Survey on Unmanned Aerial Vehicle Networks: A Cyber Physical System Perspective

Haijun Wang; Haitao Zhao; Jiao Zhang; Dongtang Ma; Jiaxun Li; Jibo Wei

Source-linked AI summary

Survey on Unmanned Aerial Vehicle Networks: A Cyber Physical System Perspective

Haijun Wang, Haitao Zhao, Jiao Zhang, Dongtang Ma, Jiaxun Li, Jibo Wei

arXiv:1812.06821v1Dec 17, 2018cs.NI

TL;DR

UAV networks combine communication, computation, and control but face mobility and resource constraints that complicate complex missions. This survey organizes the field across CPS components, coupling effects, and network hierarchies, offering a CPS perspective on cross-disciplinary challenges.

Problem
UAV mobility creates intermittent links and fluid topologies, while resource-hungry tasks exceed UAV computation and battery constraints.
Method
The paper systematically surveys communication, computation, control, their coupling effects, and UAV networks across cell, system, and system-of-system hierarchies.
Results
The survey explicitly demonstrates architectures, key techniques, applications, and mutual inspirations across UAV network components and CPS hierarchies.
Takeaways & Limitations
The survey provides beginners with a tutorial and researchers with a CPS perspective for addressing cross-disciplinary UAV network challenges.

Abstract

from arXiv · show

Unmanned aerial vehicle (UAV) networks are playing an important role in various areas due to their agility and versatility, which have attracted significant attention from both the academia and industry in recent years. As an integration of the embedded systems with communication devices, computation capabilities and control modules, the UAV network could build a closed loop from data perceiving, information exchanging, decision making to the final execution, which tightly integrates the cyber processes into the physical devices. Therefore, the UAV network could be considered as a cyber physical system (CPS). Revealing the coupling effects among the three interacted components in this CPS system, i.e., communication, computation and control, is envisioned as the key to properly utilize all the available resources and hence improve the performance of the UAV networks. In this paper, we present a comprehensive survey on the UAV networks from a CPS perspective. Firstly, we respectively research the basics and advances with respect to the three CPS components in the UAV networks. Then we look inside to investigate how these components contribute to the system performance by classifying the UAV networks into three hierarchies, i.e., the cell level, the system level, and the system of system level. Further, the coupling effects among these CPS components are explicitly illustrated, which could be enlightening to deal with the challenges in each individual aspect. New research directions and open issues are discussed at the end of this survey. With this intensive literature review, we try to provide a novel insight into the state-of-the-art in the UAV networks.

I. INTRODUCTION · A. CPS can inspire the UAV networks

UAVs are increasingly organized into cooperative networks because their agility and versatility enable broad, flexible, and robust operations. Viewing these networks as cyber physical systems highlights the coupled roles of communication, computation, and control in improving overall performance.

I. INTRODUCTION: UAVs support military and civilian applications through agile, versatile, low-cost, and easily deployable platforms.They can proactively capture information across large spatial and temporal scales and assist other unmanned or manned systems.
I. INTRODUCTION: Cooperative UAV networks provide wider coverage, greater flexibility, and robustness through redundancy compared with individual UAVs.This shift from individual to cooperative operation also creates more intractable challenges than those of single-UAV systems.
I. INTRODUCTION: Communication, computation, and control modules should be designed integrally because they affect and benefit from one another through coupling.For example, UAVs may change their locations through flight control to restore line-of-sight links when communication is degraded by shadow effects.
A. CPS can inspire the UAV networks: A UAV network forms a closed loop from data perceiving and information exchanging through decision making to final execution.Embedding these cyber processes into physical devices makes the network a cyber physical system with tightly coupled cyber and physical domains.
A. CPS can inspire the UAV networks: The survey examines UAV networks by connecting communication, computation, and control across cell-level, system-level, and system-of-system perspectives.Its structure includes backgrounds, autonomy, component coupling, and challenges or open issues.
A. CPS can inspire the UAV networks: The CPS perspective is presented as promising for significantly improving the performance of the whole UAV network system.This perspective motivates investigating coupling effects rather than treating communication, computation, and control as isolated aspects.

B. Existing surveys and motivations … A. Taking the UAV networks as CPS

Existing surveys typically isolate communication, computation, control, applications, or security, leaving their guidance and coupling in UAV networks insufficiently explained. This paper addresses that gap through the first comprehensive CPS-perspective survey, organized around the three cyber components, their coupling effects, and UAV-network hierarchies.

