Ad Hoc Networks 130 (2022) 102790

Contents lists available at ScienceDirect

Ad Hoc Networks
journal homepage: www.elsevier.com/locate/adhoc

Survey paper

A review of AI-enabled routing protocols for UAV networks: Trends,

challenges, and future outlook
Arnau Rovira-Sugranes a ,∗, Abolfazl Razi b , Fatemeh Afghah c , Jacob Chakareski d
a
School of Informatics, Computing and Cyber Systems, Northern Arizona University, Flagstaff, USA
b
School of Computing, Clemson University, Clemson, USA
c
Department of Electrical Engineering, Clemson University, Clemson, USA
d
College of Computing, New Jersey Institute of Technology, Newark, USA

ARTICLE INFO ABSTRACT

Keywords: Unmanned Aerial Vehicles (UAVs), as a recently emerging technology, enabled a new breed of unprecedented
UAV networks applications in different domains. This technology’s ongoing trend is departing from large remotely-controlled
Artificial Intelligence drones to networks of small autonomous drones to collectively complete intricate tasks time and cost-
Predictive networking
effectively. An important challenge is developing efficient sensing, communication, and control algorithms
Self-adaptive learning-based protocol
that can accommodate the requirements of highly dynamic UAV networks with heterogeneous mobility
levels. Recently, the use of Artificial Intelligence (AI) in learning-based networking has gained momentum to
harness the learning power of cognizant nodes to make more intelligent networking decisions by integrating
computational intelligence into UAV networks. An important example of this trend is developing learning-
powered routing protocols, where machine learning methods are used to model and predict topology evolution,
channel status, traffic mobility, and environmental factors for enhanced routing.
This paper reviews AI-enabled routing protocols designed primarily for aerial networks, including topology-
predictive and self-adaptive learning-based routing algorithms, with an emphasis on accommodating highly-
dynamic network topology. To this end, we justify the importance and adaptation of AI into UAV network
communications. We also address, with an AI emphasis, the closely related topics of mobility and networking
models for UAV networks, simulation tools and public datasets, and relations to UAV swarming, which serve
to choose the right algorithm for each scenario. We conclude by presenting future trends, and the remaining
challenges in AI-based UAV networking, for different aspects of routing, connectivity, topology control, security
and privacy, energy efficiency, and spectrum sharing.1

1. Introduction coordinated military attacks [13]. In addition to these commercialized

use cases, many new applications are under design and implementation
Unmanned Aerial Networks (UAVs) are an emerging technology in academia and industry. For instance, surveying and mapping [14],
that has opened its way into many fields and is expected to con- volcano monitoring [15], UAV control by the brain [16], early warning
tinue impacting the future of human life in the coming years. UAVs of severe weather [17], plant protection [18], airborne wind energy
have already been utilized in many applications to provide fast, low- harvesting systems [19], robotic herding of a flock of birds [20],
cost, on-demand, and precise monitoring and actuation services while Amazon Prime Air [21], and UPS drone delivery service [22] are only
minimizing human intervention and life-threatening risks. This covers a few examples of many projects in their infancy steps.
many applications including transportation [1], traffic control [2], Compared to piloted aircraft, satellite-based imaging, and ground-
surveillance [3], border patrolling [4], search and rescue [5], disaster
based sensing and actuation platforms, UAVs offer several advantages,
management [6], wireless network connectivity [7,8], smart agricul-
including a small size, low operational and maintenance cost, less
ture and forestry [9], and remote immersion via mobile virtual real-
human intervention requirements, less operational hazard, autonomous
ity [10–12]. Drones are also widely used in the military domain. For
control, more controlled imaging with adjustable zoom and angle of
instance, Low-Cost UAV Swarming Technology (LOCUST) is a project
view, and higher maneuverability levels [23]. Therefore, the UAV
by the US navy to utilize a swarm of autonomous drones to perform

∗ Corresponding author.
E-mail address: ArnauRovira@nau.edu (A. Rovira-Sugranes).
1
This material is based upon the work supported by the National Science Foundation under Grant No. 2008784.

https://doi.org/10.1016/j.adhoc.2022.102790
Received 12 August 2021; Received in revised form 13 December 2021; Accepted 22 January 2022
Available online 19 February 2022
1570-8705/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

their high speed, heterogeneous maneuverability levels, and obstacle-

sparse flight zones, compared to ground vehicle networks with more
predictable motion trajectories dictated by road patterns. Therefore,
communication, control, path-planning, and information acquisition
protocols that are primarily designed for ground platforms deem in-
efficient for UAV networks. For instance, connectivity of vehicular
networks along roads necessitates optimal positioning of nodes or
regulating their speed in a one-dimensional subspace that is not directly
applicable to UAV networks. Likewise, autonomous driving and colli-
sion avoidance for drones have different constraints and requirements,
compared to similar tasks in self-driving cars.
The goal of this survey paper is two-fold. First, we review UAV
networks’ features, including UAV technology, networking protocols,
and swarms, with a focus on characterizing the impact of mobility on
network topology, connectivity, and networking performance metrics.
In this respect, we highlight the importance of having an accurate
mobility prediction system for more efficient networking. Second, we
review routing protocols designed for UAV networks, emphasizing AI-
enabled routing protocols, which present better outcomes for high
mobility networks [25–27]. Furthermore, we addressed the closely re-
lated topics of mobility models for UAV networks, simulation tools and
public datasets, and relations to UAV swarming, which serve to choose
the right algorithm for each scenario, as an additional contribution of
this paper.

1.1. Motivation

Routing protocols developed for Vehicular Ad-Hoc Networks

(VANETs) and other ground networks are not well-positioned to ac-
commodate the requirements of UAV networks. As stated earlier,
drones enjoy a higher mobility degree of freedom, compared to ground
vehicles. This leads to a more vivid and fast-changing topology in
comparison to VANETs [28]. Also, the lower node density of Flying
Ad-hoc Networks (FANETs), compared to VANETs, raise connectivity
issues due to the drones’ limited communication ranges [29]. In terms
of channel modeling, fading, and diversity phenomena, FANETs benefit
from more accessible Line-of-Sight (LoS) links between UAVs and the
use of smart directional antennas for collective beamforming, and
similar techniques can offer higher gains [30,31]. From a different
point of view, it also highlights the necessity of developing more
accurate localization and tracking technology for aerial networks.
Furthermore, conventional routing protocols developed for VANETs
merely rely on the node’s prior information or current perception of the
network topology, and do not perform well in maintaining connectiv-
ity. They either impose a large overhead for constantly updating the
global network information or require a time-consuming route setup
phase. Also, UAV networks are structure-free and not consistent with
centralized routing protocols. Appropriate routing protocols for UAV
networks should have properties like low complexity, low overhead,
and preferably without the need for global knowledge and lengthy
route setup stages, as discussed in [32]. The new generation of self-
Fig. 1. The organization of this survey paper.
adaptive learning-based and topology predictive routing protocols learn
the state of the network by experiencing and predicting dominant
trends and constantly adapting to both minor or abrupt changes. This
market has experienced continued growth in the past decades, from approach leads to a higher packet delivery ratio and energy efficiency.
an estimated $19.3 billion in 2019 to a projected $45.8 billion market In this paradigm, decisions are made based on the anticipated net-
by 2025, which represents a Compound Annual Growth Rate (CAGR) work topology, and not solely based on the current state. These key
of 15.5% from 2019 to 2025 [24]. requirements promote AI-enabled routing protocols to achieve superior
Despite the following advantages in using drone technology, there performance [25–27]. Therefore, using AI methods to predict motion
still exist numerous challenges, and technical issues for implementing patterns of freely flying UAVs in a 3-dimensional space is an inte-
networking and control protocols for UAV-based infrastructures [23]. gral part of AI-based UAV protocols, while routing protocols for a
For instance, the limited payload of UAVs translates into constraints in 2-dimensional network of cars along the highway (as in VANETs) may
power consumption, communication range, and computational limits not necessarily need this computationally-expensive component. This
that in turn may cause difficulties for networking, robust control, infor- survey paper is devoted to highlighting recent developments in the AI-
mation acquisition and processing, autonomy, and task coordination. based routing protocols and analyzing their benefits and drawbacks
Another issue is the extreme dynamicity of UAV networks due to when used in realistic situations.

2
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Table 1
Most recent survey papers for routing protocols in Wireless Sensor Networks.
Year Survey Content included Drawback Application domain
SDN-based routing, monitoring, cellular, satellite,
Does not include routing protocols beyond SDN-
[33] security, placement, evaluation tools and future SDN-based UAV networks
based methods.
challenges.
2020
UAV classification, application, mobility models, The review of predictive methods is not complete.
[34] routing protocols classification, challenges and open It also excludes self-adaptive learning-based routing UAV networks
issues. protocols.
UAV communication networks issues, characteris-
[29] tics, design, applications, routing protocols, quality Does not consider learning-based routing protocols. UAV networks
of service, and future open research areas.
Network architecture and design, routing protocols
[35] including performance analysis and QoS metrics, Routing classification only includes a few methods. UAV networks
and opening issues.
UAV-UGV coordination, data gathering, monitor- Only considers routing methods for UAV-assisted
[36] ing, cellular communications, disaster management, networks, excluding routing protocols for UAV-to- UAV-assisted networks
computing and UAV-assisted routing. UAV communications.
Only considers routing protocols for UAV-aided ve-
Routing protocols for UAV-aided vehicular ad hoc
[37] hicular networks, excluding routing for UAV-to-UAV UAV-aided vehicle networks
networks with open research issues and challenges.
communications.
UAV design, architecture, routing protocols, open AI-enabled routing protocols and their impact in
2019 [38] UAV networks
issues and research challenges. dynamic networks are missing.
Architecture, mobility models, routing techniques
Self-adaptive learning-based methods are not con-
[39] and protocols with a comparative study. Future UAV networks
sidered.
challenges are also included.
UAV routing schemes, including objectives, chal-
It only briefly mentions adaptive routing protocols,
[40] lenges, routing metrics, characteristics, and perfor- UAV networks
missing most of the AI-enabled routing protocols.
mance measures, along with open issues.
Routing protocols comparison and open research Does not include AI-enabled routing protocols,
2018 [41] UAV networks
issues. among others.
Most of the routing protocols are suitable for
Single-layer and cross-layer routing, challenges and
[42] vehicular networks, but not defined for UAV Vehicular ad-hoc networks
open research directions.
2017 Networks.
UAV architectures, projects, characteristics, applica-
Does not consider AI-enabled routing protocols and
[43] tions and routing protocols, with emphasis in UAV UAV networks
future trends.
security challenges.
Position-based routing protocols with a detailed
Includes position-based routing protocols only,
[44] description and comparative study. Also, mobility UAV networks
which is just one type of UAV routing.
models and UAV applications are described.
UAV technology, UAV networking, UAV swarm for-
mation, mobility models, UAV routing protocols,
Our survey tools and public datasets to simulate real UAV NA UAV networks
network environments, future trends and remaining
issues for UAV networking.

