Unmanned Aerial Vehicles: A Survey on Civil Applications and Key Research Challenges

In plain words

Drones, or unmanned aerial vehicles (UAVs), are being used for many jobs that ordinary people and companies do: watching traffic from above, giving internet coverage to places with no phone towers, searching for missing people after a disaster, checking fields and bridges, and carrying parcels. This paper is a broad review of those civil uses. It looks at what each use needs, what has already been tried, and what is still hard.

The hard parts it highlights are practical. A drone has a battery that runs out after a short flight, so charging and battery swapping matter a lot. Drones must avoid hitting things, and a group of them must avoid hitting each other. They also have to talk to each other and to the ground in a reliable and secure way, and that brings its own networking and security problems. The paper also gives a rough picture of how big the drone market is expected to become, and it names the rules and laws that still need to be settled.

The paper contains no new method and no experiments of its own. Its use is as a map of the field and a list of open questions.

TL;DR

A survey of civil UAV applications and their challenges, with no new method and no experiments §I. It covers eight application areas: search and rescue, remote sensing, construction and infrastructure inspection, precision agriculture, goods delivery, road-traffic monitoring, surveillance, and wireless coverage §IV–§XI. It then reviews four cross-cutting challenges: charging, collision avoidance and swarming, networking, and security §XII. The authors' main claim is that the field lacked a survey that combines applications and challenges, and that the civil market will be led by infrastructure work §I §II. Most numbers in the paper are market forecasts taken from other reports, not findings of the authors.

Problem

The authors state that earlier UAV surveys focus on one vertical application at a time and do not discuss the challenges that cut across domains or practical ways to address them §I. Their own comparison tables place this survey against six earlier surveys across application areas and across new technology trends tab. 1 tab. 2.

The paper assumes the UAV is a civil vehicle that is flown for a service and that the main limits are battery life, flight safety, communication and regulation §I §XII §XIII. It notes that the legal framework for commercial use needs to be built before the services can spread §II §XIII.

Method

The survey has no algorithm of its own. Its method is how it is organised.

1. Market and classification (Part I). It reviews the market value of UAV uses §II fig. 3 fig. 4 and classifies UAVs by platform. A low altitude platform (LAP) is quasi-stationary below 10 km; a high altitude platform (HAP) sits above 10 km §III. The classification covers VTOL vehicles, aircraft and balloons for LAP, and aircraft, balloons and airships for HAP §III tab. 3.
2. Applications (Part II). For each of eight application areas it gives the domain overview, the challenges, and the research trends and future insights §IV–§XI. The overall reading structure is shown in fig. 1 and fig. 2.
3. Challenges (Part III). It treats battery and charging §XII, collision avoidance and swarming §XII, networking §XII and security §XII, then concludes §XIII.
4. Path-loss models (inside wireless coverage, §XI). It lists the air-to-ground path-loss models it reviews, grouped by environment, altitude and link. The equations are reproduced below with the paper's own labels.

Air-to-ground path loss, low altitude platform

The mean path loss is a probability-weighted mix of a line-of-sight (LOS) and a non-line-of-sight (NLOS) term, where the LOS probability depends on the elevation angle eq. 1 §XI. The paper says the LOS probability and the two mean losses are defined as follows eq. 2 eq. 3 eq. 4 §XI:

L(h,R) = P(LOS) \times L_{LOS} + P(NLOS) \times L_{NLOS} eq. 1

P(LOS) = \frac{1}{1 + \alpha \exp\left(-\beta \left[\frac{180}{\pi}\theta - \alpha\right]\right)} eq. 2

L_{LOS}(\mathrm{dB}) = 20\log\left(\frac{4\pi f_c d}{c}\right) + \zeta_{LOS} eq. 3

L_{NLOS}(\mathrm{dB}) = 20\log\left(\frac{4\pi f_c d}{c}\right) + \zeta_{NLOS} eq. 4

In these equations θ is the elevation angle, d the UAV-to-user distance, f_c the carrier frequency, c the speed of light, and α, β, ζ_LOS and ζ_NLOS are environment-dependent constants eq. 2 eq. 3 eq. 4 §XI. The paper notes that about 3% of receivers in the reviewed study were in deep fading and are not covered by the two groups the model uses §XI.

Air-to-ground path loss, UAV to indoor user

L = L_F + L_B + L_I = \left(w\log_{10} d_{out} + w\log_{10} f + g_1\right) + \left(g_2 + g_3(1 - \cos\theta_i)^2\right) + \left(g_4 d_{in}\right) eq. 5

Here L_F is free-space loss, L_B the building penetration loss and L_I the indoor loss. d_out is the UAV-to-user distance, θ_i the incidence angle and d_in the indoor distance. The reviewed model uses w = 20, g_1 = 32.4, g_2 = 14, g_3 = 15 and g_4 = 0.5 eq. 5 §XI. As the horizontal distance grows, free-space loss rises while penetration loss falls as the incidence angle shrinks, a trade-off the paper attributes to the reviewed study §XI.