B. Existing surveys and motivations: Existing UAV-network surveys usually examine either cyber issues or physical issues, such as communication, computation platforms, or control-related matters.This single-perspective treatment limits cross-disciplinary understanding of UAV networks.
B. Existing surveys and motivations: CPS surveys commonly emphasize industrial-system design, implementation, or security rather than the specific cyber–physical relations of UAV networks.Reviewed domains include energy, transportation, production, and other CPS applications.
B. Existing surveys and motivations: UAV–CPS surveys use UAVs as sensors and actuators for specialized missions but overlook broader cyber issues and coupling effects among components.Examples include cooperative target seeking and contour mapping.
C. Paper organization: The paper proposes a systematic UAV-network review from a CPS perspective, using a roadmap that combines three cyber components, their coupling effects, and UAV-network hierarchies.The paper investigates communication, computation, and control issues and advances within this structure.
C. Paper organization: The authors present the work as the first survey to comprehensively investigate key UAV-network issues from a CPS perspective.They intend the survey to provide novel insight and suggestions for addressing challenges in the field.
A. Taking the UAV networks as CPS: The UAV network forms a CPS loop in which sensing supplies data, onboard computers process information, decisions control flight and missions, and actions affect the UAVs and outside world.This describes the integration of sensing, computing, decision making, and execution in UAVs.
A. Taking the UAV networks as CPS: A UAV-based CPS comprises hardware in the physical domain and software in the cyber domain, with hardware sensing, actuating, computing, and communicating while software analyzes data and makes decisions.Hardware may include sensors, actuators, computation chips, and communication equipment; software may include embedded operating systems and applications.

B. Autonomy in the UAV networks … 1) Internal challenges:

UAV networks integrate sensing, communication, computation, and control into a CPS closed loop, while autonomy determines how responsibilities are divided between ground control stations and onboard controllers. Communication supports information exchange and coordination but faces mobility-, energy-, spectrum-, and interference-related challenges.

B. Autonomy in the UAV networks: UAV networks commonly connect one or more UAVs to a ground control station that monitors status, sets waypoints, and issues commands.The GCS may be a human pilot or a high-performance computer.
B. Autonomy in the UAV networks: In human-in-the-loop operation, UAVs cannot operate independently and flight control and mission completion rely on the pilot’s skills and the GCS–UAV data link.The GCS mainly performs computation, while communication remains central to operation.
III. CPS COMPONENTS IN THE UAV NETWORKS: A UAV network forms a CPS by integrating sensing, communication, computation, and control into a closed loop for resource allocation and mission completion.Sensing introduces physical-world data, while communication distributes and shares information for analysis and decisions.
A. Communication: Communication carries data from sensing to computation and control, supplying computational inputs and distributing or sharing computational outputs.This data flow supports both centralized and distributed decision-making scenarios.
A. Communication: UAV communication disseminates observations, tasks, and control information and can improve coordination and safety, but its demands vary substantially across applications.Robust communication is difficult because of UAV mobility and energy limits, spectrum scarcity, and malicious interference.
(a) Communication demands: Communication demands shift with autonomy: low-autonomy networks depend heavily on UAV–GCS links, whereas fully autonomous networks rely more on UAV–UAV information exchange.Control-related, mission-oriented, and normal data messages also differ in delay, delay variance, and bandwidth requirements.
(b) Communication challenges: UAV mobility spans three-dimensional operating space, varied pitch, roll, and yaw attitudes, and speeds from static to high speed, expanding operating scope while complicating communication.Static UAVs suit applications such as aerial base stations and objective surveillance, but mobility is preferred in many cases because it can reduce deployment counts.
1) Internal challenges:: Energy limits constrain transmission power and connectivity, while returning UAVs for charging and replacement intensify topology changes and contribute to intermittent links, fluid topology, Doppler effects, antenna-alignment complexity, and vanishing nodes.UAVs also face spectrum variations, scarcity, and outside interference when using unlicensed bands under fixed spectrum assignment policies.

2) External challenges: … B. Computation

The section surveys external challenges and communication foundations in UAV networks, emphasizing spectrum variability, heterogeneous links and channels, antenna and mobility design, and computation as the core of the closed loop. It identifies cognitive radio, aerial mobility models, and computation platforms, decision entities, architectures, and intelligent algorithms as key considerations.