The rest of this paper is organized as presented in Fig. 1. In Sec- Here, we closely discuss the most notable papers from the last
tion 2, we investigate recently published survey papers to highlight the three years only, since newer papers usually tend to improve previous
new content and additional aspects covered by our paper. Next, we re- reviews and cover the most recent developments. In addition to recent
view the UAV technologies used in military, industrial and commercial studies, we also consider two fairly older papers for their remarkably
applications in Section 3. In Section 4, we emphasize the role of AI unique content. One paper is [45], which provides a thorough review of
methods in improving the performance of UAV networking. Section 5 routing protocols in inter-vehicle communication systems. This paper
presents commonly used networking protocols, and UAV swarm forma- covers broadcast-based routing, multicast, and geocast-based routing,
tion methods are presented in Section 6. In Section 7, commonly-used as well as unicast-based routing protocols, which is perhaps the most
UAV mobility models, and their impact on key networking character- complete review for routing protocols developed for vehicular net-
istics, including connectivity, channel models, network topology, and works. Another review is a seminal paper by Gupta et al. [23] published
routing efficiency, is investigated. Fundamentals of conventional and in 2015, which offers a broad outlook and comprehensive review of
AI-enabled routing protocols, along with their stability under dynamic important issues in UAV networks. Also, it reviews the concept of
conditions, are provided in Section 8. Tools, public datasets, and re- routing in networks subject to severe delays and disruptions, which is
mote experimentation infrastructures for testing routing protocols are unique among the published papers.
reviewed in Section 9. Future trends and remaining issues are discussed A survey paper by Awang et al. in 2017 [42] provides a review
in Section 10, followed by concluding remarks in Section 11. of routing protocols for vehicular ad hoc networks describing existing
single-layer and cross-layer routing algorithms. It offers a fluent review
2. Related work of routing protocol for VANETs along with a clear description of
the advantages and disadvantages of each method. However, most of
There are a few recent review papers that survey routing protocols the methods mentioned in this paper are designed for VANETs, and
for ground and aerial networks. Other related papers that review UAV not suitable for FANETs with substantially different constraints and
networks survey a broader set of aspects. Table 1 summarizes key topics requirements.
covered in these surveys, along with key topics missing in each paper. A review of routing protocols and security challenges in UAV net-
To our knowledge, no paper provides a comprehensive and up-to-date works is provided in [43]. This paper reviews different routing proto-
review of AI-based routing protocols for aerial networks, which is our cols developed for dynamic networks. Nevertheless, it does not include
central focus. an important and emerging trend of AI-enabled routing protocols.

3
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Another paper [44] provides a comprehensive review of position-

based routing protocols for UAV networks. It nicely classifies routing
protocols with a detailed description of each category. Also, the routing
algorithms are compared based on various criteria and performance
metrics. However, only position-based routing protocols are mentioned,
excluding many other types of routing protocols developed for UAV
networks.
In 2018, a paper titled ‘‘Routing protocols for Unmanned Aerial
Vehicles" [41] compared the routing protocols from different stand-
points. This paper sheds light on which methods are suitable for UAV
Networks under different network conditions and application-oriented
requirements. However, the provided review is not comprehensive, and
many efficient routing protocols such as hierarchical, probabilistic and
AI-enabled methods are not discussed.
In 2019, a survey paper offered a complete review of UAV network
design, architectures, routing protocols, open issues, and research chal-
lenges [38]. Deterministic, stochastic, and social-network-based routing
protocols are discussed, along with a qualitative comparison of their
major features, characteristics, and performance. However, AI-enabled
routing protocols and their role in accommodating dynamic networks
are missing.
In [39], a comparative review of major existing routing protocols Fig. 2. Military, industrial and commercial drones.
developed for FANETs, along with a careful analysis of their perfor-
mance under different design constraints and planning strategies, is
provided. However, the important class of self-adaptive learning-based more advanced technology, compared to commercial products [47,48].
methods is not discussed. Similarly, [40] offers a comprehensive review For their proven performance and success in reducing casualties, the
of routing schemes in FANETs, including objectives, challenges, rout- military sector invests heavily in the research and development of
ing metrics, characteristics, performance measures, and open issues. military drones [49]. UAVs in the military can be categorized based
However, most of the AI-enabled routing protocols are not discussed. on their weight, range, speed, and specific capabilities. Based on their
The following are a few survey papers published in 2020. A re- weight, we can differentiate class I, class II and class III drones. Class
cently published paper [33] reviews Software-Defined Network (SDN) I refers to micro, mini, or small drones that weigh less than 150 kgs.
and Network Function Virtualization (NFV) for UAV-assisted monitor- Class II includes tactical drones between 150 and 600 kgs. Lastly, class
ing, cellular, and satellite communications systems. More specifically, III are strategic drones weighing more than 600 kgs [46].
this paper reviewed SDN-based routing. However, it only considers Drones in industrial settings are used in a broad range of applica-
Air-to-Ground scenarios. tions such as smart agriculture, forestry, mining, construction, weather
A review of mobility models and routing algorithms for FANETs is and climate control, power plants, structural monitoring (buildings,
provided in [29,34], with the inclusion of UAV communication net- dams, and bridges) and the energy sector (oil and gas refineries). In
works issues. Nonetheless, they exclude an important class of learning- addition to high definition cameras, industrial drones are typically
based routing algorithms. A comparative analysis of emerging routing equipped with different types of sensors, including but not limited to
protocols for UAV networks under different conditions is provided LiDAR, Global Positioning System (GPS), range finders for collision
in [35]. However, it includes only the position-based methods and ig- avoidance, Positive-Intrinsic-Negative (PIN) diodes for motion detec-
nores different implementations of proactive, reactive, and AI-enabled tion, and pressure gauges [50]. Environmental and climate sensors such
routing protocols. as temperature, humidity, air pollution, and chemical sensors are also
Two recent works [36,37] offer an exhaustive review of communica- embodied in industrial drones when necessary.
tion protocols, applications, and security issues of UAV-assisted ground More basic drones are also used by ordinary people for regular tasks,
and vehicular networks. However, their center of attention is UAV- ranging from hobby and entertainment to more complicated monitoring
assisted routing, which excludes many routing protocols developed for tasks. The applications of commercial drones are countless and include
the more general class of UAV-to-UAV communication in FANETs. In shipping and delivery, inspection, real state, security, insurance, life
summary, almost all of the previously published surveys do not pay habitat monitoring, border patrolling, structural monitoring, enter-
enough attention to the emerging class of AI-enabled routing protocols, tainment, sports monitoring, fire monitoring, flood prediction, smart
which can be considered the most appropriate class of routing protocols agriculture, forest monitoring, volcano monitoring, fishery, weather
for extremely dynamic UAV Networks. report, etc. Some deliverables from drones include 3D maps, orthomo-
saic, and actionable reports [51]. Commercial drones must meet the
3. UAV technology: military, industrial and commercial drones regulations set by the Federal Aviation Administration (FAA) for safe
operation. For instance, commercial drones should have a maximum
Like many other technologies, the use of drones initiated in military weight of 55 pounds and should operate at or below 400 ft above the
domain, and soon afterward, made its way into industrial and commer- ground when in uncontrolled (Class G) airspace. Otherwise, specific
cial application. One of the key motivating factors was using drones authorizations should be obtained for flying in controlled airspace
in risky environments, and inaccessible areas with harsh conditions to (Class B, C, D, and E) [52].
minimize human risk. Different types of UAV technology, including military, industrial,
Military-grade drones typically utilized advanced features (e.g., and commercial drones, are displayed in Fig. 2.
stealth), custom-built sensors, equipment, and weaponry, appropriate
for military and reconnaissance missions in hostile environments [46]. 3.1. AI features
For instance, hyperspectral and Light Detection and Ranging (LiDAR)
sensors, AI-based object recognition, quantum cryptography, and multi- UAV technology has evolved in recent years. Modern UAVs are
spectral targeting systems with infrared sensors are usually utilized in equipped with onboard computation boards powered by embedded

4
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

circuitry, Graphical/Tensor Processing Units (GPU/TPU), and Field-

Programmable Gate Array (FPGA) boards that allow running light-
weight deep learning (DL) algorithms for AI applications [53]. These AI
chips, along with different sensors, allow UAVs to realize some levels of
intelligence to improve performance in various applications, including
those presented in Fig. 3. An alternative way is using UAVs with high
payload capacity (e.g., xFold rigs Dragon X12 U11 drone with a payload
size of 100 lbs) that can carry onboard huge computation servers. The
motivation for embedding AI features into UAV technology comes from
the importance of realizing low latency and fast processing of data for
real-time applications. Therefore, using deep learning with AI-enabled
chips by the UAVs can offer superior performance than streaming the
raw data and running the applications on the ground-based processing
centers [54]. This approach also substantially reduces the communica-
tion cost and satisfies the low-power requirements of UAVs compared
to aggregating the information by the UAVs and running the appli-
cations on the cloud. Online processing also is desirable for real-time
applications. However, this is not a general recipe for all applications.
There exist some scenarios (e.g., low-cost single-hop communication for
a single drone with relatively huge computation load), where raw video
streaming and offloading the computation load to the cloud servers
is advantageous [55]. Regardless of the computation load distribution
between the drone and the ground servers, the benefits of AI algorithms
are achievable. Fig. 3. AI-enabled UAV applications.

