25] Wenyu Dong, Guangquan Xu, Shicheng Feng, Hongpeng Bai, Quan Z. Sheng, and Xi Zheng. 2026. Escorting the Confidentiality and Integrity of UAVs: What Exactly Can Trusted Execution Environment Offer? ACM Computing Surveys 58, 3, Article 72 (2026), 35 pages. doi:10.1145/3763788 [26] Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. 2017. CARLA: An Open Urban Driving Simulator. In Proceedings of the 1st Annual Conference on Robot Learning (Proceedings of Machine Learning Research, Vol. 78). PMLR, 1–16. https://proceedings. mlr.press/v78/dosovitskiy17a.html [27] Dian Fan, Feifei Gao, Bo Ai, Gongpu Wang, Zhangdui Zhong, Yansha Deng, and Arumugam Nallanathan. 2019. Channel Estimation and SelfPositioning for UAV Swarm. IEEE Transactions on Communications 67, 11 (2019), 7994–8007. [28] Zhengru Fang, Yanchao Deng, Guoliang Xue, and et al. 2025. PACP: Priority-Aware Collaborative Perception for Autonomous Driving. IEEE Transactions on Mobile Computing (2025). doi:10.1109/TMC.2024.3377829 Early Access 2024; vol. 23, no. 12, 2025. [29] Zhengru Fang, Zhenghao Liu, Jingjing Wang, Senkang Hu, Yu Guo, Yiqin Deng, and Yuguang Fang. 2025. Task-Oriented Communications for Visual Navigation with Edge-Aerial Collaboration. In Proc. IEEE Global Communications Conference (GLOBECOM). [30] Zhenyu Fu, Juan Liu, Yuyi Mao, Long Qu, Lingfu Xie, and Xijun Wang. 2025. Energy-Efficient UAV-Assisted Federated Learning: Trajectory Optimization, Device Scheduling, and Resource Management. IEEE Transactions on Network and Service Management 22, 2 (April 2025), 974–988. [31] Hui Gao, Ruohan Cao, Wenjun Xu, Caixia Yuan, and Hsiao-Hwa Chen. 2025. Space-Air-Ground Integrated Networks With Task-Driven Connected Intelligence. IEEE Wireless Communications 32, 2 (April 2025), 254–261. [32] Junhui Gao, Yan Pan, Qianru Wang, Wenzhe Hou, Yiqin Deng, Liangliang Jiang, and Yuguang Fang. 2026. TriDeliver: Cooperative Air-Ground Instant Delivery with UAVs, Couriers, and Crowdsourced Ground Vehicles. arXiv:2604.09049 [cs.RO] https://arxiv.org/abs/2604.09049 Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

33

[33] Tilahun M. Getu, Georges Kaddoum, and Mehdi Bennis. 2024. Semantic Communication: A Survey on Research Landscape, Challenges, and Future Directions. Proc. IEEE 112, 11 (Nov. 2024), 1649–1685. [34] Jing Guo, Salman Durrani, Xiangyun Zhou, and Zesong Fei. 2024. Design and Performance Analysis of Cache-Enabled Multicast in UAV-Assisted Cellular Networks. IEEE Transactions on Communications 72, 7 (July 2024), 4459–4475. [35] Xiaoyu Guo, Xiaowei Song, Dan Zeng, Zhen Dong, Xiang Yu, Lu Liu, and Yuguang Fang. 2025. Integrated Energy-Efficient Planning and Management Framework for Autonomous Long-Endurance Flight of Hydrogen Fuel Cell/Battery Hybrid UAVs. IEEE/ASME Transactions on Mechatronics (Jan. 2025). Early Access. [36] Qiaojian Han, Haixia Zhang, Yueheng Li, Dongfeng Yuan, and Xiaoyan Ma. 2025. Physics Informed Digital Twin for RIS-Assisted Wireless Communication System. IEEE Wireless Communications 32, 3 (June 2025), 106–112. [37] Shoufei Han, Xiaojing Liu, MengChu Zhou, Kun Zhu, Liang Zhao, Aiiad Albeshri, and Abdullah Abusorrah. 