2) External challenges:: UAVs face spatial and temporal spectrum variation across mission terrains, while cellular, Wi-Fi, Bluetooth, and sensor networks compete for the same spectrum.These conditions create spectrum scarcity and outside interference in UAV mission areas.
2) External challenges:: Cognitive radio enables UAVs to opportunistically access licensed or unlicensed bands and adapt working frequencies to mitigate spectrum variation and interference.Dynamic spectrum access supports applications including traffic surveillance, crop monitoring, border patrolling, and disaster management.
(c) Communication basics: UAV communication design must consider communication links, channel models, antenna designs, and mobility models.These basics collectively shape communication-related evaluation and system design.
1) Communication links:: CNPC links ensure safe UAV operation, whereas data links support mission communications with the GCS and among UAVs, including image and video transmission.Both link types include UAV-ground and UAV-UAV channels with distinct characteristics.
2) Channel models:: UAV-ground channels depend on three-dimensional terrain and obstacles, while UAV-UAV channels are mainly line-of-sight but can experience minimal multipath fading and high Doppler frequencies.Two-ray, Markov, frequency-shift estimation, and diversity approaches are discussed, while more precise mission-area models remain needed.
3) Antenna designs:: Antenna design requires jointly considering antenna type and number, combining omnidirectional coverage with directional transmission when appropriate.Omnidirectional antennas radiate in all directions, whereas directional antennas send signals through a desired direction.
4) Mobility models:: Traditional MANET mobility models can be adapted for UAVs but cannot capture correlated aerial mobility and smooth turns constrained by mechanics and aerodynamics.Circular, smooth-turn, flight-plan-based, and multi-tier models are among the alternatives suited to different UAV applications.
B. Computation: Computation is the core of the UAV network’s closed loop, supporting data analysis, decision making, feedback, and the formation of memories, knowledge, and experiences.Key issues include onboard platforms, GCS-versus-UAV decision entities, distributed or centralized architectures, and intelligent optimization algorithms.

(a) Computation platforms … (c) Intelligent algorithms

The survey presents UAV computation as a combination of hardware, software, decision-making entities, and intelligent algorithms that together determine autonomy, mission execution, and communication demands. It contrasts basic low-autonomy platforms with higher-autonomy systems that support distributed decisions and intelligent mission functions.

(a) Computation platforms: UAV computation platforms combine chip modules, embedded software, operating systems, and functional algorithms, with configurations varying by application scenario and autonomy.The platforms include the MCU and software supporting sensors, communications, computation, flight control, and mission functions.
1) Hardware platform:: Civilian UAVs commonly use ARM-based STM32 or Mega2560 MCUs with low frequencies and limited computing power, supporting basic flight control but not high-speed parallel intelligence.STM32 operates at about 200MHz, while Atmel chips can operate as low as 20MHz.
1) Hardware platform:: Qualcomm, Intel, Nvidia, Leadcore Technology, and ZEROTECH introduced computation suites for smart drones, including Snapdragon Flight, Edison for Arduino, Jetson TX1/TX2, and LC1.These platforms were developed to promote UAV intelligence.
2) Software platform:: UAV software platforms include firmware, functional software, algorithms, and operating systems that connect hardware with sensors, communications, computation, and complex AI applications.Examples of operating systems include Linux, Windows, and Macintosh; deep learning can support high-performance CPU use.
(b) Decision making entity: Decision-making entities may be UAVs or the GCS, with low-autonomy UAVs relying on continual GCS decisions and higher-autonomy UAVs operating independently.Centralized decision making is simpler and reduces onboard processing requirements, whereas distributed decision making can avoid single-point failure and enable parallel processing.
(b) Decision making entity: Decision-making manner determines communication demands: consensus-based distributed decisions require more communication than decisions based only on individual UAV status.Consensus requires information exchange among UAVs, while individual-based decisions depend on each UAV’s own status.
(c) Intelligent algorithms: Intelligent algorithms support path planning, task allocation, machine vision, image recognition, and adaptive decision making through bio-inspired methods and supervised, unsupervised, reinforcement, and transfer learning.Artificial intelligence can reduce dependence on GCS-UAV communication by improving autonomy, recognition, and adaptability to dynamic environments.
(c) Intelligent algorithms: Bio-inspired methods model natural collective behavior: PSO, ACO, and GA have been applied to UAV path planning and task allocation, with GA producing superior trajectories to PSO in one comparison.The comparison concerned parallel GA and PSO for real-time path planning.

C. Control … (b) Flight control algorithms

The paper frames UAV control as the final stage translating sensing, communication, and computation decisions into actuator instructions for task execution. It surveys flight controllers and flight-control algorithms, emphasizing stronger computation, lower energy use, redundancy, and combined control methods.