AI hardware in modern UAVs consists of computing, storage, and

networking parts [56]. Computing has been developing rapidly in
recent years. However, storage and networking aspects still need more and interpret common patterns and anticipate future states and events
research to satisfy UAVs’ diverse requirements. Particularly, there is a when making decisions.
need for long-term storage and networking protocols for linking equip- The benefits of using AI in UAVs are countless. For example, AI-
ment to servers. Regarding the AI chip design, various technologies are based decision-making with real-time data allows continuous feedback
available and under development, including GPU, TPU, reconfigurable in inaccessible areas to keep functions alive [65]. Also, training nodes
Neural Processing Unit (NPU), neuromorphic chip architectures, and by experience, in most cases, would result in more accurate results than
analog memory-based technologies. Based on the application-specific taking actions blindly. Gathering information can facilitate resource
requirements and constraints, we can incorporate one or some of these management for energy optimization and trajectory design to avoid
designs. Lastly, we must acknowledge that producing AI hardware is obstacles. However, these benefits come at a higher computational
a complex process [57]. Consequently, many of the tech leaders, such cost. In contrast to sophisticated scenarios, the value added by the
as Apple, Google, Microsoft, Intel, IBM, Nvidia, etc., are competing to AI methods in more straightforward scenarios, especially with no or
design and build the most innovative AI technology, and we expect to limited training dataset, is negligible. Therefore, each scenario should
witness more breakthrough developments in the coming years [58]. be carefully investigated to analyze the benefits and drawbacks of using
AI methods.
3.2. Drone manufacturers In [66], applications of AI methods to UAV-enabled wireless net-
working are listed. This paper summarizes different learning approaches
including supervised and unsupervised learning, reinforcement learn-
Drone market is expected to grow to $63.6 billion by 2025, with
ing and federated learning. Some areas where researchers introduced
2.4 million global shipments by 2023, increasing at a 66.8% Compound
AI-based solutions include positioning and detection [67–69], chan-
Annual Growth Rate (CAGR) [59]. A large portion of the commercial
nel estimation [70,71], virtual reality applications [11,72,73], imag-
drone market share in the US (about 77%) belongs to the DJI company.
ing [74], autonomous path control [75,76], scheduling and resource
Intel, Vuneec, Parrot, GoPro, 3DR, HolyStone, Autel, SenseFly, and
allocation [77], security [78] and sensing [79], as shown in Fig. 3.
Kespry are among the top 10 drone market shares in the US [60]. Also,
Our paper focuses on the applications of AI in routing protocols while
many companies use drones to provide 3rd party aerial solutions for
mentioning how AI is embedded into UAV technology, networking
different applications. For instance, PrecisionHawk is a North Carolina-
protocols and swarms, in general.
based company that offers smart agriculture solutions [61]). Fortem
Fig. 4 presents a scenario on the importance of including AI tech-
Technologies offers AI-enabled interceptor drones that can hunt intrud- niques for optimal routing. Node 1 intends to send a packet to node 5
ing drones [62]. The number of these companies grows larger than 100, through the optimal path. The edge metrics represent an arbitrary per-
and a list of such companies can be found in [63]. Commercial drones formance metric such as distance, energy consumption, delay, bitrate,
can offer service in many domains including emergency response, or a combination of these metrics. Consider a network of freely mov-
disaster relief, disease control, fighting crime, etc. [64]. ing UAVs that transmit their info through queued and delay-tolerant
communication. The network topology can change substantially during
4. Artificial intelligence in UAV networks the transmission session while the data packets are waiting in the
intermediate nodes’ transmission buffer. Therefore, the optimal path, if
AI-based solutions help to solve complex problems related to UAV found by the source node based on the initial network topology using a
networking and operation by integrating computational intelligence typical shortest path algorithm, may not remain optimal throughout the
into different aspects of UAV networks. The key idea is to incorporate transmission. In Fig. 4, the blue and red circles, respectively, present the
AI algorithms into networking and control protocols to assist UAVs in original and the updated positions of the nodes (after motions shown
perceiving the networks’ and environments’ overall conditions based by dashed green arrows). A conventional algorithm would determine
on their limited observations. AI can also empower UAVs to process (1-2-3-5) as the optimal path from source node 1 to destination 5

5
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

(QoS) both from the technical and feasibility, and economic points of
view. Some efforts have been made to create a nationwide and high-
speed broadband wireless network for public safety communications.
For example, FirstNet offers a solution to deploy, operate, maintain, and
improve the first high-speed, nationwide wireless broadband network
dedicated to public safety that could apply to UAV networks in the
near future [83]. The potential broadband wireless technologies include
WiFi, 4G Long Term Evolution (LTE), 5G (with the 3rd Generation
Partnership Project (3GPP) standard on 5G communication for drones),
satellite communications, and dedicated public safety systems such
as TETRA and APCO25. Also, Long Range Wide Area Network (Lo-
RaWAN), which enables long-range communications at low powers,
when high throughput is not necessary [84–86].
The following is a review of the most commonly used protocols in
Fig. 4. Scenario showing the importance of predictive routing [80].
UAV networks, emphasizing their capability in handling the dynamicity
of the network topology.

5.1. Wi-Fi

Most commercial UAVs use Wi-Fi (IEEE 802.11 series) for their
communications, especially to a ground station (e.g., command and
control commands in the uplink and video streaming in the down-
link), as a low-cost, scalable, and affordable solution. Wi-Fi-based UAV
networks can also be used for wireless backhauling [87]. Also, inter-
drone communications can be powered by Wi-Fi provided that one
node is defined as the Access Point (AP) to implement a local WLAN.
This node may or may not provide access to the Internet. Apparently,
one drawback of Wi-Fi is handling mobility and hand-off between
the base stations, limiting the operation range of drones within direct
access to the AP to a few miles. Although the throughput of Wi-Fi
(theoretically between 54 Mbps for 802.11a to as high as 2.4 Gbps for
802.11ax [88]) is relatively lower than LTE and 5G, it is sufficient for
most applications, including real-time high-resolution video streaming.
Fig. 5. UAV Networking from different standpoints. In scenarios that long-range connectivity is required, Wi-Fi loses the
game to licensed wireless systems when such networks are available.
Some drones develop their proprietary communication protocol on
(represented by blue arrows) based on the original positions (blue top of Wi-Fi. For instance, the XFold spy x8 KDE U3 drone by Xfold
circles). In contrast, a predictive routing selects the path shown by red Rigs [89] comes with a Futaba Commercial 14-channel.
color (1-2-4-5), taking into account the predicted network topology
change (i.e., the position of the nodes when met by the data packets),
while the packet is waiting in the transmission buffer of node 2. 5.2. LTE

5. UAV networking protocols LTE systems offer airborne connectivity beyond the LoS communi-
cations for drones. They improve the throughput and network connec-
Different communication protocols can be used to transfer data tivity due to the hard and soft hand-over mechanism [90]. Recent years
among drones, to/from satellite and aerial control units through Air have witnessed a surge of activities in using terrestrial LTE networks to
to Air (A2 A) communications, and from Air to Ground (A2G) (Fig. 5). provide connectivity to UAVs. A collaborative project has been initiated
Ground-based stations include standalone control units, larger servers, by FAA and the National Aeronautics and Space Administration (NASA)
Internet gateways, and edge computers. As shown in Fig. 6, networking in the U.S. since January 2017 to build a system using LTE technology.
can be studied from different perspectives, especially from the commu- To better understand the potential of LTE for small UAVs, the 3GPP has
nication, computation, and scheduling requirements and constraints.
formed a study group to investigate enhanced LTE support for aerial
Most of the aerial monitoring platforms utilize ground-based or web-
vehicles since March 2017 [91]. The most notable drawback of using
based servers for bulk processing, where drones collect and offload
LTE and other cellular systems is the need to register drone transmitters
raw information into processing units. However, alternative methods
with a service provider, which increases the operation cost and restricts
such as on-the-fly processing using embedded light-weight GPUs/TPUs,
the operation of drones to areas covered by the service provider. This
Mobile Edge Computing (MEC), and fog computing for accelerated
and near real-time processing are gaining more momentum recent is why using cellular systems is not as popular as Wi-Fi for drones.
years [81,82]. Another key issue is that the LTE propagation plannings typically
In most cases, the operation area is vast, far beyond the Line of Sight aim to serve the ground users; hence, the propagation maps are not
(LoS) communication range of a single UAV, hence using networked optimized for aerial nodes. Therefore, LTE radio planning requires
UAV platforms is unavoidable to ensure connectivity. One of the key substantial revisions to serve UAV networks, especially when they
design questions is choosing the best wireless technology (e.g., WiFi, scale up to large networks at higher altitudes and with a high-varying
LTE bands) with enough capacity and an acceptable Quality of Service topology.

6
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Fig. 6. UAV networking: different aspects.

5.3. 5G protocols, most of which have not yet been commercialized. For in-
stance, the idea of spectrum sharing and spectrum leasing for drones is
Similar to LTE, 5G is also considered for drone communications proposed to extend the connectivity of drones for high-speed and tem-
when higher bitrates beyond 2.4 Gbps are required. It also enables the porary service when wireless coverage is accessible, which integrates
concept of Internet of Things (IoT) for drones, where a drone serves WiFi and Cellular access [101,102]. Also, beamforming can extend
as a thing [92]. UAV-assisted communications have several promising the communication range further and reduce the interference [103].
advantages such as facilitating on-demand deployment, high flexibility Indeed, UAVs provide a realistic scenario for distributed and coopera-
in network reconfiguration, and enabling long-range LoS communi- tive beamforming, since transmit antennas can be spatially separated
cation links. In some scenarios, drones serve as 5G radio stations among UAVs [104]. The authors of this paper have proposed AI-
(also called AP) to extend the coverage of 5G networks for ground enabled routing [105], compression [106], and task coordination [107]
users, especially for sensitive applications, such as public safety and protocols to minimize the unnecessary information exchange among
post-disaster management [93–95]. the UAVs and prolong their mission time.