2024. Joint Association, Deployment and Flight Trajectory Optimization for Multi-UAV-Enabled Large-Scale Mobile Edge Computing. IEEE Transactions on Mobile Computing 23, 12 (Dec. 2024), 13207–13221. [38] Arash Heidari, Nima Jafari Navimipour, Mehmet Unal, and Guodao Zhang. 2023. Machine Learning Applications in Internet-of-Drones: Systematic Review, Recent Deployments, and Open Issues. ACM Computing Surveys 55, 12 (March 2023), 1–45. [39] Yunhao Hou, Bochao Zou, Min Zhang, Ran Chen, Shangdong Yang, Yanmei Zhang, Junbao Zhuo, Siheng Chen, Jiansheng Chen, and Huimin Ma. 2025. AGC-Drive: A Large-Scale Dataset for Real-World Aerial-Ground Collaboration in Driving Scenarios. arXiv preprint arXiv:2506.16371 (2025). doi:10.48550/arXiv.2506.16371 [40] Gangqiang Hu, Donglin Zhu, Jiaying Shen, Jialing Hu, Jianmin Han, and Taiyong Li. 2024. Reliable Incentive Mechanisms for Federated Learning in Drone Companies’ Collaborative Training. Drones 8, 10 (2024), 567. doi:10.3390/drones8100567 [41] Bang Huang, Eddine Youcef Belmekki, and Mohamed-Slim Alouini. 2026. High-Altitude Platforms in the Low-Altitude Economy: Bridging Communication, Computing, and Regulation. arXiv preprint arXiv:2602.19784 (2026). doi:10.48550/arXiv.2602.19784 [42] Yuhong Huang, Haiyu Ding, Weiyan Chen, Luting Kong, Wei Deng, Xin Li, Yang Liu, Guizhen Wang, and Liang Liu. 2025. Towards a Low-Altitude Aerial Intelligent Network: Vision, Challenges, and Key Technologies. China Communications 22, 9 (Sept. 2025), 1–21. [43] Min Hui, Jian Chen, Long Yang, Lu Lv, Hai Jiang, and Naofal Al-Dhahir. 2024. UAV-Assisted Mobile Edge Computing: Optimal Design of UAV Altitude and Task Offloading. IEEE Transactions on Wireless Communications 23, 10 (Oct. 2024), 13633–13647. [44] Peng Jiang, Ming Li, Rang Liu, Wei Wang, and Qian Liu. 2025. Joint Waveform and Beamforming Design in RIS-ISAC Systems: A Model-Driven Learning Approach. IEEE Transactions on Communications 73, 8 (Aug. 2025), 6469–6483. [45] Yihang Jiang, Xiaoyang Li, Guangxu Zhu, Hang Li, Jing Deng, Kaifeng Han, Chao Shen, Qingjiang Shi, and Rui Zhang. 2025. Integrated Sensing and Communication for Low Altitude Economy: Opportunities and Challenges. IEEE Communications Magazine (2025). Early Access. [46] Chanuka A. S. Hewa Kaluannakkage and Rajkumar Buyya. 2025. Incentive Based Federated Learning: Architectural Elements and Future Directions. arXiv preprint (2025). arXiv:2510.14208v2. [47] Ayshika Kapoor and Dheeraj Kumar. 