C. Control: Control closes the UAV network’s loop by translating prior sensing, communication, and computation decisions into actuator instructions that affect the physical world.Flight control is presented as a precondition for task execution.
(a) Flight controllers: Flight controllers or autopilot systems provide autonomous attitude stabilization, waypoint generation, and mission planning through coordinated hardware and software.They enable UAVs to progress from remotely controlled aircraft to autonomous and intelligent aircraft.
1) Paparazzi:: Paparazzi supports multi-copters, fixed-wing aircraft, helicopters, and hybrid aircraft with dynamic mission-state flight plans and waypoint variables for operator-independent missions.It is described as the first open-source drone project encompassing autopilot and ground-station software.
2) PIXHAWK:: Pixhawk combines PX4-FMU and PX4-IO on one computer-vision-based autopilot module, while Phenix Pro adds RTOS, ROS, extensive interfaces, and FPGA acceleration for vision and deep-neural-network applications.Phenix Pro supports more than twenty interfaces, including mmWave radar and thermal-camera interfaces.
3) Phenix Pro: / 4) OcPoC: / 5) DJI A2:: OcPoC uses a Xilinx Zynq processor and FPGA performance for enhanced I/O and processing, runs ArduPilot, and processes sensor data in real time; DJI targets reliable multirotor autopilots at aerial-photography users.DJI A2 is not open-source, and its processor and sensor details are unavailable.
6) NAVIO2:: Navio2 improves performance and redundancy with dual IMU chips and a Raspberry Pi board, enabling Pixhawk functions plus more powerful computation for higher-level missions.It is the latest version of the Navio autopilot family.
7) Trinity:: AscTec Trinity provides up to three redundancy levels, automatically detecting and compensating for errors and handling sub-autopilot failures through total autopilot-system redundancy.This makes Trinity more reliable, robust, and safe than systems with only sub-module redundancy.
(b) Flight control algorithms: Quadrotor flight-control algorithms address nonlinear instability, fast-response requirements under disturbances, and under-actuation; linear and nonlinear methods can be combined because no single algorithm controls quadrotors optimally.Examples include PID with sliding mode control, feedback linearization with backstepping, and MPC with robust feedback linearization.

IV. UAV NETWORKS OF THREE CPS HIERARCHIES · A. Cell level · UCL USL USoS

The survey organizes mission-oriented UAV networks into three CPS hierarchies—UCL, USL, and USoS—whose increasing scale and complexity add communication and platform-level integration. At the cell level, a single UAV forms a self-contained CPS through sensing, computing, communicating, controlling, and extending capabilities.

IV. UAV NETWORKS OF THREE CPS HIERARCHIES: Mission scale and complexity determine whether deployment uses a single UAV, a UAV swarm, or a more systematic multi-platform network.Single-UAV networks suit simple, low-scale missions, while interactive UAV swarms support larger, more complicated missions through redundancy and swarm intelligence.
IV. UAV NETWORKS OF THREE CPS HIERARCHIES: USoS is an organic combination of UAV types represented as “hardware + software + communication network + platform.”The hierarchy evolution of UAV networks is illustrated in Fig. 7.
A. Cell level: UCL is the minimum indivisible CPS unit, and a single UAV can close a data-driven loop for resource allocation.The UAV interacts with a ground control station and potentially other UAVs when constructing a USL.
A. Cell level: At the cell level, a UAV integrates perceiving, computing, communicating, controlling, and extending functions.These capabilities support interaction with the GCS and other UAVs while forming higher-level networks.
A. Cell level: Packet dispatching illustrates how a single UAV uses prior information, onboard sensing, and path planning to address the logistics “last mile” problem.The UAV departs from a distribution station, initially plans a rough route, and collects real-time data during cruise.
UCL USL USoS: UCL operates as “hardware + software,” USL adds an inter-UAV communication network, and USoS extends integration across platforms and systems.Each hierarchy builds a progressively larger closed loop, from self-operation to multi-UAV coordination and cross-platform interoperability.

(a) Architecture … 2) Organization architecture:

The UAV cyber-physical system links physical entities with a cyber shell that senses, computes, controls, and communicates, while its organization ranges from simple ground-centered links to flexible multi-UAV topologies. At the system level, communication enables cooperation, shared information, distributed decisions, and larger-scale closed-loop operation, but may also become a bottleneck.