5.4. 6G 6. UAV swarms

The demand for higher throughput and bigger numbers of devices The concept of using UAV swarms is gaining more momentum in
never stops, and 6G is on the way to serve these requirements. 6G, recent years. The idea is to use a sheer number of drones, in most cases
the next generation for wireless communication, is expected to provide miniaturized and limited-capability drones, to collectively perform a
intelligent, secure, reliable, and limitless connectivity at rates 100 times complicated mission with no or minimal operator intervention. This
faster than 5G [96]. Similar to 5G, we expect that the UAV networks’ approach mitigates the drawbacks of using a single drone, such as lim-
diverse requirements such as low latency, reliability, and energy effi- ited allowable payload and limited sensing and actuation capabilities. A
ciency will be better served by 6G networks, and aerial nodes will be an general architecture for task order in swarm environments is presented
integral part of 6G networks. Also, network intelligence is envisioned in [108].
to be a key feature of 6G, which can assist in many connectivity-
related applications [97], for example, in a blockchain-based solution 6.1. Swarming advantages
for UAV communications [98]. Some of the challenges brought by the
futuristic concept of connected sky include high mobility, interference, The main advantages of using UAV swarms include shortening the
and connection to down-tilted antennas. It is expected that aerial nodes, task completion time, extending the coverage area, and reducing the
when integrated into terrestrial nodes, will be instrumental in covering operation cost. In the military domain, UAV swarms also increase the
such issues and enhancing the 6G user experience. tactical mission’s success rate by eliminating the reliance of the mission
on a single drone’s functionality, which can be subject to cyber-attacks
5.5. AI benefits in UAV networking and hijacking by an adversary. Using UAV swarms can also increase
the unpredictability of the mission and overwhelm the enemy’s defense
Artificial intelligence for UAV networking can help with the relia- system with a large number of potential targets in an interactive
bility, connectivity, and security of wireless communication by offering battleground. Further, UAV swarms are used to collectively find and
data-driven solutions for key challenges of interference management, fight enemy targets [109]. These ideas were behind the US navy’s LO-
mobility management, and handover, cyber–physical attacks, and au- CUST project to design UAV swarm attacks [13]. Also, swarm systems
thentication [99]. For instance, [100] uses AI to predict transmission can be equipped with anti-jamming systems to more efficiently block
success and failures, to anticipate and avoid networking issues. cyber attacks [110]. UAV swarms which incorporate an authentication
Among the aforementioned communication protocols, the most ap- protocol using Physical Unclonable Functions (PUFs) can authenticate
propriate one should be selected based on the application-specific multiple devices at once with high scalability while being resistant to
constraints and requirements. Several research efforts have been de- mobility, and multiple cyberattacks such as physical capture, cloning
voted to implementing new networking algorithms on top of these attacks, eavesdropping, and man-in-middle attacks [111,112].

7
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Fig. 7. Different architectures for UAV swarms.

The use of UAV swarms is not limited to military applications. a central controller’s function [114]. This architecture takes advantage
They enable search and rescue missions over big areas [113]. Some of the high mobility and networking efficiency of M2M communication
agriculture duties such as watering or identifying sick plants are time- in LTE, and 5G wireless systems [122–125].
consuming, and using UAV swarms with minimal operator intervention
would increase the efficiency of precision agriculture [114]. More
applications for UAV swarms can be found in [115]. 6.3. Artificial swarm intelligence

6.2. Swarm architectures

Artificial swarm intelligence, also known as swarm AI, is a technol-
ogy that combines real-time inputs and uses algorithms to optimize the
UAV swarm refers to scenarios when a sheer number of drones, most
overall performance of the swarm [126]. The initial idea comes from
of which with similar sensing, networking, and actuation capabilities,
collectively perform a designated monitoring or operation tasks [116]. the swarm intelligence found in natural systems, including ant colonies,
For example, a swarm of drones can carry out a coordinated military bird flocking, animal herding, etc.
attack [13]. UAV swarms function in the following different ways from For UAV networks, it represents an aerial system that uses AI
the control and coordination point of view. and data-driven methods to control the drones to achieve the des-
ignated goal [127]. In military drones, AI can be used to enhance
6.2.1. Infrastructure-based swarm architecture UAV swarm operation by increasing the range, accuracy, mass, coor-
In this architecture, a central node, mostly a Ground Control Station dination, intelligence, and speed, with a potential impact on security
(GCS), collects and processes real-time information from all swarm and strategic stability [128]. Other examples include AI-based flight
members; and sends back control commands (e.g., navigation, sensor control for autonomous drones for real-time positioning without a
actuation, sampling rate, camera control, etc.) to the UAVs, as shown centralized controller [129], as well as flight control for detection,
in Fig. 7(a). For instance, pre-planned paths can be revised during the localization, and tracking tasks while relying only on local spatial,
mission to avoid collisions when facing unexpected obstacles based on
temporal, and electromagnetic information [130]. In the upcoming
the live video streaming by member UAVs [117]. The key advantages
years, we expect to see how UAV swarm AI can revolutionize many
of this architecture include (i) feasibility of mission for low-capability
existing systems and create a new breed of applications. A key concern
UAVs by offloading the computation load to a more capable GCS,
(ii) global optimality of the resulting decisions, and (iii) no need for is the negative impacts that may be brought by the excessive power
complex networking algorithms for inter-UAV communications [118]. of autonomous UAVs that can jeopardize people’s privacy and security.
Also, realizing asymmetric security protocols such as Public Private This issue is more critical when drones’ control units are hijacked by
Infrastructure (PKI), which requires Central Authority (CA) to allocate cyber attacks. Therefore, a hot research area is developing security and
digital signatures is feasible [119]. However, it suffers from the well- privacy-preserving protocols for next-generation UAV networks.
known drawbacks of central systems such as (i) sensitivity to the GSC
malfunction, hijack, or cyber threats, and (ii) the restriction of the
mission area to the accessible range of the GCS. 7. Impact of mobility on communications

6.2.2. Ad-hoc structure-free architecture In this section, we investigate the impact of node mobility on data
This architecture allows direct communication between UAVs with transmissions. We first review popular UAV mobility models as well as
no need for APs or routers, and can utilize distributed decision making the techniques used to predict UAV mobility and network topology. We
(Fig. 7(b)). This method eliminates the sensitivity of the mission to the also review the challenges that mobility brings to connectivity control
GCS function and relaxes the constraints on the mission area. The cost
and optimal routing.
paid is the need for more capable UAVs for local decision making and
a routing protocol to accommodate dynamic network topology [120,
121]. 7.1. UAV mobility models

6.2.3. Hybrid swarm architecture

This architecture makes use of a cellular network to connect UAVs Mobility models are used to describe, model, and emulate UAV
while using distributed decision-making without the need for a cen- motion trajectories. Generative modes typically incorporate location,
tral controller (Fig. 7(c)). This architecture leverages the strengths of speed, and direction changes as model parameters. The following is
network-based and structure-free architectures by enabling long-range the list of commonly-used mobility models that facilitate the analysis
missions with reliable networking among drones while not relying on of UAV networks.

8
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

7.1.1. Random WayPoint (RWP) [131,132] 7.1.7. Distributed pheromone repel model [143]
This segment-wise model includes linear and independent motions This model uses a pheromone map to guide UAVs in reconnaissance
with constant speed and direction between a set of points called way- scenarios. Each UAV maintains its own pheromone map, and scans the
points. Also, UAVs decide on their next action based on some fixed area of the corresponding map. The UAVs share information every few
probabilities, and their motion does not depend on neighbor nodes. seconds to create a global pheromone map. UAVs turn right, left, or go
According to [133], RWP has two important variants, Random Walk straight with probabilities based on the pheromone smell. UAVs prefer
Model (RWM) and Random Direction Model (RDM) [134]. This model areas with a low pheromone smell, so new areas are scanned. This
can be used for both rotary and fixed-wing drones, and is the most model results in good scanning properties, but does not consider the
appropriate for missions with pre-path planning.
network connectivity between UAVs that serve different areas.