2025. Federated Learning for Urban Sensing Systems: A Comprehensive Survey on Attacks, Defences, Incentive Mechanisms, and Applications. IEEE Communications Surveys & Tutorials 27, 2 (April 2025), 1293–1325. [48] Pawan Kumar, Kunwar Pal, and Mahesh Govil. 2026. Comprehensive Review of Path Planning Techniques for Unmanned Aerial Vehicles (UAVs). ACM Computing Surveys 58, 3, Article 73 (2026), 44 pages. doi:10.1145/3737280 [49] Joohyung Lee, Faranaksadat Solat, Tae Yeon Kim, and H. Vincent Poor. 2024. Federated Learning-Empowered Mobile Network Management for 5G and Beyond Networks: From Access to Core. IEEE Communications Surveys & Tutorials 26, 3 (2024), 2176–2212. [50] Jian Li, Lan Zhang, Kaiping Xue, Yuguang Fang, and Qibin Sun. 2023. Secure Transmission by Leveraging Multiple Intelligent Reflecting Surfaces in MISO Systems. IEEE Transactions on Mobile Computing 22, 4 (2023), 2387–2401. [51] Peishi Li, Ming Li, Rang Liu, Qian Liu, and A. Lee Swindlehurst. 2025. MIMO-OFDM ISAC Waveform Design for Range-Doppler Sidelobe Suppression. IEEE Transactions on Wireless Communications 24, 2 (Feb. 2025), 1001–1015. [52] Xingyu Li, Wenzhe Zhang, Linfeng Liu, and Jia Xu. 2025. Exploring the Robustness: Hierarchical Federated Learning Framework for Object Detection of UAV Cluster. IEEE Transactions on Mobile Computing 24, 10 (Oct. 2025), 9489–9505. [53] Yongxiao Li, Feroz Khan, Manzoor Ahmed, Aized Amin Soofi, Wali Ullah Khan, Chandan Kumar Sheemar, Muhammad Asif, and Zhu Han. 2025. RIS-Based Physical Layer Security for Integrated Sensing and Communication: A Comprehensive Survey. (2025). doi:10.48550/arXiv.2503.17721 [54] Tianhao Liang, Nanchi Su, Yuqi Ping, Guangyu Lei, Xinglin Chen, Longyu Zhou, Tingting Zhang, Qinyu Zhang, and Tony Q. S. Quek. 2026. UAV Control and Communication Enabled Low-Altitude Economy: Challenges, Resilient Architecture and Co-Design Strategies. arXiv preprint arXiv:2604.04044 (2026). doi:10.48550/arXiv.2604.04044 [55] Yuemin Liu, Shuai Liu, Xin Liu, Zechen Liu, and Tariq S. Durrani. 2023. Sensing Fairness-Based Energy Efficiency Optimization for UAV-Enabled Integrated Sensing and Communication. IEEE Wireless Communications Letters 12, 10 (2023), 1702–1706. [56] Zhiyan Liu, Xu Chen, Hai Wu, Zhanwei Wang, Xianhao Chen, Dusit Niyato, and Kaibin Huang. 2025. Integrated Sensing and Edge AI: Realizing Intelligent Perception in 6G. IEEE Communications Surveys & Tutorials (July 2025). Early Access. [57] Zhiyan Liu and Kaibin Huang. 2025. Semantic-Relevance Based Sensor Selection for Edge-AI Empowered Sensing Systems. arXiv preprint arXiv:2503.12785 (2025). doi:10.48550/arXiv.2503.12785 [58] Zhiyan Liu, Qiao Lan, and Kaibin Huang. 2025. Over-the-Air Fusion of Sparse Spatial Features for Integrated Sensing and Edge AI over Broadband Channels. IEEE Transactions on Wireless Communications 24, 4 (2025), 2999–3013. doi:10.1109/TWC.2025.3527331 Manuscript submitted to ACM

34

