PaperlayerView exact revision on arXiv

Source-linked AI summary

Design Challenges of Multi-UAV Systems in Cyber-Physical Applications: A Comprehensive Survey, and Future Directions

Reza Shakeri, Mohammed Ali Al-Garadi, Ahmed Badawy, Amr Mohamed, Tamer Khattab, Abdulla Al-Ali, Khaled A. Harras, Mohsen Guizani

arXiv:1810.09729v1cs.ROcs.AIeess.SY

TL;DR

Multi-UAV systems could extend UAV-enabled CPS applications, but their design involves coverage, navigation, energy, connectivity, security, and scalability challenges. This survey synthesizes these challenges, reviews algorithms and testbeds, and maps them to CPS applications. It concludes with application-specific comparisons and future directions for scalable and decentralized multi-UAV systems.

  • Problem

    Multi-UAV CPS applications require coordinated coverage, tracking, navigation, networking, energy management, and secure operation beyond the capabilities of a single UAV.

  • Method

    The survey classifies multi-UAV design challenges, reviews algorithms and practical testbeds, and maps challenges to important CPS applications using qualitative and quantitative comparisons.

  • Results

    The survey reports coverage and tracking comparisons, including that PFA maintains the highest coverage ratio among the compared mobility-pattern algorithms, at the cost of more drones, distance, and time complexity.

  • Takeaways & Limitations

    Future multi-UAV CPS research should address scalable and decentralized architectures, wireless connectivity, topology control, safety, security, and energy-efficient operation.

Abstract

from arXiv · show

Unmanned Aerial Vehicles (UAVs) have recently rapidly grown to facilitate a wide range of innovative applications that can fundamentally change the way cyber-physical systems (CPSs) are designed. CPSs are a modern generation of systems with synergic cooperation between computational and physical potentials that can interact with humans through several new mechanisms. The main advantages of using UAVs in CPS application is their exceptional features, including their mobility, dynamism, effortless deployment, adaptive altitude, agility, adjustability, and effective appraisal of real-world functions anytime and anywhere. Furthermore, from the technology perspective, UAVs are predicted to be a vital element of the development of advanced CPSs. Therefore, in this survey, we aim to pinpoint the most fundamental and important design challenges of multi-UAV systems for CPS applications. We highlight key and versatile aspects that span the coverage and tracking of targets and infrastructure objects, energy-efficient navigation, and image analysis using machine learning for fine-grained CPS applications. Key prototypes and testbeds are also investigated to show how these practical technologies can facilitate CPS applications. We present and propose state-of-the-art algorithms to address design challenges with both quantitative and qualitative methods and map these challenges with important CPS applications to draw insightful conclusions on the challenges of each application. Finally, we summarize potential new directions and ideas that could shape future research in these areas.

I. INTRODUCTION

UAVs offer mobile, flexible sensing for diverse civilian, industrial, and military applications, while multi-UAV coordination addresses the coverage limits of individual drones in CPS monitoring.

  • UAV Systems: UAVs support civilian, industrial, environmental, agricultural, surveillance, disaster-relief, and military applications through mobility, adaptive altitude, flexible deployment, and adjustable usage.They can gather diverse sensor data, including temperature, humidity, geographic information, thermographic images, and visual images.
  • UAV Systems: Satellite remote sensing provides slow large-scale monitoring, whereas camera sensor networks require widespread deployment and maintenance and may not support fine-grained structural monitoring.These two technologies represent contrasting trade-offs between monitoring accuracy and intrusive deployment or maintenance.
  • UAV Systems: CPSs integrate computational and physical processes through computing and communication services that connect sensors, devices, users, applications, and surroundings.Their applications span numerous domains, and their development is linked to increasingly capable UAV systems.
  • UAV Systems: A single UAV generally cannot cover every point in a practically sized area, motivating coordinated multi-UAV systems for coverage, tracking, path planning, networking, and energy management.Multi-UAV coordination introduces trade-offs between reducing coverage time, limiting the number of drones, maintaining target coverage, and minimizing mechanical energy consumption.