7.1.2. Levy walk/flight [135]

This mobility model is similar to the random walk mobility model, 7.1.8. Hybrid Mobility model with Pheromones (H3MP) [144]
with a distinction that the steps-lengths are not constant, rather random This model is suitable for search and rescue applications. This
values that follow a power-law distribution. The truncated Levy walk hybrid mobility model combines Markov chains and pheromones to
model incorporates tendency of the node to stay in confined area with adapt to dynamic environments. Markov chains guide UAVs to promis-
some randomness and hence has been shown to be a reasonable model ing areas, and pheromones guide information sharing that allows
for human walk and cellular users based on statistical analysis, com- mobility management through UAVs. Results show the superiority
pared to pure random models (like random walk and Brownian motion) of this method in detecting and tracking targets, compared to other
and also random way points, as stated in [136,137]. We conjecture pheromone-based methods.
that the rotary drones in some missions may have mobility patterns
to human and follow similar models. It is recognized by former works
and some believe that Levy flight model is able to mimic movement 7.1.9. UAV fleet mobility model [7]
patterns of aerial nodes over a larger time span where mixed effects This mobility model incorporates the remaining energy level, the
may be experienced [138]. area coverage, and network connectivity into the mobility decision
criterion. After receiving information from its neighbors, each UAV
7.1.3. Gauss Markov Mobility Model (GMMM) [139] determines its next movement based on these criteria. The direction
This model is used to simulate the movement of UAVs in swarms, and the speed of the UAVs are calculated using weighted vectors
which incorporates controlled randomness to the speed and direction considering neighbor UAVs. Results show that this method outperforms
equations. It prevents sudden stops and sharp turns within the simula- random motion methods in terms of coverage and connectivity.
tion region, to realize smooth and more realistic trajectories, especially
These models clearly are more appropriate to design motion paths
for fixed-wing UAVs, and targeted missions.
for specific missions, and less appropriate for modeling general UAVs
networks. Overall, a proper mobility model should be adopted for each
7.1.4. Semi random circular movement [133,140]
application based on the drone types, the mission requirements, and
Used to simulate the curved movements of UAVs when they hover
the utilized path planning method.
at a constant altitude (e.g., for collecting imagery). It has a uniform
node spatial distribution and outperforms random waypoints (in terms
of connectivity and scanning coverage over 2D disk). The nodes move 7.2. Mobility prediction
around co-centered circles to cover destinations located on the circle
perimeters. Uniform distributions are used for directional and angular
Network nodes use radar-based and visual target tracking to per-
velocities and the node pause at each destination. Once a circle scan is
ceive the network topology, at least in their close neighborhood. The
finished, the node switches to the next circle.
purpose of mobility prediction is to go one step beyond and antici-
pate the future locations of the objects that form the network. These
7.1.5. Mission Plan-Based mobility model (MPB) [141]
In this mobility model, UAVs are aware of trajectory information methods can be categorized into the following two mainstream trends,
and move according to a predetermined path to the mission area, where data-driven, and model-based methods.
potential information is available. Start and end points are randomly
assigned, but the velocity and the flight time are given. 7.2.1. Data-driven
This approach includes data mining and fuzzy methods, where
7.1.6. PaPaRaZzi Mobility model (PPRZM) [142]
frequent motion patterns are exploited by analyzing relatively large
It is a stochastic mobility model that combines various models with
datasets. These methods indirectly capture the influence of the natural
a Markov state diagram. Each UAV chooses a movement type from
and human-made textures, user behavioral habits on the spatial and
a set of predefined motion patterns with different parameters. Each
temporal variations of node mobility. For instance, TAPASCologne is a
motion state’s parameters are initialized randomly according to a given
project to collect and publish datasets of vehicle motion patterns in
distribution and remain unchanged until the transition to another state
the city of Cologne, in Germany with application to cellular network
with different motion parameters. A set of common motion patterns
includes ‘‘Stay-at’’ ‘‘Way-Point’’, ‘‘Eight’’, ‘‘Scan’’, and ‘‘Oval’’, which design [145,146]. Similar data-driven methods are proposed to model
are used in the original implementation of PPRZM. Results show that the motion patterns of pedestrians [147–149], vehicles [150,151],
this model outperforms the RWP model in terms of geometric and net- animals [152] and other mobile users. In a similar line of research, node
work performance, since it brings the flexibility of switching between mobilities are not exploited directly, rather traffic distribution trends
different modes. are extracted [153,154]. Although these methods are useful in the
Most of the aforementioned models consider independent motion network planning phase, their primary goal is to characterize general
trajectories for each drone. However, in some scenarios (e.g., UAV properties of the motion of a population of objects in different settings.
swarms), the motion of network members can be highly correlated. Sev- Therefore, they are not well suited for predicting the position of single
eral models are proposed to address this concept and realize correlated objects, which is influenced by the real-time node decisions. For such
trajectories. scenarios, model-based prediction methods are more appropriate.

9
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

7.2.2. Model-based topology is known for a central node, which can share with network
In this approach, the smoothness of the motion paths is used to members. Therefore, the source node can execute a local algorithm to
predict the future locations of mobile objects, typically in an online find the optimal end-to-end path. In distributed algorithms, the nodes
fashion. These methods include piecewise segment methods [155], are not aware of the entire network topology, and they have partial
Hidden Markov Models (HMM) [149], Levy flight process [136–138], knowledge about their local subnetwork topology. In these methods,
Bayesian methods [156,157], Manifold learning [158], Kalman filter- either sequential methods are used to break down finding the optimal
ing [159,160], fuzzy zone-based method [156] and mixture Gaussian end-to-end path into piece-wise sub-problems; or methods like ‘‘hello
models [156]. Each method relies on a model that is appropriate for a messages’’ are used to discover the network topology.
different class of mobile objects, including pedestrians, indoor mobile
Routing algorithms can also be classified into deterministic and
users, vehicles, etc. Indeed, these methods rely on an underlying model,
probabilistic algorithms based on the optimization approach or equiv-
which is customized for a specific object with a different mobility
alently the resulting route’s randomness (Fig. 8, layer 2). Deterministic
model such as random walk [161], Random Waypoint (RWP) [162],
HMM [149], Gaussian Morkov Mobility Model (GMMM) [163], Brown- routing refers to a non-stochastic decision-making policy, where the
ian motion [164], Linear model via Durbin’s curve [165], and mixture resulting routes are fully and explicitly determined under given assump-
models [166]. Consequently, they are not applicable to a network of tions, network conditions, and decision rules. On the other hand, in
heterogeneous nodes and fail in balancing between the randomness and probabilistic algorithms, the resulting routes are probabilistic; hence
predictability of node’s mobility. the actual paths are selected in run-time by the nodes based on a set of
rules and probabilities.
7.3. Mobility-related networking challenges From a different standpoint, we can classify routing protocols into
static and dynamic routing protocols (Fig. 8, layer 3). In static routing
High mobility nodes, especially when not properly predicted, pose protocols, the route is established based on the initial network topology
critical challenges to the communication performance in terms of con- without considering the changes which occur during the transmission.
nectivity and routing optimality. Mobility of nodes translates to the These algorithms are appropriate for static networks and low-volume
network dynamicity that can disrupt the information exchange by transmissions. On the other hand, in dynamic routing protocols, the
losing connectivity and undermining the routing efficiency. In extreme resulting end-to-end path can change over time to accommodate node
cases, the network can breakdown into isolated islands. Different ap-
mobility. Therefore, they are more suitable for UAV networks and will
proaches can be taken to overcome the loss of connectivity in UAV
be discussed in more detail in this paper. Dynamic routing algorithms
networks. One main approach is topology control to avoid connectivity
have different variants based on how paths are determined in response
issues, which can be seen as jointly optimizing the networking and
to network topology changes. Main categories include proactive, re-
control aspects. For instance, the idea of the dynamic placement of new
nodes in locations to cover connectivity holes is introduced in [167]. active, hybrid, position-based, topology predictive, and self-adaptive
ML methods can assist with achieving this goal by modeling and learning-based routing methods.
predicting network topology, traffic mobility, spectrum availability, To investigate the performance of different routing protocols in
and channel states. An online learning procedure is used in [168] UAV networks, we first present the key characteristics of these routing
to adjust the UAVs to their radio transmission parameters, based on methods in Table 2. Next, we review different implementations of each
the perceived topology while revising their flight paths. The authors category and investigate their use for highly dynamic UAV networks.
of [169] propose a Chaotic Ant Colony Optimization approach (CA-
COC) to maximize the coverage area while preserving the connectivity.
Another method to improve network connectivity is ECORA [170]. This 8.1. Conventional routing protocols
method uses geographic protocols considering positioning prediction
and link expiration time by excluding links with approaching expiration In this section, we briefly review routing protocols that were mainly
time from the path selection algorithm. These methods usually aim introduced for low-speed ad hoc networks. These routing protocols
to enhance network connectivity by controlling the network topology. do not adapt to high mobility and abrupt changes we find in UAV
Recently, the idea of using predictive and self-adaptive learning-based networks. Therefore, most of them are not applicable for high-speed
routing protocols gained a lot of attention to use ML methods to UAV applications.
enhance networking efficiency. One approach is predicting network
topology changes and incorporating the predicted network topology
into the networking decisions [105,171]. These methods are more 8.1.1. Static routing protocols
appropriate for separating the networking layer from mission-based Static protocols are mainly designed for networks with static or
path-planning algorithms. Also, prediction-based positioning is more slow-varying topology, meaning that the optimal end-to-end path for
appropriate for scenarios that the nodes have limited connectivity any source–destination pair does not change over time. Static routing
or on-demand link establishments which are costly to share location algorithms consider the initial network topology when finding the
information, nodes which are prohibited from sharing their locations best path. Generally, in this approach, the global network topology
for privacy and security concerns, nodes which belong to different
is known to a central node (which can also be shared with network
and even adversary and competing networks, and nodes whose motion
members). Therefore, the optimal paths for all source–destination
trajectories are subject to constant changes due to using autonomous
nodes are calculated and programmed in terms of routing tables. In
on-the-fly control systems.
other words, each intermediate node receives a packet, passes the
8. Routing protocols for UAV networks packet to the next node determined by the routing tables based on
the destination address. These algorithms are suitable for structured
This section reviews routing protocols, emphasizing the role of ML networks, but some modified versions are proposed for UAV net-
methods to accommodate the requirements of UAV networks. works. Examples of such algorithms include shortest path algorithm
Routing protocols can be categorized in many different senses, as (Dijkstra’s and Bellman–Ford’s algorithms) [172], shortest-path-aided
shown in Fig. 8. Regardless of their reaction to topology changes, back-pressure [173], Multi-Level Hierarchical Routing (MLHR) [174],
routing algorithms can be divided into centralized and distributed algo- Load Carry and Deliver Routing (LCDR) [174], and Data-Centric Rout-
rithms (Fig. 8, layer 1). In centralized algorithms, the global network ing [174].

10
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Fig. 8. Routing classification based on different criteria and dynamic types of routing protocols.

Table 2
Types of routing.
Routing protocol Central/distributed Deterministic/Probabilistic Scalable Mobility Global info Discovery message AI aspect
Static Central Deterministic Yes Static Yes No No
Position-based Distributed Deterministic Yes Low speed No No No
Proactive Distributed Deterministic Bigger overhead Low speed Yes Hello message No
Reactive Distributed Deterministic Longer RREQ Low speed No RREQ & RREP No
Hybrid Distributed Deterministic Complexity Low speed No Hello message No
Hierarchical Distributed Deterministic Yes, into clusters Low speed Yes Hello message No
Probabilistic Distributed Probabilistic Adds complexity Low speed Yes No No
Topology predictive Distributed Deterministic Yes Dynamic No Depends Yes
Self-adaptive learning-based Distributed Deterministic Yes Dynamic No Depends Yes