Y. Deng et al.

[59] Mingquan Lu, Zheng Yao, Yuan Shen, Xingxing Li, and Zhipeng Wang. 2025. Space-Air-Ground Collaborative Positioning, Navigation and Timing Architecture and Key Technologies for Low-Altitude Economy. China Communications 22, 9 (Sept. 2025), 48–80. [60] Yanan Ma, Yiqin Deng, Zhengru Fang, Longzhi Yuan, Xianhao Chen, and Yuguang Fang. 2026. RAISE: Optimizing RIS Placement to Maximize Task Throughput in Multi-Server Vehicular Edge Computing. IEEE Transactions on Wireless Communications 25 (2026), 9185–9199. [61] Yanan Ma, Zhengru Fang, Yihang Tao, Yu Guo, Yiqin Deng, Xianhao Chen, and Yuguang Fang. 2026. Birdcast: Interest-Aware BEV Multicasting for Infrastructure-Assisted Collaborative Perception. arXiv preprint arXiv:2604.00701 (2026). [62] Vuk Marojevic, Ismail Guvenc, Rudra Dutta, Mihail L. Sichitiu, and Brian Floyd. 2020. Advanced Wireless for Unmanned Aerial Systems: 5G Standardization, Research Challenges, and AERPAW Experimentation Platform. IEEE Vehicular Technology Magazine 15, 2 (2020), 22–30. [63] Siqi Meng, Shaohua Wu, Jiaming Zhang, Junlan Cheng, Haibo Zhou, and Qinyu Zhang. 2025. Semantics-Empowered Space-Air-Ground-Sea Integrated Network: New Paradigm, Frameworks, and Challenges. IEEE Communications Surveys & Tutorials 27, 1 (Feb. 2025), 140–183. [64] Silvia Mura, Dario Tagliaferri, Marouan Mizmizi, Umberto Spagnolini, and Athina Petropulu. 2025. Optimized Waveform Design for OFDM-Based ISAC Systems Under Limited Resource Occupancy. IEEE Transactions on Wireless Communications 24, 6 (June 2025), 5241–5254. [65] Huan X. Nguyen, Kexuan Sun, Duc To, Quoc-Tuan Vien, and Tuan Anh Le. 2025. Digital Twin for O-RAN Toward 6G. IEEE Communications Magazine 63, 3 (2025), 174–181. [66] Zhaolong Ning, Hao Hu, Xiaojie Wang, Lei Guo, Song Guo, Guoyin Wang, and Xinbo Gao. 2023. Mobile Edge Computing and Machine Learning in the Internet of Unmanned Aerial Vehicles: A Survey. ACM Computing Surveys 56, 1 (Aug. 2023), 1–31. [67] O-RAN next Generation Research Group (nGRG). 2024. Research Report on Digital Twin RAN Use Cases. Technical Report RR-2024-07. O-RAN Alliance. https://mediastorage.o-ran.org/ngrg-rr/nGRG%20RS01%20Digital%20Twin%20Use%20Case%20Research%20Report%20v1.0.pdf Release date: 2024.05. [68] Sicong Peng, Bin Li, Lei Liu, Zesong Fei, and Dusit Niyato. 2025. Trajectory Design and Resource Allocation for Multi-UAV-Assisted Sensing, Communication, and Edge Computing Integration. IEEE Transactions on Communications 73, 4 (Jan. 2025), 2847–2861. [69] Quoc-Viet Pham, Ming Zeng, Thien Huynh-The, Zhu Han, and Won-Joo Hwang. 2022. Aerial Access Networks for Federated Learning: Applications and Challenges. IEEE Network 36, 3 (May 2022), 159–166. [70] Minh K. Quan, Pubudu N. Pathirana, Mayuri Wijayasundara, Sujeeva Setunge, Dinh C. Nguyen, Christopher G. Brinton, David J. Love, and H. Vincent Poor. 2025. Federated Learning for Cyber Physical Systems: A Comprehensive Survey. IEEE Communications Surveys & Tutorials (2025). Early Access. [71] Kathiroli Raja, Kottilingam Kottursamy, Vishal Ravichandran, Sivaganesh Balaganesh, Kapal Dev, Lewis Nkenyereye, and Gunasekaran Raja. 2025. An Efficient 6G Federated Learning-Enabled Energy-Efficient Scheme for UAV Deployment. IEEE Transactions on Vehicular Technology 74, 2 (Feb. 2025), 2057–2066. [72] Amir Hossein Fahim Raouf, Donggu Lee, Mushfiqur Rahman, Saad Masrur, Gautham Reddy, Cole Dickerson, Md Sharif Hossen, Sergio Vargas Villar, Anıl Gürses, Simran Singh, Sung Joon Maeng, Martins Ezuma, Christopher Roberts, Mohamed Rabeek Sarbudeen, Thomas J. Zajkowski, Magreth Mushi, Ozgur Ozdemir, Ram Asokan, Ismail Guvenc, Mihail L. Sichitiu, and Rudra Dutta. 