B. Cyber-physical systems

The survey situates multi-UAV systems within CPS applications and examines architectures, design challenges, testbeds, and application-specific trade-offs. It contrasts centralized, distributed, and hybrid control while emphasizing scalability, connectivity, and reliable operation.

  • Cyber-physical systems: CPSs combine computational and physical potentials and require robust, reliable, predictable, synchronized, distributed, and responsive operation in applications such as healthcare, traffic control, and automotive safety.Reliability and predictability are identified as central to successful real-world CPS deployment.
  • Cyber-physical systems: The survey maps multi-UAV design challenges to transportation, infrastructure inspection, surveillance, goods delivery, wireless and cellular systems, and medical and healthcare applications.It also compares approaches and draws conclusions about application-specific challenges.
  • Cyber-physical systems: Centralized architectures provide global environmental knowledge and accurate navigation but require continuous communication with all UAVs, limiting scalability for large swarms and monitoring areas.The paper motivates trade-offs between centralized and distributed processing and control.
  • Cyber-physical systems: Hybrid architectures push some computation toward UAVs while retaining lightweight high-level central control, potentially improving scalability through distributed monitoring and autonomous navigation.The DbeG testbed is presented as an inexpensive, flexible, controllable multi-UAV CPS testbed with localization capabilities.
  • Cyber-physical systems: Decentralized architectures scale to large-area coverage but create challenges in maintaining connectivity and accurate coverage while preserving safety and robustness.The GRASP testbed demonstrates multi-robot control algorithms for autonomous navigation and control.

III. AREA AND TARGET COVERAGE

The section distinguishes camera-based target coverage from area coverage, then organizes area-coverage work around fixed deployment and mobile scanning objectives.

  • Target coverage: Coverage requires at least one camera to view a target with acceptable quality within its sensing range.For full target coverage, one camera must see the target fully from an angle; coverage maximization is generally NP-complete.
  • Area coverage: Area coverage aims to monitor an entire region regardless of the objects or targets inside it.This objective is typically practical in controlled or reasonably small, dense areas.
  • Area coverage: Area-coverage studies classify objectives into fixed camera deployment and area scanning.Fixed deployment minimizes cameras while maximizing event-detection probability, whereas area scanning uses mobile cameras to detect events over time.
  • Area scanning: Mobile UAV scanning can subdivide a work area into cells and use graph-based routes to sweep the region.Related approaches generate waypoints so individual UAVs maximize visited area within a bounded region and fixed time interval.

B. Target Coverage

Target coverage spans fixed, dimensional, and mobile targets, with algorithm choices trading coverage, computational cost, camera count, prediction, and adaptability. Results show distinct trade-offs across target counts, camera FoV, sensing range, and mobility models.

  • Fixed-target coverage: Fixed-target coverage seeks minimal camera placement for covering targets, or maximal target coverage with a fixed camera count.
  • Fixed-target coverage: Clustering-based k-camera and cluster-first heuristics approach near-optimal camera counts as coverage range, viewing angle, or target count increases.
  • Fixed-target coverage: Greedy and D-Smp achieve perfect coverage in tested scenarios, whereas SSKCAM and fuzzy coverage leave some targets uncovered.Fuzzy coverage has lower execution time, while SSKCAM scales better than D-Smp as target numbers increase.
  • Fixed-target coverage: Wider FoV values reduce and nearly converge the required camera count and algorithmic complexity, while Greedy and D-Smp provide higher coverage fractions than SSKCAM and fuzzy coverage.
  • Fixed-target coverage: Increasing camera sensing range reduces the required camera count across algorithms; fuzzy coverage and SSKCAM also reduce execution time, whereas D-Smp increases exponentially.
  • Dimensional targets: Dimensional targets require coverage beyond point or center-based models, including one-dimensional pipelines and two-dimensional infrastructure views.
  • Mobile target tracking: Mobile-target tracking must account for temporal movement, environmental visual challenges, and the difficulty of matching agile targets with mechanical UAVs.Tracking is affected by illumination variation, occlusion, viewpoint variation, background clutter, and UAV adaptability limits.
  • Mobile target tracking: Prediction-based PFA achieves the highest coverage ratio across mobility patterns, but requires more drones, traveled distance, and time complexity than alternatives.All algorithms maintain coverage above 60%; LIFA has the lowest time complexity and drone efficiency, while PFA and PIFA use prediction.