8.1.2. Proactive routing protocols 8.1.3. Reactive routing protocols

Proactive routing protocols can be implemented as table-driven Reactive routing protocols create on-demand routing information.
methods, where optimal paths are found for all source–destination pairs It means that the route discovery process is executed only when a
transmission session has to be established. The main benefit of this
based on the global network topology. The routing tables are filled
approach is its reduced overhead, especially in the low-traffic regime.
accordingly at all nodes to guide the packets link-by-link to their final
On the other hand, in case of a route failure, the re-establishment of
destinations. In applications like UAV networks, routing tables should
a new route can take a long time. The following are some important
be updated if optimal paths are changed due to varying network topol- implementations of reactive routing protocols: Dynamic Source Routing
ogy or link budgets. The main advantage of this approach is that the (DSR) [185], Ad hoc On Demand Distance Vector (AODV) [186],
route can easily be calculated and established. In case of a link break- Dynamic Topology-Multipath AODV (DT-MAODV) [187], Associativity-
down, new links can be re-established quickly. However, it may impose Based Routing (ABR) [188], Signal Stability-based Adaptive routing
a large overhead for time-consuming topology exploration and path dis- (SSA) [189], Message Priority and Fast Routing (MPFR) [190], Dy-
covery, such as using hello packets to learn the network topology. Some namic Backup Routes Routing Protocol (DBR2P) [191], Dynamic
important implementations of proactive routing algorithms designed MANET On-demand (DYMO) [192] and Time Slotted On-demand
for UAV networks include Optimized Link State Routing (OLSR) [175], Routing (TSOR) [193].
Directional Optimized Link State Routing (DOLSR) [176], Multidimen-
8.1.4. Hybrid routing protocols
sional Perception and Energy Awareness OLSR (MPEAOLSR) [177],
Hybrid routing protocols combine proactive and reactive routing
Dynamic Dual Reinforcement Learning Routing (DDRLR) [178], Desti-
features. The route is initially determined with a proactive protocol.
nation Sequenced Distance Vector (DSDV) [179], BABEL [180], Cluster However, a reactive routing protocol is activated when a substan-
head Gateway Switch Routing (CGSR) [181], Wireless Routing Protocol tial network topology is recognized or a previously established route
(WRP) [182], Topology Broadcast based on Reverse Path Forward- breaks. Some important implementations of hybrid routing algorithms
ing (TBRPF) [183] and Better Approach To Mobile Ad hoc Network designed for UAV networks are Zone Routing Protocol (ZRP) [194] and
(BATMAN) [184]. Temporarily Ordered Routing Algorithm (TORA) [195].

11
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

8.1.5. Position-based routing protocols • Predictive Dijkstra’s [80,105]: This routing protocol assumes
Position-based routing protocols find the optimal route based on that the intermediate nodes’ locations when the packet is sup-
the location information. For example, the next node can be selected posed to meet them are predicted using ML methods. Then, it
based on its distance to the current node or to the destination. The incorporates this predictive information into the path selection
key disadvantage of these methods are their dependence on accurate criterion based on Dijkstra’s shortest path algorithm. Results show
positioning and tracking systems. Some important implementations of superior performance compared to the standard Dijkstra’s algo-
position-based routing algorithms proposed for UAV networks include: rithm, especially when higher velocities are applied. The achieved
Greedy Perimeter Stateless Routing (GPSR) [196], Greedy–Hull–Greedy performance gain is dependent upon the prediction accuracy. Two
(GHG) [197], Greedy–Random–Greedy (GRG) [198], greedy forward- important shortcomings of this method is its reliance to accurate
ing [199], Energy-Balanced Greedy forwarding Routing (EBGR) [200], trajectory prediction methods, and the need for global location
Greedy Distributed Spanning Tree Routing (GDSTR) [201], Cross-layer
information exchange.
Link quality and Geographical-aware beaconless Opportunistic routing
• Predictive greedy routing [32]: Distance-based greedy routing
(Xlingo) [202], Adaptive Forwarding Protocol (AFP) [203], Reactive-
algorithm solves the issues of [105] and relies solely on the UAVs’
Greedy-Reactive (RGR) [204], scoped flooding and delayed route re-
local observations of their surrounding subnetwork. Each node
quest RGR [205], beaconless opportunistic routing [206], Location-
estimates the location of its neighbors (e.g., using model-based
Oriented Directional MAC (LODMAC) [207], Extremely Opportunistic
Routing (ExOR) [208] and Location-Aided Routing (LAR) [209]. object motion trajectory prediction) and selects the next node
that makes maximum progress toward the destination node. This
8.1.6. Hierarchical routing protocols method adapts to highly dynamic network topology. Moreover, it
Hierarchical routing protocols consider nodes arranged hierarchi- has low complexity and low overhead with no need for an initial
cally, where the lower layers can form clusters. Each node typically route setup. Simulation results show considerable improvement,
holds information only about its neighbors stored in a table that is compared to centralized shortest path routing algorithms.
updated through hello packets. Each cluster head communicates with • Predictive Optimized Link State Routing (P-OLSR) [217,218]:
the rest of the cluster heads to select the best path. Cluster-Based Rout- This routing protocol is an advanced version of OLSR routing
ing Protocol (CBRP) [210], Enhanced Cluster head Gateway Switch protocol. This algorithm exploits GPS information and calculates
Routing (ECGSR) [211] and Fisheye State Routing (FSR) [212,213] are an Expected Transmission (ETX) count metric to estimate the
examples of recently-developed algorithms for UAV networks. quality of the link when finding the optimal path. Numerical
results show that the P-OLSR outperforms other algorithms such
8.1.7. Probabilistic routing protocols as OLSR and BABEL under dynamic network topology.
Probabilistic routing protocols find multiple routes from source to • Geographic Position Mobility Oriented Routing (GPMOR)
destination, which can be selected based on probabilistic mechanisms [219]: This routing protocol proposes an efficient and effective
to cope with network congestion and security. Some examples of these geographic-based routing protocol that uses the Gauss–Markov
algorithms used in UAV networks are random walk routing [214] and mobility model to predict the movement of UAVs to eliminate the
MIMO-based random walk routing [215].
impact of highly dynamic movements. This approach selects the
next hop according to the mobility relationship in addition to the
8.2. AI-enabled routing protocols
Euclidean distance to make more accurate decisions. Experiment
results show that this approach provides effective and accurate
In this section, we study the AI-enabled routing protocols, which
use the learning power of ML algorithms for optimal route path se- data forwarding solutions. Moreover, it decreases the impact of
lection based on a more accurate perception of the network topology, intermittent connectivity and achieves a better latency and packet
channel status, user behavior, traffic mobility, etc. These algorithms delay rate than other position-based routing protocols.
bridge the two networking and AI research areas to implement modern • Mobility Prediction Clustering Algorithm (MPCA) [220]: This
networking, especially for dynamic UAV networks. These algorithms routing algorithm is appropriate for clustered UAV networks. It
can be viewed as state of the art and are not included in most previous finds the highest node reliability to select the cluster head. Then,
survey papers. The following is a fairly comprehensive list of AI-based it predicts the network topology using the Trie data structure
algorithms. dictionary prediction and link expiration time mobility model.
Also, this approach ensures the stability of cluster formation.
8.2.1. Topology predictive routing protocols • Robust and Reliable Predictive (RARP) [221]: This routing
The main feature of topology-predictive routing protocols is using protocol combines omnidirectional and directional transmission
ML algorithms to predict the node’s motion trajectories (as an approx- schemes with dynamic angle adjustment. This method features a
imate of the network topology, if the communication range of nodes is hybrid use of unicasting and geocasting routing protocols using
known) and incorporate them into the path selection mechanism. the location and trajectory information. The intermediate node
Here, we review some of the proposed routing protocols that use locations are predicted using 3-D estimation; then, directional
mobility or trajectory prediction approaches to enhance the perfor- transmission is used toward the predicted location, enabling a
mance of routing algorithms for UAV networks. longer transmission range and tracking topology changes. The
authors show that their method reduces path re-establishment
• Learning-based Adaptive Position MAC protocol [216]: This
routing protocol proposes an adaptive hybrid communication pro- and service disruption time and achieves higher successful packet
tocol by integrating a novel Position-Prediction-based directional delivery rates.
MAC protocol (PPMAC) and a Self-learning Routing Protocol • Scoped Flooding and Mobility Prediction-based RGR (SFM-
based on Reinforcement Learning (RLSRP). The performance re- PRGR) [222]: This algorithm is a modified version of RGR. This
sults show that the proposed PPMAC overcomes the directional method associates with data packets mobility prediction informa-
deafness problem, which happens when the transmitter fails to tion, including velocity, direction, and timestamp, to compute the
communicate with the receiver due to having the receiver’s an- distance between the current node and its neighbors. Then, if the
tenna oriented in a different direction. Also, RLSRP provides next hop is out of range, the approach directly switches to GGF
an automatically evolving and more effective routing scheme, to save dropped data packets, making it a better approach for
appropriate for autonomous FANETs. dynamic networks.

12
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

• Q-learning-based Geographic adhoc routing protocol (QGeo)