2025. Wireless Datasets for Aerial Networks. arXiv preprint arXiv:2510.08752 (2025). doi:10.48550/arXiv.2510.08752 [73] Daniel Romero, Pham Q. Viet, and Raju Shrestha. 2024. Aerial Base Station Placement via Propagation Radio Maps. IEEE Transactions on Communications 72, 9 (Sept. 2024), 5349–5364. [74] Shital Shah, Debadeepta Dey, Chris Lovett, and Ashish Kapoor. 2017. Airsim: High-fidelity visual and physical simulation for autonomous vehicles. In Field and service robotics: Results of the 11th international conference. Springer, 621–635. [75] Arnolnt Spyros, Periklis Chatzimisios, Dimitrios Kavallieros, Theodora Tsikrika, Stefanos Vrochidis, and Yiannis Kompatsiaris. 2026. A Comprehensive Survey and Taxonomy of Cybersecurity Challenges and Proactive Measures for IoD. ACM Computing Surveys 58, 7 (2026), 1–35. [76] Geng Sun, Yixian Wang, Zemin Sun, Qingqing Wu, Jiawen Kang, Dusit Niyato, and Victor C. M. Leung. 2024. Multi-Objective Optimization for Multi-UAV-Assisted Mobile Edge Computing. IEEE Transactions on Mobile Computing 23, 12 (Dec. 2024), 14803–14820. [77] Yan Sun, Yinqiu Liu, Shaoyong Guo, Ruichen Zhang, Jiacheng Wang, Xuesong Qiu, Geng Sun, Weifeng Gong, Dusit Niyato, and Qihui Wu. 2025. A Synergy of Computing Power Networks and Low-Altitude Economy Intelligent Communications: Challenges, Design Principles, and Research Directions. arXiv preprint arXiv:2509.23810 (2025). doi:10.48550/arXiv.2509.23810 [78] Yao Tang, Guangxu Zhu, Wei Xu, Man Hon Cheung, Tat-Ming Lok, and Shuguang Cui. 2025. Integrated Sensing, Computation, and Communication for UAV-Assisted Federated Edge Learning. IEEE Transactions on Wireless Communications 24, 4 (April 2025), 2647–2662. [79] Akbar Telikani, Arupa Sarkar, Bo Du, Fendy Santoso, Jun Shen, Jun Yan, Jianming Yong, and Emily Yap. 2025. Unmanned Aerial Vehicle-Aided Intelligent Transportation Systems: Vision, Challenges, and Opportunities. IEEE Communications Surveys & Tutorials (Jan. 2025). Early Access. [80] The ns-3 Project. 2026. ns-3.47 Released. https://www.nsnam.org/news/2026/02/16/ns-3-47-released.html Official release page. [81] Chen Wang, Xiao Tang, Daosen Zhai, Ruonan Zhang, Nurzhan Ussipov, and Yan Zhang. 2025. Energy-Efficient Federated Learning Through UAV Edge Under Location Uncertainties. IEEE Transactions on Network Science and Engineering 12, 1 (Jan. 2025), 223–236. [82] Jiahao Wang, Xiangyu Cao, Jiaru Zhong, Yuner Zhang, Zeyu Han, Haibao Yu, Chuang Zhang, Lei He, Shaobing Xu, and Jianqiang Wang. 2026. Griffin: Aerial-Ground Cooperative Detection and Tracking Dataset and Benchmark. In AAAI, Vol. 40. 9867–9875. [83] Jie Wang, Qiuming Zhu, Zhipeng Lin, Junting Chen, Guoru Ding, Qihui Wu, Guochen Gu, and Qianhao Gao. 2024. Sparse Bayesian Learning-Based Hierarchical Construction for 3D Radio Environment Maps Incorporating Channel Shadowing. IEEE Transactions on Wireless Communications 23, 10 (2024), 14560–14574. doi:10.1109/TWC.2024.3416447 Manuscript submitted to ACM