(a) Architecture: The UCL combines physical entities—operating environments and onboard sensors, actuators, and interacting devices—with a cyber shell providing sensing, computation, control, and communication.The physical entities manipulate the physical process, receive control commands, and exert effects through actuators; the cyber shell interfaces them with the information world.
1) Constitute architecture:: The cyber shell digitalizes physical entities and bridges them to the outside information world, binding the physical and cyber domains together.This connection enables the information world to control physical entities.
2) Organization architecture:: UCL organization is usually infrastructure-based, with the GCS as the center and each UAV connecting directly to and receiving intervention from the ground.Even with multiple UAVs, they generally do not communicate directly or use the GCS as a relay; the GCS controls each UAV separately.
B. System level: USL enables multiple UAVs to coordinate across spatial, temporal, or functional dimensions through a communication network, building a larger-scale closed loop.Its capabilities include mission coordination, joint path planning, cooperative control, monitoring and diagnosing, and data interoperation.
1) Constitute architecture:: Communication in USL supports automatic wider-range data flow, shared belief maps, global decision making, swarm intelligence, interoperability, and deeper resource configuration.Each UAV can maintain a complete belief map of others through observations and interactions.
2) Organization architecture:: USL organization is more diverse and flexible than UCL, comprising infrastructure-based, star or multi-star, flat or hierarchical ad hoc, and other distributed configurations.Infrastructure-based designs may cause link blockage, higher latency, and large-bandwidth downlink requirements, whereas ad hoc arrangements support UAV-to-UAV communication and distributed decisions.
2) Organization architecture:: The three USL organization architectures can be selected by mission, UAV quantity, and UAV performance, and can transform into one another through self-adaptation.Self-adaption algorithms have been studied to improve communication-network performance by supporting transformations among the architectures.
2) Organization architecture:: USL coordinates shared computation resources, selects decision-making entities, assigns computation tasks, and provides real-time dynamic control, while communication can enhance or bottleneck the closed loop.Technical requirements span communication networking, computation coordination, and unified cyber-physical control; shared information helps UAVs make optimal decisions, but the network may constrain system performance.

1) Communication network: · 2) Computation coordination:

The UAV network’s communication design must support harsh, mobile, power-constrained aerial links, while computation coordination assigns tasks across distributed entities according to task requirements and topology. Centralized, decentralized, and consensus-based decision mechanisms address latency, resilience, and coordination needs.

1) Communication network:: Communication networking can use IEEE 802.11, IEEE 802.15.4, 3G/LTE, and infrared, but a broadly applicable solution for harsh aerial links remains open.The solution must account for UAV mobility and power limitations.
1) Communication network:: Communication-network design spans physical, data link, network, transport, and cross-layer concerns.Examples include channel modeling, antenna and spectrum selection, MAC and channel allocation, routing and QoS, congestion control, and flow control.
2) Computation coordination:: Computation resources are distributed across the GCS and UAVs, requiring tasks to reach appropriate decision-making entities at suitable times.Where and how these decisions should be made remains an open problem involving multiple issues.
2) Computation coordination:: Task requirements determine placement: onboard controllers handle pitch, roll, and yaw, whereas computation-intensive image recognition is transferred to the GCS.Moderate tasks such as path planning and mission-related processing are also part of the coordination problem, although the passage is truncated.
2) Computation coordination:: Under infrastructure-based topology, the GCS generally makes centralized decisions using global system information.The cited work optimizes path planning and task allocation through centralized strategies, including a genetic algorithm and Voronoi diagram for concurrent assignment and trajectory planning.
2) Computation coordination:: Star and ad hoc topologies increase UAV participation because multi-hop UAV-GCS links can add latency and GCS failure should not paralyze the system.Distributed consensus can still support UAV-level decisions after receiving global or cluster information, including distributed path consensus for multi-robot coordination.
2) Computation coordination:: Cooperative control coordinates UAV positions and velocities while prioritizing collision avoidance with other UAVs and obstacles.The passage frames safe flight as a prerequisite for complicated cooperative missions.

3) Cooperative control: · C. System of system level

At the system-of-system level, UAV networks coordinate heterogeneous UAV resources and services through cloud platforms that enable data convergence, interoperability, and operational optimization. Cooperative control also encompasses multi-UAV collision avoidance and collective motion for transportation and complex group tasks.