[223]: This is an ML-based geographic routing scheme to re-
duce network overhead in high-mobility scenarios. The basic idea
is that nodes make geographic routing decisions distributively,
utilizing a reinforcement learning method without knowing the
entire network topology. It consists of location estimation, a
neighbor table, and a Q-learning module. The location estimation
module updates the current location information reported by
the GPS or other localization methods. Their results show that
QGeo provides a higher packet delivery rate and a lower network
overhead than previously reported routing protocols.
• Predictive Ad-hoc Routing fueled by Reinforcement learning
and Trajectory Knowledge (PARRoT) [224]: This is another
ML-powered routing protocol, which exploits mobility control
information for integrating knowledge about the future motion of
the mobile agents into the routing process. Each agent estimates
its own future position based on the current position and prop-
agates the result to other nodes. This algorithm achieves higher
robustness and a significantly lower end-to-end latency compared
to similar algorithms previously reported. This algorithm is ap-
propriate for separating networking from path planning layer
or when the paths are planned on the fly, since the nodes are
assumed to be unaware of their future locations. Fig. 9. Illustration of the RL-based routing [227].
• Fuzzy Logic Reinforcement Learning-Based Routing Algo-
rithm (FLRLR) [225]: This algorithm uses fuzzy logic to de-
termine the neighbor nodes in real-time. Then, by using the
• Q-Routing [228]: The first proposed Q-Routing protocol operates
future reward method of reinforcement learning, this method
based on learning from experience. Each node stores the expected
reduces the average number of hops through continuous training.
time to the destination through any of its neighbors as Q-values
Simulation results show lower average numbers of hops and
high link connectivity, compared to the Ant Colony Optimization in a Q-table. Each node selects the next node that minimizes the
(ACO) algorithm. expected travel time to the destination. Once a packet is received
by a node, it sends back the real travel time to the previous node
8.2.2. Self-adaptive learning-based routing protocols to updates its Q-values for the next round.
Most learning-based routing protocols use Reinforcement Learning • Predictive Q-Routing (PQ-Routing) [230]: This method is an ex-
(RL) to make routing decisions by continued and online learning of the tension of the conventional Q-Routing that addresses the
environment and their decision consequences on desired performance exploration–exploitation trade-off and fine-tunes the routing poli-
metrics such as delay, throughput, energy efficiency, and fairness. A key cies for the low network loads. Their approach was based on
advantage of RL-based algorithms is their abstract formulation which learning and storing new optimal policies under decreasing load
brings independence from topology prediction and channel estimation. conditions and reusing the learned best experiences by predicting
This comes from the concept of learning from experience. the traffic trend. Their idea was to re-investigate the paths that
The concept of RL for optimized routing is shown in Fig. 9. Initially, remain unused for a while due to the congestion-related delays.
the scenario is represented by state 𝑠1 , where node or agent 𝐴1 has The probing frequency is an adjustable parameter to be tuned
two candidate neighbors 𝐴2 and 𝐴3 to send its packet. Consequently, based on the path recovery rate estimate. Their proposed results
we must select between action 𝑎1 or 𝑎2 based on the reward expected showed that PQ-Routing outperformed the Q-Routing in terms
for each action 𝑎 at state 𝑠, defined as 𝑄(𝑠, 𝑎). Once we select the of both learning speed and adaptability. However, PQ-Routing
appropriate action, the agent 𝐴1 obtains an immediate reward from requires large memory for the recovery rate estimation. Also,
the environment, 𝑟1 or 𝑟2 , respectively. Next, we start the same pro- it is not accurate in estimating the recovery rate under varying
cess in a new state 𝑠2 , where decisions are made based on the new topology change rates (e.g., when nodes start to moving faster).
environmental conditions and the learned policy in terms of actions- Furthermore, this method only works for delays arising from the
rewards relations. The end goal is to find an optimal policy in which
queuing congestion and not delays coming from the network
the cumulative reward over time is maximized by assigning optimal
topology change.
actions to each state [226].
• Dual Reinforcement Q-Routing (DRQ-Routing) [231]: The key
RL-based routing was first introduced in [228], where Q-Routing
idea of this algorithm is to use forward and backward explorations
considered packet forwarding as an application of Q-learning. This
by the sender and receiver of each communication hop, by ap-
method demonstrated superior performance compared to a non-
pending information to the packets they receive from their neigh-
adaptive algorithm based on the pre-computed shortest paths [229].
The essence of Q-Routing is gauging the impact of routing strategies bors. Simulation results prove that this method learns the opti-
on a desired performance metric by investigating different paths in the mal policy more than twice faster than the standard Q-Routing.
exploration phase, and using the discovered best paths in the exploitation A comparative analysis of learning-based routing algorithms is
phase. Exploration imposes an overhead to the system, but is critical provided in [232], where the performance of the self-adaptive
for finding newly emerged optimal paths, especially when the network Q-Routing and dual reinforcement Q-Routing algorithms are com-
topology undergoes substantial changes. An inherent challenge is to pared against the conventional shortest path algorithms. Their
adaptively solve the trade-off between the exploration and exploitation results showed that the Q-learning approach performs better than
times to accommodate the dynamicity of the network topology. the traditional non-adaptive approach under scenarios with in-
The following is a summary of learning-based routing protocols creasing traffic that causes node and link failures. However, Q-
mainly based on Q-Routing to show they evolved over time to better Routing does not always guarantee finding the shortest path and
serve dynamic UAV networks. does not explore multiple forwarding options in parallel.

13
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

• Credence-based Q-Routing (CrQ-Routing) and Probabilistic Q-network algorithm as a traffic balancing strategy. Then, the
Credence-based Q-Routing (PCrQ-Routing) [233]: These two protocol makes routing decisions based on the evaluation of each
methods dynamically capture the traffic congestion to improve wireless link. Results show an improving performance in terms of
the learning process to select less congested paths. Both methods packet delivery ratio and end-to-end delay.
adapt to the current network conditions much faster than the • Q-learning based Multi-objective optimization Routing pro-
conventional Q-Routing. tocol (QMR) [242]: This novel Q-learning-based multi-objective
• Full-echo Q-Routing [228]: Another technique proposed to ac- optimization routing protocol adaptively adjusts the learning pa-
celerate the learning speed of conventional Q-Routing is the rameters based on the dynamicity of the network. The authors
full-echo approach. In conventional Q-Routing, each node only proposed a new mechanism to explore some undiscovered poten-
updates the Q-values for its selection (the best neighbor). In tial optimal paths while exploiting the acquired knowledge by
contrast, in the full-echo routing, a node gets each neighbor’s re-estimating neighboring relationships to select the more reliable
estimate of the total time to the destination, which helps update next hop. Results show a higher packet arrival ratio, lower delay,
the Q-values accordingly for each of the neighbors. and energy consumption than the preceding Q-learning-based
• Full-echo Q-Routing with Adaptive Learning Rate [234]: A routing methods.
more recent work added adaptive learning rates to the full-echo • Q-Learning-based Fuzzy Logic for Multi-Objective Routing
Q-Routing to improve the exploration performance. Results show algorithm in Flying Ad Hoc Networks (QLFLMOR) [243]: This
that this technique reduces the oscillations of the full-echo Q- multi-objective Q-learning-based fuzzy logic algorithm facilitates
Routing for high load scenarios. the selection of the routing paths in terms of the per-link and over-
• Adaptive Q-Routing with Random Echo and Route Memory all path performances. The optimal routing path to the destination
(AQRERM) [235]: An extension of the previous work is AQRERM, is determined by each UAV using a fuzzy system with link-level
which improves the performance of the baseline method in terms and path-level parameters. The link-level parameters include the
of the overshoot and settling time of the learning process, as well transmission rate, energy state, and flight status between neigh-
as the learning stability. bor UAVs, while the path-level parameters include the hop count
• Poisson’s probability-based Q-Routing (PBQ-Routing) [236]: and successful packet delivery time. Simulation results show that
This approach uses forwarding probability and Poisson’s proba- the proposed method can maintain low hop count and energy
bility for decision making and controlling transmission energy for consumption and prolong the network lifetime.
intermittently connected networks. Results show that the delivery • Adaptive UAV-assisted Geographic Routing with Q-Learning
probability is almost twice bigger than for Q-Routing, and the (QAGR) [244]: This algorithm uses fuzzy-logic and Depth-First-
overhead ratio is reduced to half. Search (DFS) algorithms to calculate the global path routing.
• Delayed Q-Routing (DQ-Routing) [237]: This routing protocol As it is designed for UAV-Assisted networks, the vehicle on the
uses two sets of value functions to carry out random delayed ground maintains a fix-sized Q-table that converges with a well-
updates to reduce the overestimation of the value function and designed reward function and forwards the routing request to the
improve the rate of convergence. Experiments show that this optimal node by looking up the Q-table filtered according to the
method also improves the learning rate. global routing path. Results show a good performance in packet
• QoS-aware Q-Routing (Q2 -Routing) [238]: This method in- delivering and end-to-end delay.
cludes a variable learning rate based on how big are variations • Fully-Echoed Q-Routing with Simulated Annealing Inference
in Q-values and ensures the traffic Quality of Service (QoS). for Flying Adhoc Networks (FESAIQ-Routing) [227]: This rout-
Results show that this method outperforms the well-known ad- ing protocol is a full-echo Q-Routing with an adaptive learning
hoc routing algorithms in dynamic environments under QoS rate controlled by Simulated Annealing (SA) optimization, where
constraints. the temperature parameter captures the influence of the nodes’
• Q-Network Enhanced Geographic Ad-Hoc Routing Protocol mobility on the Q-value update rates. The soft variation of the
Based on GPSR (QNGPSR) [239]: This routing protocol uses Q- exploration rate not only optimizes the exploration rate, but also
network as an approximator to estimate the quality of different accommodates abrupt changes in the network dynamicity. Simu-
routing paths. Then, it makes forwarding decisions based on the lation results show better performance than previous state-of-art
estimated Q-values. Also, the neighbor topology information is in- Q-Routing algorithms.
troduced to estimate the environment and node states. QNGPSR is
trained off-line when the signal propagation model is determined. A summary of some of these routing protocols is presented in
Therefore, online-learning is not necessary, which reduces the Tables 3 and 4 with some characteristics for each routing protocol,
computational load. Results show a higher packet delivery ratio as well as some comparative results to provide an idea of how these
and a lower end-to-end delay compared to the original GPSR. routing protocols perform under certain circumstances.
• Adaptive and Reliable routing protocol with deep reinforce-
ment learning (ARdeep) [240]: This is a deep RL-based adaptive 9. Tools and public datasets
and reliable routing protocol that formulates routing decisions
with a Markov Decision Process model to characterize the net- In this section, we review tools and public datasets available for
work variations automatically. It considers link status, the packet simulating real UAV networking environments. We investigate their
error ratio, the expected connection time of the link, the re- features and capabilities, emphasizing their use for testing networking
maining energy of nodes, the distance between the node and solutions (e.g., routing protocols) for UAV networks under different
the destination when making routing decisions to precisely infer conditions.
the network environment and make more appropriate forwarding
decisions. Simulation results show that ARdeep outperforms the 9.1. Simulation tools
existing QGeo and conventional GPSR routing protocols.
• Traffic-aware Q-Network enhanced routing protocol based UAV simulation tools emulate virtual environments to model UAV
on GPSR (TQNGPSR) [241]: Traffic-aware Q-network enhanced flights in close-to-reality situations. It gives the convenience of evaluat-
geographic routing protocol based on Greedy Perimeter Stateless ing the performance of UAV networks in virtual environments at much
Routing (GPSR). This algorithm uses the congestion information lower costs and trouble before testing in real scenarios. The choice of
of neighbors and evaluates the quality of a wireless link by the the appropriate simulator depends on both the testing objective and

14
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Table 3
Performance comparison table for topology predictive routing protocols.
Algorithm Objective performance Results
Energy Delay Throughput Connectivity
Compared to DMAC, LODMAC, OLSR, GPMOR, RARP, IMAC+OLSR:
∙ Lowest network delay and longest path lifetime
PPMAC+RLSRP [216] ✓ ✓ ✓ ∙ Highest route setup success and data delivery ratio.
∙ Better successful throughout without retransmissions.