Collaborative Air-Ground SCCSI for LAE

35

[84] Qunshu Wang, Shaoyong Guo, Celimuge Wu, Chengwen Xing, Nan Zhao, Dusit Niyato, and George K. Karagiannidis. 2025. STAR-RIS Aided Covert Communication in UAV Air-Ground Networks. IEEE Journal on Selected Areas in Communications 43, 1 (2025), 245–259. [85] Xiaojie Wang, Qi Guo, Zhaolong Ning, Lei Guo, Guoyin Wang, Xinbo Gao, and Yan Zhang. 2024. Integration of Sensing, Communication, and Computing for Metaverse: A Survey. ACM Computing Surveys 56, 10 (May 2024), 1–38. [86] Xiangyu Wang, Donglin Yang, Ziqin Wang, Hohin Kwan, Jinyu Chen, Wenjun Wu, Hongsheng Li, Yue Liao, and Si Liu. 2025. Towards Realistic UAV Vision-Language Navigation: Platform, Benchmark, and Methodology. In The Thirteenth International Conference on Learning Representations. [87] Xin Wang, Bo Yi, Saru Kumari, Chien-Ming Chen, Shalli Rani, Keqin Li, and Jianhui Lv. 2025. Bridging Terrestrial and Non-Terrestrial Networks: A Novel Architecture for Space-Air-Ground-Sea Integration System. IEEE Wireless Communications 32, 3 (June 2025), 20–27. [88] Xiaojie Wang, Zhonghui Zhao, Ling Yi, Zhaolong Ning, Lei Guo, F. Richard Yu, and Song Guo. 2024. A Survey on Security of UAV Swarm Networks: Attacks and Countermeasures. ACM Computing Surveys 57, 3 (Nov. 2024), 1–37. [89] Yixian Wang, Geng Sun, Zemin Sun, Jiacheng Wang, Jiahui Li, Changyuan Zhao, Jing Wu, Shuang Liang, Minghao Yin, Pengfei Wang, Dusit Niyato, Sumei Sun, and Dong In Kim. 2025. Toward Realization of Low-Altitude Economy Networks: Core Architecture, Integrated Technologies, and Future Directions. IEEE Transactions on Cognitive Communications and Networking 11, 5 (2025), 2788–2820. [90] Dexing Wei, Lun Zhang, Quan Liu, Hao Chen, and Jian Huang. 2024. UAV Swarm Cooperative Dynamic Target Search: A MAPPO-Based Discrete Optimal Control Method. Drones 8, 6 (2024), 214. doi:10.3390/drones8060214 [91] Dingzhu Wen, Yong Zhou, Xiaoyang Li, Yuanming Shi, Kaibin Huang, and Khaled B. Letaief. 2025. A Survey on Integrated Sensing, Communication, and Computation. IEEE Communications Surveys & Tutorials (2025). Early Access. [92] Sheng Wu, Ning Chen, Ailing Xiao, Haoge Jia, Chunxiao Jiang, and Peiying Zhang. 2024. AI-Enabled Deployment Automation for 6G Space-AirGround Integrated Networks: Challenges, Design, and Outlook. IEEE Network 38, 6 (Nov. 2024), 219–226. [93] Xiaoyu Xia, Sheik Mohammad Mostakim Fattah, and Muhammad Ali Babar. 2023. A Survey on UAV-Enabled Edge Computing: Resource Management Perspective. ACM Computing Surveys 56, 3 (Oct. 2023), 1–36. [94] Bingnan Xiao, Xichen Yu, Wei Ni, Xin Wang, and H. Vincent Poor. 2025. Over-the-Air Federated Learning: Status Quo, Open Challenges, and Future Directions. Fundamental Research 5, 4 (2025), 1710–1724. [95] Yue Xiao, Ziqiang Ye, Mingming Wu, Haoyun Li, Ming Xiao, Mohamed-Slim Alouini, Akram Al-Hourani, and Stefano Cioni. 2024. Space-AirGround Integrated Wireless Networks for 6G: Basics, Key Technologies, and Future Trends. IEEE J. Sel. Areas Commun. 