C. Lessons learned

UAV coverage in CPS applications must address area and target coverage while balancing altitude-dependent image quality, coverage extent, and target mobility. These challenges support applications requiring rapid, low-cost visual sensing of real-world areas and objects.

  • Coverage challenges: UAV-enabled CPS coverage is classified into area coverage, which covers an entire area, and target coverage, which focuses on objects or targets.
  • Coverage challenges: Higher-altitude imaging widens camera coverage but lowers image quality, whereas lower-altitude imaging narrows coverage while improving quality.
  • Coverage challenges: Tracking agile targets requires lightweight, flexible UAVs and practical prediction of target mobility because matching target speed can be difficult.
  • CPS applications: The survey maps UAV areas and target-coverage use cases, design challenges, and requirements across CPS applications.
  • CPS applications: Medical and healthcare systems are identified as a CPS application domain for UAV technologies.

IV. PATH/TRAJECTORY PLANNING

Multi-UAV missions require coordinated path planning to cover key areas and objects while conserving energy, reducing latency, avoiding obstacles, and preventing collisions. The design space includes spatial, temporal, and vertical planning choices.

  • Path-planning requirements: Multi-UAV missions need optimal path planning to assign which UAV moves where and when, while supporting swarm management and collision avoidance.The objective includes prolonging system lifetime and minimizing response time after failure events.
  • Path-planning requirements: Path planning should minimize mechanical navigation energy while keeping UAVs obstacle-free and preventing inter-UAV collisions.
  • Path-planning requirements: Re-optimization and graph-based path-planning techniques are reported, but the cited approaches assume obstacle-free paths and do not address UAV collisions.
  • Collision avoidance: Collision-avoidance planning is examined through proper path selection and degrees of freedom.
  • Planning dimensions: The planning categories include spatial horizontal, temporal, and spatial vertical path planning.

A. Collision avoidance

Collision avoidance in multi-UAV surveillance requires coordinated path or timing changes, sensing, and timely maneuver decisions to prevent conflicts with UAVs, targets, and obstacles.

  • A. Collision avoidance: Moving-target surveillance creates collision risk because sudden target movements can temporarily affect the behavior of the multi-UAV system.
  • A. Collision avoidance: Spatial path planning avoids intersecting flight paths but can delay target monitoring and increase power consumption.The example reroutes one drone along a longer path in the same 2D plane.
  • A. Collision avoidance: Temporal planning separates drone movements across different time instances, although some targets may remain uncovered.Time-division tracking gives one drone priority during the first time instance.
  • A. Collision avoidance: Collision-free path planning can be formulated as nonlinear optimization minimizing total energy or accelerating real-time autonomous navigation.Greedy methods such as selecting the shortest path first are alternatives when optimization complexity is high.
  • A. Collision avoidance: Collision avoidance combines environmental sensing and detection with a subsequent avoidance-and-maneuver step.Candidate sensing methods include ADS-B, radar, and infrared sensors.

B. Swarm Formation

Multi-UAV formation and energy planning must jointly support coordinated, collision-free missions while accounting for connectivity, movement costs, battery limits, and alternative energy sources.