Cellular base station to UAV

L(d,\theta) = L_{ter}(d) + \eta(\theta) + X_{uav}(\theta) = 10\alpha\log(d) + A(\theta - \theta_o)\exp\left(-\frac{\theta - \theta_o}{B}\right) + \eta_o + \mathcal{N}(0, a\theta + \sigma_o) eq. 6

L_ter(d) is the mean terrestrial loss at distance d from the base station, η(θ) the excess aerial loss, and X_uav(θ) a zero-mean Gaussian shadowing term whose spread depends on the depression angle θ eq. 6 §XI. A, B, θ_o and η_o are fitting parameters eq. 6 §XI.

Air-to-ground path loss, high altitude platform

L = P(LOS) \times L_{LOS} + P(NLOS) \times L_{NLOS} eq. 7

P(LOS) = a - \frac{a - b}{1 + \left(\frac{\theta - c}{d}\right)^{e}} eq. 8

L_{LOS}(\mathrm{dB}) = 20\log(d_{km}) + 20\log(f_{GHz}) + 92.4 + \zeta_{LOS} eq. 9

L_{NLOS}(\mathrm{dB}) = 20\log(d_{km}) + 20\log(f_{GHz}) + 92.4 + L_s + \zeta_{NLOS} eq. 10

The model covers 2 to 6 GHz in four built-up area types (suburban, urban, dense urban, urban high-rise) §XI. a to e are empirical parameters, d_km is in km, f_GHz is in GHz, L_s is a random shadowing term that depends on the elevation angle, and ζ_LOS and ζ_NLOS are environment-dependent losses eq. 8 eq. 9 eq. 10 §XI.

Surveillance: decentralised border patrol alg. 1

The paper reproduces one algorithm, which the authors attribute to a reviewed decentralised patrol study alg. 1 §X. In our own words:

# Decentralised border patrol, one UAV per perimeter segment  [alg. 1]
# Input:  agreement with a neighbour, own segment, perimeter endpoints
# Output: patrol direction on the own segment

def patrol_step(agreement, neighbour, segment):
    # Meet the neighbour at the shared border if they agreed  [alg. 1]
    if agreement_with(neighbour):
        border = shared_border_position(segment, neighbour)
        travel_with(neighbour, to=border)
        # Monitor the own segment in the set direction  [alg. 1]
        direction = patrol_direction(segment)
        # Turn around when the perimeter endpoint is reached  [alg. 1]
        if at_perimeter_endpoint(segment):
            direction = reverse(direction)
    return direction

Hyperparameters

The survey has no training or experimental settings. The table lists every numeric value the text states, plus the unknown ones that an implementer would need.

Results

The survey reports no experiments. Its numbers are market forecasts and figures from the works it reviews. All of them are author-reported, and the paper credits the source of each one.

Paper vs code

No official code. The paper has no code or project link in its text, and the arXiv abstract has none. A GitHub repository search for the title's keywords returned no results. The paper is a survey, so there is no implementation to compare with the text.

Pitfalls

• Market numbers are forecasts from other reports. The PwC, SDI, Business Intelligence and Bard Center figures are taken from those sources and are not checked by the authors §II. Several are for years that have already passed or are close, so treat them as historical.
• Two different "$45 billion" statements. The abstract refers to a market of more than 45 billion dollars for UAV usage. The body says the addressable market for UAV uses is over 127 billion and that civil infrastructure is expected to be 45 billion of it §II. Do not merge the two.
• No review protocol. The text does not state how the surveyed papers were chosen, which databases were searched, or which years were covered beyond "recent literature" §I. A reader cannot reproduce the coverage.
• Path-loss notation is inconsistent. Eq. 6 writes the shadowing as a normal with spread a·θ + σ_o, but the text describes an angle-dependent standard deviation σ_uav(θ). The symbols a and σ_o are not defined in the text eq. 6 §XI.
• Same symbol, two roles. In eq. 2, α is both the environment constant and the offset inside the exponent eq. 2. Check the original study before implementing it.
• Angle and unit conventions are not stated for the HAP model. Eq. 2 uses (180/π)·θ, so θ is in radians there. Eq. 8 uses θ directly, and the text does not say which unit applies eq. 8 §XI.
• Deep-fade receivers are excluded from the LAP model. The model is fitted to the first two groups, and the paper says the third group is only about 3% of receivers §XI.
• Battery figures lack a source. The 45–60 minute charging time is not attributed to any study in the text, while the one-minute swap time comes from the reviewed swapping systems §XII.
• Typos that change meaning. The security section writes "VFV" for network function virtualisation (NFV) and "LLT" for low-power lossy networks (LLN) §XII. A stray non-Latin character appears in the wireless-coverage text §XI. Check the original paper before quoting these parts.
• One figure is reprinted from another work. The general UAV-system architecture figure is marked as reprinted from a separate paper §XII. Its licence is not the authors' own, so it must not be reused under this paper's licence.
• Dated content. The survey is a 2018 snapshot. Regulation, market size and the list of current platforms have changed since then.

Lineage

• Builds on: the surveys it compares itself with in its introduction tables. These include hayat2016survey, gupta2016survey, motlagh2016low, mozaffari2018tutorial, khawaja2018survey and aja2018 tab. 1 tab. 2. None of them has a wiki card, so nothing from their content is used here.
• Compared against: the same six surveys, plus the introduction's review of FANET and public-safety communication work, which the paper cites but does not reproduce §I.
• Superseded by: not known from wiki content.