3) Cooperative control:: UAV collision-avoidance research uses either centrally scheduled predefined routes or real-time planning for emergency cases and unknown environments.Predefined scheduling considers obstacles, restricted airspace, and altitude limitations, whereas real-time planning addresses emergency situations.
3) Cooperative control:: Collective motion coordinates UAVs as cohesive groups for transportation, light shows, collective mapping, and searching.The surveyed literature includes classifications and characterizations of formation types and architectures for collective movement.
3) Cooperative control:: USL improves robustness, flexibility, and activity range over UCL through UAV redundancy and loose single-hop connections with the GCS.USL applications include network coverage and line relays linking remote areas.
C. System of system level: A UAV cloud service platform integrates and schedules diverse USLs or UCLs, enabling heterogeneous data integration, exchange, sharing, and cross-system interoperability.The platform supports a closed loop involving comprehensive perception, analysis, and decision making.
C. System of system level: In USoS, each UAV accesses a uniform cloud interface that manages its resources and services while accepting sensed information, computing resources, and actuation abilities.The platform maintains UAV information about available resources and services.
C. System of system level: The USoS architecture includes three abstraction layers and supports cloud services, UAV performance management, operational optimization, preventive maintenance, energy efficiency, and manufacturing or upcycle instructions.These functions span internal performance management and external operational services.
C. System of system level: The platform provides UAV capabilities as external services, including sensing, actuation, camera capturing, and video recording.UAVs can also obtain firmware upgrades, remote maintenance, and service subscriptions from the platform.
C. System of system level: USoS functions as a UAV ecosystem in which heterogeneous UAVs are orchestrated and their data and services converge.Its architecture aims to optimize internal UAV operation performance while providing services externally.

1) Constitute architecture: · 2) Organization architecture: · 1) Flexible distributed computation:

The UAV network’s system-of-systems architecture combines UAV resources, heterogeneous interfaces, cloud services, and distributed data processing. Its flexible computation ecosystem assigns processing across edge, fog, and cloud resources according to mission requirements.

1) Constitute architecture:: The UAV layer supplies user-facing resources as a service and supports hardware interaction through middleware and data links.The service layer realizes cloud services through communication interfaces, data manipulation, storage, data fusion, distributed computing, and big data analytics.
1) Constitute architecture:: The service layer uses network interfaces for continuous streams and web services for control commands, cloud communication, and data exchange.UAV-originated streams are partly stored in the cloud and partly distributed among individual entities for analytics and distributed computing.
1) Constitute architecture:: USoS connects networked UAVs to a cloud service platform, whose organization depends on whether it is deployed over a LAN or the Internet.The two representative architectures are LAN-based and Internet-based platforms.
2) Organization architecture:: Heterogeneous network interfaces converge UAV networks with the cloud platform, using specialized UAVs in LAN deployments or existing cellular, satellite, and Wi-Fi infrastructures.These interfaces are indispensable in both organization scenarios.
2) Organization architecture:: USoS processes richer multidimensional data to extract information and knowledge for improved decision making and resource optimization.Data services can improve the platform’s and UAVs’ abilities to control and optimize resources.
1) Flexible distributed computation:: Edge, fog, and cloud computing can flexibly process data according to mission requirements and provide data, customized, and specialized intelligent services.These technologies are used to build the USoS cloud service platform.
1) Flexible distributed computation:: Cloud computing shifts computation-intensive tasks from network-edge devices to a virtually resource-rich platform for storage, advanced analysis, and big data analytics.Cloud platforms provide infrastructure, platform, and software services and can communicate with drones over the Internet.
1) Flexible distributed computation:: Edge computing processes data near UAVs and objectives, reducing data flow, bandwidth utilization, and latency while supporting agile connections, real-time analysis, and security.The three-layer ecosystem combines cloud computation, fog resources, and UAV sensing and actuation; shifting tasks toward the edge is expected to decrease network latency and load.

2) Resource and service provision: … 1) Communication contributes to computation:

The UAV network is presented as a service-oriented cloud platform whose resources and capabilities support applications across UAV hierarchies. Its communication and computation components are tightly coupled through data exchange, shared decisions, and reciprocal performance enhancement.

2) Resource and service provision:: UAV IaaS combines onboard sensors, actuators, payloads, memory, processors, and other internal resources with external resource or service providers.External components may include ground sensors, cloud-connected objects, and cloud computing infrastructures.
2) Resource and service provision:: The PaaS middleware integrates UAVs with the cloud and provides an application-building layer.The cloud service framework is illustrated in Fig. 17.
2) Resource and service provision:: UAV and collaborative services can be integrated into applications, reducing the time and cost of developing collaborative UAV applications.UAV services use UAV capabilities, whereas collaborative services support collaborative UAV operations.
2) Resource and service provision:: UAV services are selected according to available UAV resources and mission tasks, including sensing services that acquire data or trigger thresholds.Sensing services convey sensor data to a broker service and may support either value acquisition or threshold-based triggering.
2) Resource and service provision:: The UAV cloud’s ubiquitous platform enables networks at any hierarchy to access services anywhere, anytime, and through any method.The text describes military USoS deployments, civilian Internet-based platforms such as Dronemap Planner, and applications spanning agriculture, transportation, meteorology, and geography.
V. COUPLING EFFECTS IN THE UAV NETWORKS: The UAV network’s dataflow-based closed loop tightly couples cyber-domain and physical-domain components.Sensors introduce data from the physical world into the cyber domain, while actuators feed decisions back to affect the physical world.
A. Computation and communication: Communication performance can constrain real-time, high-throughput computation, whereas higher throughput and lower latency improve computation ability and cooperative intelligence.Communication also supports consensus beliefs that improve the tractability of optimal policy making in multi-agent POMDPs.
1) Communication contributes to computation:: Communication contributes input data for computing and decision-making, while computed decisions are shared through the network to reach consensus.Examples include state dissemination for collision avoidance, path planning, and task allocation.