  • B. Swarm Formation: Formation control seeks to maintain a specified formation along generalized trajectories while avoiding collisions, potentially requiring a global view of all UAVs.
  • B. Swarm Formation: Existing swarm methods include genetic, ant-colony, evolutionary-game particle-swarm, and geometric path-planning algorithms.
  • C. Energy Planning and UAV energy profiling: Energy profiling models aggregate consumption for vertical and horizontal movements during target detection or tracking.The models are intended to estimate energy use for efficient coverage.
  • C. Energy Planning and UAV energy profiling: Choosing a path by movement energy alone can conflict with wireless connectivity requirements between participating UAVs.Horizontal paths may be preferable for communication even when vertical paths require fewer extra movements.
  • C. Energy Planning and UAV energy profiling: Multi-UAV planning seeks simultaneous collision-free routes that minimize total mechanical energy while also accounting for communication functions.Experiments estimate energy consumed during horizontal, upward, and downward movement, as well as video streaming and sensor measurements.
  • C. Energy Planning and UAV energy profiling: Battery replacement remains a practical problem for wireless nodes, especially when batteries are deployed in dangerous environments.
  • C. Energy Planning and UAV energy profiling: Energy harvesting uses sources such as solar, vibration, and thermal effects, while solar panels are presented as more logical than RF harvesting for moving UAVs.Panel size, weight, stability, sunlight availability, and location geometry require careful consideration.
  • C. Energy Planning and UAV energy profiling: SWIPT divides time between information and energy transfer through a single antenna for systems with fixed sensor nodes.

D. Lessons learned

The survey frames multi-UAV CPS deployment around collision avoidance, swarm organization, energy planning, and application-specific path-planning requirements.

  • D. Lessons learned: Collision avoidance requires sensing environmental data, making timely decisions, and matching designs to UAV type, power, payload, sensing dimension, and target motion.
  • D. Lessons learned: Multi-UAV CPS systems need suitable formation organization and reformation mechanisms.
  • D. Lessons learned: High mobility creates dynamic, lightly and alternately connected UAV networks, while 3D formation control remains insufficiently investigated.
  • D. Lessons learned: UAVs have limited onboard energy for communication, mobility, control, processing, and payload, making energy planning application dependent.
  • D. Lessons learned: Future work includes wireless charging and computational-intelligence methods such as deep reinforcement learning for path and battery scheduling.
  • D. Lessons learned: Infrastructure inspection requires multi-UAV coordination, synchronization, low-delay collision avoidance, energy management, and motion-aware path planning.
  • D. Lessons learned: Surveillance planning must handle arbitrary target counts, uncertainty, accidental risk, and real-time response needs.
  • D. Lessons learned: Goods-delivery navigation must account for dynamic weight, payload size, fragile-item departure and landing, and dense-urban collision avoidance.

V. IMAGE ANALYSIS AND VISION-BASED TECHNIQUES

UAV image analysis supports detection, localization, tracking, and decision making in CPS monitoring, but multi-UAV mobility introduces dynamic-background, denoising, and image-stitching challenges.

  • V. IMAGE ANALYSIS AND VISION-BASED TECHNIQUES: Images are widely used for UAV object representation, while vision methods support target detection, localization, tracking, and decision making.
  • V. IMAGE ANALYSIS AND VISION-BASED TECHNIQUES: CPS monitoring requires flexible, autonomous multi-UAV systems because these environments are highly dynamic.
  • V. IMAGE ANALYSIS AND VISION-BASED TECHNIQUES: UAV mobility creates image-analysis issues involving dynamic backgrounds, motion artifacts, and stitching images from multiple UAVs.
  • V. IMAGE ANALYSIS AND VISION-BASED TECHNIQUES: The survey discusses multi-UAV visual monitoring of CPSs together with related lessons learned.

A. Multi-UAVs for visual monitoring of CPSs

Multi-UAV visual monitoring combines coordinated data collection, robust image analysis, and information visualization for CPS applications. The section emphasizes deep learning, real-time processing, multi-view sensing, and augmented reality as central design directions.