42, 12 (Dec. 2024), 3327–3354. [96] Jinpeng Xu, Lin Bai, Xin Xie, and Lin Zhou. 2025. Collaborative Secret and Covert Communications for Multi-User Multi-Antenna Uplink UAV Systems: Design and Optimization. IEEE Transactions on Wireless Communications 24, 7 (2025), 6020–6035. [97] Qichao Xu, Ruidong Li, Yihao Qi, Zhou Su, and Dongfeng Fang. 2025. Trust-Enhanced Game Incentive for Secure Quantum Federated Learning in UAV-Assisted Wireless Networks. IEEE Journal on Selected Areas in Communications (2025). Early Access. [98] Weijie Yuan, Yuanhao Cui, Jiacheng Wang, Fan Liu, Lin Zhou, Geng Sun, Tao Xiang, Jie Xu, Shi Jin, Dusit Niyato, Sinem Coleri, Sumei Sun, Shiwen Mao, Abbas Jamalipour, Dong In Kim, Mohamed-Slim Alouini, and Xuemin Shen. 2025. From Ground to Sky: Architectures, Applications, and Challenges Shaping Low-Altitude Wireless Networks. arXiv preprint arXiv:2506.12308 (2025). doi:10.48550/arXiv.2506.12308 [99] Yongs Zeng, Qingqing Wu, and Rui Zhang. 2019. Accessing from the Sky: A Tutorial on UAV Communications for 5G and Beyond. Proceedings of the IEEE 107, 12 (2019), 2327–2375. [100] Hongming Zhang and Lajos Hanzo. 2020. Federated Learning Assisted Multi-UAV Networks. IEEE Transactions on Vehicular Technology 69, 11 (Nov. 2020), 14104–14109. [101] Jianshan Zhang, Xu Yang, Rui Ming, Haibo Luo, Xiaoding Wang, and Md. Jalil Piran. 2025. A Time-Driven UAV Deployment and Content Caching Approach for Multi-UAV-Enabled Consumer Electronics Network. IEEE Transactions on Consumer Electronics 71, 2 (2025), 3941–3951. [102] Lan Zhang, Li Yan, Yuguang Fang, Xuming Fang, and Xiaoxia Huang. 2019. A Machine Learning-Based Defensive Alerting System Against Reckless Driving in Vehicular Networks. IEEE Transactions on Vehicular Technology 68, 12 (2019), 12227–12238. [103] Ruichen Zhang, Hongyang Du, Dusit Niyato, Jiawen Kang, Zehui Xiong, Abbas Jamalipour, Ping Zhang, and Dong In Kim. 2024. Generative AI for Space-Air-Ground Integrated Networks. IEEE Wireless Communications 31, 6 (2024), 10–20. [104] Xuhui Zhang, Wenchao Liu, Jinke Ren, Huijun Xing, Gui Gui, Yanyan Shen, and Shuguang Cui. 2025. Latency Minimization for UAV-Enabled Federated Learning: Trajectory Design and Resource Allocation. IEEE Internet of Things Journal 12, 14 (July 2025), 27097–27112. [105] Qin Zhen, Shoushuai He, Hai Wang, Yuben Qu, Haipeng Dai, Fei Xiong, Zhenhua Wei, and Hailong Li. 2024. Air-Ground Collaborative Mobile Edge Computing: Architecture, Challenges, and Opportunities. China Communications 21, 5 (2024), 1–16. [106] Jie Zheng, Jipeng Xu, Hongyang Du, Dusit Niyato, Jiawen Kang, Jiangtian Nie, and Zheng Wang. 2024. Trust Management of Tiny Federated Learning in Internet of Unmanned Aerial Vehicle Networks. IEEE Internet of Things Journal 11, 12 (2024), 21046–21060. [107] Zijian Zheng, Yansha Deng, Wenqiang Yi, Hyundong Shin, and Arumugam Nallanathan. 2025. Over-the-Air Computation Enabled SemiAsynchronous Wireless Federated Learning. IEEE Transactions on Communications 73, 10 (2025), 8919–8936. [108] Yilin Zhu, Yang Kong, Yingrui Jie, Shiyou Xu, and Hui Cheng. 2023. GRACO: A Multimodal Dataset for Ground and Aerial Cooperative Localization and Mapping. IEEE Robotics and Automation Letters 8, 2 (Feb. 2023), 966–973.

Manuscript submitted to ACM