2) Computation boosts communication: · B. Computation and control

Computation strengthens UAV-network communication by measuring channels, adapting waveforms, applying intelligent algorithms, and reducing communication overhead. It also enhances flight control through real-time planning, faster disturbance response, learning-based methods, and improved processor utilization.

2) Computation boosts communication:: Computation measures and models UAV communication channels, enabling waveform decisions based on fresh measurements and improving communication in unknown environments.A waveform library may be extended with optimal decisions for occasional and unknown environments; repeated learning iterations improve communication.
2) Computation boosts communication:: Communication and computation promote each other: computation unlocks communication capabilities, while communication enables information exchange for computational decisions.The section describes this relationship as positive feedback between computation and communication.
2) Computation boosts communication:: Intelligent algorithms enhance communication and networking across network layers to address UAV requirements such as low latency and high transmission rates.These requirements arise from frequent topology and status changes and high-resolution image or video transmission.
2) Computation boosts communication:: Multi-agent decision technologies reduce communication quantity and overhead by making bounded-communication decisions without sacrificing task performance.Communication may be limited, unavailable, or dangerous, so multi-agent teams must use it effectively.
B. Computation and control: Onboard computation directly supports UAV control because computed digital control inputs are translated into continuous signals and fed to actuators.Control decision making is usually undertaken by the UAV’s onboard computer.
B. Computation and control: Greater computation power facilitates real-time optimal path planning, formation control, faster responses to disturbances, and reduced dependence on system models.Learning algorithms can help reduce nonlinear control-design difficulties and support suitable control actions.
B. Computation and control: Intelligent flight control combines learning with conventional control algorithms, requiring flight data for training rather than prior mathematical model knowledge.These methods address deficiencies of classic controllers and open novel control theories.

1) Computation-enhanced flight control: … 3) The promotions between communication and control:

The survey examines computation-enhanced flight and formation control, then explains how communication and control constrain, depend on, and promote one another in UAV networks. These couplings motivate intelligent control, communication-aware coordination, and mobility-based networking strategies.

1) Computation-enhanced flight control:: Fuzzy control offers robustness, adaptability, fault tolerance, and easy realization without requiring an accurate process model.Its limitation is noted but truncated in the supplied passage.
1) Computation-enhanced flight control:: Reinforcement learning improves outdoor altitude control of multi-agent quadrotors when handling nonlinear disturbances compared with classical linear control techniques.The passage reports a significant improvement but gives no numerical value.
2) Computation-enhanced formation control:: Bio-inspired and artificial-intelligence algorithms enhance formation control for large-scale UAV swarms by supporting cooperative flight and formation optimization.PSO is highlighted as reflecting the three-dimensional flying behavior of UAV groups.
2) Computation-enhanced formation control:: Neural networks can learn UAV dynamics, including aerodynamic friction, enabling cooperative controllers to maintain desired formations under unmodeled dynamics and bounded unknown disturbances.The supplied passage also mentions neural-network-based adaptive consensus control, but the sentence is truncated.
C. Communication and control: Communication and control are tightly coupled: flight control drives three-dimensional mobility, while control can address networking problems through physical-domain changes such as location adjustments.The interaction extends beyond communication-parameter tuning.
1) The dependency between communication and control:: Flight and formation control depend on telemetry and emergency commands, while communication relies on topology control and distance maintenance among UAVs.Longer inter-UAV distances can require more transmission power or hops, which shortening distances can alleviate when mission constraints permit.
2) The constraints between communication and control:: Communication-constrained control accounts for delays, limited bandwidth, and limited range, while control-constrained communication addresses mobility-induced networking challenges.Communication delays can reduce flight-control performance and increase collision risk, making collision avoidance and path planning communication-aware.
3) The promotions between communication and control:: Mobility control promotes communication by changing UAV locations, using store-carry-forward operation and relays to extend connectivity beyond waveform and protocol adjustments.The survey identifies this geographic flexibility as evidence of cyber-physical integration; returning UAVs can also increase throughput by carrying data.