  • Visual monitoring process: Multi-UAV visual monitoring requires informative image and video collection, robust analysis algorithms, and suitable information visualization.The collection process should balance data quality, quantity, implementation cost, collaboration, and informative viewpoints.
  • Vision-based analysis: Deep learning, particularly deep convolutional neural networks, supports image recognition, segmentation, object detection, and localization.The survey cites applications including vehicle counting, traffic monitoring, and civil infrastructure inspection.
  • Vision-based analysis: Real-time object detection is important for infrastructure inspection, surveillance, and search and rescue using UAV imagery.Cloud-based approaches can move recognition to the cloud while retaining low-level detection and short-term navigation onboard.
  • Recent advances and future directions: Future directions include cloud-supported distributed processing, deep learning, multi-view image stitching, incomplete-view handling, and multi-sensor data fusion.These directions target cooperative multi-UAV image analysis for more effective CPS monitoring.
  • Path planning, context awareness, and flight control: Localization and mapping require sophisticated vision methods that can operate in real time on power-constrained UAV devices.The survey identifies deployment within constrained devices and real-time response as design challenges.
  • Site monitoring and safety inspection: Augmented reality can add informative and interactive content to UAV-captured imagery, but few studies have investigated UAV-AR integration.The survey identifies AR integration as a direction for future multi-UAV monitoring systems.

CPS

The CPS applications represented in the supplied study mappings span vision-based monitoring, inspection, navigation, search and rescue, and autonomous flight functions. These examples connect UAV image-analysis methods with operational CPS objectives.

  • Vision-based CPS applications: Vision-based UAV studies address object detection, recognition, tracking, localization, and translational-state estimation across CPS applications.The mapped entries include machine-learning models, visual localization, and target-tracking methods.
  • Infrastructure and inspection: Civil infrastructure applications include monitoring, surveying, safety inspection, 3D-model generation, and detection of vehicles or solar panels.The study mappings associate these objectives with UAV imagery and image-analysis methods.
  • Agriculture and general analysis: Agricultural and environmental examples include oil-palm-tree classification, agriculture-object recognition, and general image analysis.The mapped methods include statistical analysis, convolutional support vector approaches, and neural-network architectures using histograms and convolution.
  • Navigation and response: UAV vision methods are also mapped to path planning, search and rescue, construction monitoring, indoor navigation, and autonomous landing.The listed approaches include CNN-based analysis, 3D-model generation, recurrent neural networks, and unsupervised perception models.

B. Lessons learned

The survey identifies imagery, networking, connectivity, quality of service, spectrum use, security, redundancy, and failure detection as intertwined multi-UAV design challenges. Dynamic mobility and constrained resources make coordination and reliable communication difficult.

  • Vision-based techniques: UAV image analysis supports target detection, localization, and tracking, but accurate automatic processing requires advances in signal processing, machine learning, and computer vision.Visible-band and near-infrared cameras are identified as relevant UAV sensing instruments.
  • Networking and connectivity: Multi-UAV networks must maintain connectivity while moving, although temporary disconnection may be allowed when drones collect data outside coverage and later return.Connectivity requirements depend on mission and environment.
  • Networking and connectivity: Existing swarm formation approaches often maintain formations and avoid collisions without jointly considering network connectivity or distributed coverage.This limitation can constrain operation in metallic structures, urban environments, and areas with poor wireless connectivity.
  • Quality of service: QoS design must address application-specific bandwidth, video quality, throughput, delay, energy limits, and changing wireless topology.Adaptive video and image encoding can trade available throughput against video distortion to reduce end-to-end delivery delay.
  • Quality of service: Cross-layer optimization coordinates parameters across network layers to improve end-to-end performance under resource constraints such as delivery delay.The survey describes video-distortion minimization frameworks as examples.
  • Spectrum management: Cognitive radio is presented as a potential response to spectrum insufficiency by sensing unoccupied bands and dynamically adjusting operating parameters.The approach is discussed for deployments involving several multi-UAV systems.
  • Security and reliability: Multi-UAV design must address hacking, loss of control, privacy, safety, security, and the energy and hardware costs of redundant UAVs.Redundancy can improve reliability and coverage but reduces network efficiency through additional energy and hardware consumption.
  • Failure detection: Failure detection mechanisms can support safer operation, including boundary-based detection that triggers UAV landing in an indoor testbed.The survey also references unreliable failure detectors for networked UAV systems.

D. Lessons learned

Dynamic UAV connectivity, QoS, security, and application-specific reliability requirements remain central challenges for CPS deployments. The survey maps these issues across surveillance, delivery, wireless networking, and healthcare scenarios.