VI. CHALLENGES AND OPEN ISSUES … C. Computation offloading

The paper identifies open issues across UAV-network CPS modeling, heterogeneous-network convergence, and computation offloading. Key challenges involve coupling cyber and physical processes, accommodating dynamic interactions and network diversity, and managing resource-intensive workloads under UAV resource constraints.

VI. CHALLENGES AND OPEN ISSUES: UAV networks remain challenging to design because their CPS integration spans multidisciplinary techniques and raises open issues in both networking and CPS.
A. CPS model of the UAV networks: CPS modeling must couple discrete, asynchronous computations with continuous, synchronous physical processes across multiple interacting disciplines.
A. CPS model of the UAV networks: Dynamic UAV behaviors exhibit spatio-temporality and non-determinism, while autonomous flight processes interact continuously under strict spatial-temporal constraints.
A. CPS model of the UAV networks: Existing CPS models often ignore time or treat it as non-functional, while best-effort communication services conflict with CPS characteristics.
B. Convergence of heterogeneous networks: Heterogeneous-network convergence is significant because UAVs integrate diverse networks while differing from traditional networks in dynamicity, heterogeneity, and embeddedness.
B. Convergence of heterogeneous networks: Hybrid-network research still faces challenges including node access, channel switching, seamless service handover, network security, and quality of service.
C. Computation offloading: Computationally intensive tasks such as pattern recognition, natural language processing, and video preprocessing consume substantial resources and energy beyond UAVs’ limited computation and battery capacities.
C. Computation offloading: Game-theoretic offloading can trade execution time against energy consumption, making offloading valuable for resource-constrained UAVs and worthy of further development for complex missions.

D. Resource scheduling … VII. CONCLUSION

The survey identifies resource scheduling, energy efficiency, security and privacy, performance evaluation, and social impacts as major challenges in UAV networks viewed as cyber physical systems. It concludes that a CPS perspective organizes the field across cyber components and network hierarchies while motivating further cross-disciplinary research.

D. Resource scheduling: Resource scheduling must coordinate heterogeneous sensing, communication, computation, and actuation resources to satisfy applications and improve UAV-network performance, but resource modeling and management remain immature.The resources are varied, numerous, and strongly heterogeneous, while physical evolution and sensor observations are uncertain.
E. Energy efficiency: Energy efficiency is constrained by limited UAV power because communication and motion consume most energy, requiring separate and joint optimization of communication, flight, and whole-system lifetimes.The two lifetimes should be orchestrated to achieve comparable durations because failure of either can prevent mission fulfillment.
E. Energy efficiency: Computation can improve energy efficiency by deciding transmissions, selecting power, and planning lower-energy trajectories while preserving QoS and mission fulfillment.The passage cites MDP/POMDP decisions about whether, what, and whom to communicate as one example.
F. Security and privacy: UAV networks are vulnerable across physical and cyber layers because sensing, communication, computation, and control form an interdependent loop, while open-air wireless operation lacks mature security standards.Privacy risks include exposure of geographical and personally identifiable data that could enable profiling or physical attacks through storage, communication, or actuation modules.
G. Performance evaluation: Performance evaluation remains difficult because CPS-specific theoretical foundations, guidelines, and universally accepted verification standards are lacking despite component-level indicators and testing platforms.Existing modeling, simulation, verification, human-computer interaction, and complex-system platforms can be reused; MacroLab is one cited example.
H. Unwelcomed social impacts: UAV deployment creates societal risks involving privacy, safety, and psychological disturbance, requiring national and international regulation to uphold basic human privacy rights.Surveillance can threaten freedom and human rights when citizens are constantly monitored.
H. Unwelcomed social impacts: Temporary urban wireless-access and surveillance missions can endanger the public through crashes, environmental hazards, system errors, and shared airspace with manned aircraft, while persistent surveillance causes psychological pressure.The passage attributes crashes to power depletion, air turbulence, lightning, and unforeseen system errors, and describes profound effects in surveilled or war-torn areas.
VII. CONCLUSION: The survey structures UAV-network research around communication, computation, control, and three CPS hierarchies, offering beginners a tutorial and researchers a CPS perspective on complex interdisciplinary challenges.The three hierarchies are the cell level, system level, and system of system level; the field is described as promising, ongoing, and emerging.