  • Network connectivity: Dynamic node positions and changing velocities can produce irregular connections and unstable links in multi-UAV networks.Connectivity challenges vary with mission type and environment.
  • Network connectivity and security: Multi-UAV systems must provide seamless sessions while managing delay tolerance, energy limitations, communication range, control, and coordination requirements.IoT integration can additionally enlarge UAV vulnerability surfaces.
  • Surveillance: Surveillance applications integrated with IoT face expanded vulnerability surfaces, especially in unattended environments.The survey identifies security issues as a concern for such deployments.
  • Goods delivery: Goods-delivery systems require secure, reliable connectivity for tracking and monitoring, while UAV deployment can increase data-delivery delay and IoT vulnerability.The survey presents these as simultaneous delivery-system design considerations.
  • Wireless networking: Flying-base-station applications require secure connectivity, protection against ghost control, and deployment trade-offs for optimal wireless coverage.Ghost control refers to unauthorized UAV control through spoofed control or navigation signals.
  • Medical and healthcare systems: Healthcare UAV systems require secure, reliable connectivity and real-time response, creating a trade-off between stronger security and system flexibility.The human-life priority makes connectivity requirements especially stringent in these use cases.
  • Flight control: Autonomous flight control depends on trustworthy flight-dynamics models and must be examined for practicality and operability in real flights.Flight control is described through kernel control, command generation, and flight scheduling, with PID and learning-based strategies among proposed approaches.

A. Proportional–Integral–Derivative (PID) Controller

The section contrasts conventional and learning-based UAV control, while identifying centralized multi-UAV control as scalable only with continuous connectivity and motivating distributed alternatives.

  • A. Proportional–Integral–Derivative (PID) Controller: PID controllers adjust UAV control parameters against a reference flight path but have limitations in optimality and robustness.
  • A. Proportional–Integral–Derivative (PID) Controller: Learning-based control adapts to environmental changes and unexpected or aggressive conditions by learning from real-world experience.
  • A. Proportional–Integral–Derivative (PID) Controller: Learning inputs can include images, light imaging detection and ranging data, or both, enabling mappings from sensory patterns to high-level control instructions.
  • A. Proportional–Integral–Derivative (PID) Controller: Centralized navigation and control requires continuous connectivity with a command and control server and limits swarm scalability.
  • A. Proportional–Integral–Derivative (PID) Controller: Future distributed architectures must maintain connectivity and coverage while supporting coordinated decisions, secure communication, energy efficiency, and timely monitoring.
  • A. Proportional–Integral–Derivative (PID) Controller: Coverage and tracking research increasingly considers unpredictable, directional, and three-dimensional targets, including dense crowds.

C. Path and trajectory planning

The section identifies path planning, networking, security, and machine learning as coupled challenges for practical multi-UAV CPS deployments. It emphasizes efficient trajectories, cross-layer processing, resilient operation, and lightweight onboard intelligence.

  • C. Path and trajectory planning: Future coverage and tracking methods should optimize UAV paths using fewer drones and less energy to extend flight times.
  • C. Path and trajectory planning: Cross-layer mobile-edge computing for video analysis is proposed to jointly optimize end-to-end delay and energy consumption in multi-UAV systems.
  • C. Path and trajectory planning: UAVs’ wireless operation, limited computation, and open-area deployment expose multi-UAV systems to malicious attacks and leave safety and security solutions incomplete.
  • C. Path and trajectory planning: Machine learning supports autonomous navigation, object detection and tracking, path planning, obstacle avoidance, aggressive maneuvers, and multi-UAV collaboration.
  • C. Path and trajectory planning: Nano-UAVs currently have limited onboard capability for vision-based autonomous navigation, motivating lightweight algorithms and stronger onboard computation.
  • C. Path and trajectory planning: The survey concludes that future work spans scalable decentralized architectures, connectivity, coverage of unpredictable targets, path planning, safety, security, and related challenges.
Loading 1810.09729v1…