Compared to conventional Dijkstra’s algorithm:

Predictive Dijkstra’s [80,105] ✓ ∙ Up to 25% decrease in end-to-end delay for 100 nodes.

Compared to conventional Dijkstra’s and static greedy:

∙ Higher probability of success (up to 100%).
Predictive greedy routing [32] ✓ ✓
∙ Energy consumption reduction (up to 30%).

Compared to OLSR and BABEL:

∙ Cuts down the outage time by at least 85%.
P-OLSR [217,218] ✓ ✓
∙ Achieves a more stable goodput.

Compared to GPSR and GLSR:

∙ Better packet delivery ratio (up to 250%).
GPMOR [219] ✓ ✓
∙ Lower average delay of the network (up to 50%).

Compared with LID, HD and WCA:

∙ Up to 500% increase in clusterhead duration.
MPCA [220] ✓ ✓
∙ Up to 70% lower reaffiliation frequency.

Compared to conventional AODV:

∙ Up to 30% increase in route setup success rate.
RARP [221] ✓ ✓
∙ Higher average path lifetime.

Compared to AODV, RGR, and MPRGR:

∙ 10% increase in packet delivery ratio.
SFMPRGR [222] ✓ ✓
∙ 20% reduction end-to-end delay.

Compared to GPSR and QGrid:

∙ Up to 50% increase in packet delivery ratio
QGeo [223] ✓ ✓
∙ 20% reduction in end-to-end delay.

Compared to AODV, OLSR, GPSR and B.A.T.M.A.N:

∙ Higher packet delivery ratio (at least by 45%).
PARRoT [224] ✓ ✓
∙ Lower end-to-end latency

Compared to Fuzzy Logic and ACO:

∙ Up to 20% lower average number of hops.
FLRLR [225] ✓ ✓
∙ Around 30% higher connectivity.

the list of features offered by each simulator. Some simulators incorpo- best simulators currently available. By combining AI components, these
rate the Motion of Capture (MOCAP), which allows simulating UAVs’ simulators allow researchers to simulate their algorithms in near-real
natural movements [257]. Another tool is MAVLink, a lightweight mes- environments, with the opportunity to develop better algorithms for
saging protocol for communicating with drones to test communication AI-based networking and control tasks.
protocols and algorithms. Software In The Loop (SITL) is a hardware-
free simulation environment that facilitates simulating real-time UAV
9.2. Experimental platforms
operations. It includes a c++ code to directly implement autopilot
operation on the user’s computer for testing [258].
The list of tools for simulating UAV networking is large and still Experimental platforms enable testing networking protocols in em-
growing. A comparative analysis of some popular simulation tools, ulated network setups. Most experimentation platforms can be exe-
including X-Plane [247], FlightGear [248] (compatible with MATLAB cuted on a standalone computer or High-Performance Computing (HPC)
Simulink), Gazebo [249], JMavSim [250], Microsoft AirSim [251], servers. However, larger experimentation platforms typically consist of
and UE4Sim [252], is presented in Table 5. The first four simulate custom-built hardware with a set of simulation and operation software
the UAV motions solely based on physics laws and do not support packages, programming environment, and web-based user interface to
MOCAP. On the other hand, Microsoft AirSim and UE4Sim support enable remote experimentation for the research community.
MOCAP by using Unreal Engine 4 (UE4), an open-source tool that The most commonly-used experimental platforms are Network Sim-
simulates UAV movement using physics along with a high-quality ulator (NS-3) [261], and OPNET [262]. NS-3 is an open-source, free,
trajectory creation engine. AirSim is considered a platform for both AI and discrete-event network simulator for Internet and networking sys-
research and training [259]. AirSim is empowered with deep learning, tems, enabling testing different layers of networking protocols, includ-
computer vision, and reinforcement learning features to generate and ing routing protocols in MAC and Network layers. OPNET is an open
utilize training datasets [260]. Also, UE4Sim simulator benefits from network simulator used to simulate the function and performance of
a built-in and robust DL-based approach for real-time autonomous different networking systems. It is known for its power and versatility to
driving that does not require manually collected training data. For create and simulate different network topologies. OPNET Technologies
this reason, Microsoft AirSim and UE4Sim are considered two of the considers requests for free access for academic use.

15
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Table 4
Performance comparison table for self-adaptive learning-based routing protocols.
Algorithm Objective performance Results
Energy Delay Throughput
In comparison to Q-Routing and Dual Q-Routing:
Adaptive Q-Routing Full-
✓ ∙ Smaller average delivery time.
echo [234]

Compared to Q-Routing, DRQ-Routing and AQFE:

∙ Better overshoot and settling time of the learning.
AQRERM [235] ✓
∙ Lower average delivery time.

Compared to Q-Routing, Epi-R, PRoPHET and HBPR:

∙ Delivery probability almost gets doubled.
PBQ-Routing [236] ✓ ✓ ✓
∙ Lower energy use and overhead reduced to half.

Compared to AODV and EQ-Routing:

∙ Lower packet delay.
Q2 -Routing [238] ✓ ✓
∙ Higher packet success ratio.

Compared to Q-Routing:
∙ Much higher average reward of DQ-routing.
DQ-Routing [237] ✓
∙ More likely to choose best action.

Compared to OLSR, AODV and GPSR:

∙ Up to 65% lower end-to-end delay.
QNGPSR [239] ✓ ✓
∙ Up to 35% higher packet delivery ratio.

Compared to QGeo and conventional GPSR:

∙ Up to 30% higher packet delivery ratio.
ARdeep [240] ✓ ✓
∙ Up to 30% lower average end-to-end delay.

Compared with AODV, OLSR, GPSR, and QNGPSR:

∙ Outperforms in terms of the packet delivery ratio, end-to-end delay, and
TQNGPSR [241] ✓ ✓
throughput.

Compared to other Q-learning based routing methods:

QMR [242] ✓ ✓ ✓ ∙ Higher packet arrival ratio, lower delay and energy consumption.

Compared to conventional fuzzy logic and Q-value-based AODV:

QLFLMOR [243] ✓ ✓ ✓ ∙ Lower hop count and energy consumption and longer network lifetime.

Compared to ARPRL [245], U2RV [246], GPSR and AODV:

∙ Up to 300% higher in packet delivering ratio.
QAGR [244] ✓ ✓ ∙ Around 50% lower end-to-end delay and up to 90% reduction in average
number of hops.

Compared to Q-R, REE-R, PE-R, AFEQ-R and SAHQ-R:

∙ Reduction in energy consumption between 7% to 82%.
FESAIQ-Routing [227] ✓ ✓
∙ Increase of up to 264% in packet delivery ratio.

Table 5
UAVs simulation tools.
Simulator Free access ROS Interface MOCAP MAVLink SITL Obstacles Usability
X-Plane [247] No No No Yes Yes Yes Medium
FlightGear [248] Yes No No Yes Yes Yes Medium
Gazebo [249] Yes Yes No Yes Yes Yes Easy
JMavSim [250] Yes Yes No Yes Yes No Easy
Microsoft AirSim [251] Yes No Yes Yes Yes Yes Medium
UE4Sim [252] Yes No Yes No No Yes Medium

Table 6
UAVs experimental platforms.
Experimental platform Interface Free access MAVLink Compatible pilot software
QGround-Control [253] Graphical Yes Yes PX4 Autopilot, ArduPilot
Mission Planner [254] Graphical Yes Yes PX4 Autopilot, Ardupilot
MAVProxy [255] Command Yes Yes Ardupilot
UGCS [256] Graphical Yes, but limited Yes DJI, Innoflight, Micropilot, Mikrokopter, Microdrones, Parrot

Some other simulators are developed specifically for UAV net- is tunable to account for a large range of communication fidelity and
works. For instance, ROS-NetSim [263] is a Robot Operating System complexity.
(ROS) package, which acts as an interface between robots (UAVs in Another UAV-specific simulator is UB-ANC University at Buffalo’s
this case) and network simulators. ROS-NetSim accurately replicates Airborne Networking and Communications Testbed (UB-ANC) [264],
Perception-Action-Communication (PAC) loops. Moreover, ROS-NetSim which is an open platform that facilitates rapid testing and repeatable

16
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Table 7
UAVs trajectory datasets.
Trajectory datasets Size/Length of flight Description
Large-scale, aggressive indoor flight dataset collected using a custom-built quadrotor
Blackbird [272] 10 h of flight data from 168 flights.
platform. Over 17 flight trajectories and 5 environments at velocities up to 7.0 m/s.
Trajectories, proxy meshes and images generated for path planning on real and
KAUST [273] 11.172 Gb including all material. synthetic scenes. It includes a benchmarking tool allowing new trajectories to output
camera images for reconstruction.
420k frames representing 79 min of Multi-purpose synthetic dataset for low altitude drone flights. Data corresponding to
Mid-Air [274] drone flight records extracted out of flight records of a flying quadcopter recorded in different climate conditions, including
more than 5 h of flight. a test set for benchmarking.
Over 27 flight sequences, with more Visual-inertial odometry dataset from a drone racing quadrotor with fast laps around a
UZH-FPV [275]
than 10 km of flight distance. racetrack trajectories, as well as free-form indoor and out trajectories around obstacles.

comparative evaluation of airborne networking and communications 9.3. Trajectory datasets

protocols at different layers of the protocol stack. It enables the flexible
deployment of novel communications and networking protocols, with Trajectory datasets consist of recorded data from real-time UAV
emphasis on modularity and extensibility. flights that are useful for simulation purposes. The idea is to test new
A recently established experimentation center is Aerial Experimen- methodologies or protocols using trajectory data that mimic real-world
tation and Research Platform for Advanced Wireless (AERPAW) in the situations. The bigger the size of the dataset, the more variability
North Carolina State University (NCSU) [265,266]. This NSF-funded of flight trajectories is incorporated. In Table 7, we compare some
center integrates drones and 5G wireless technology to provide in- UAV trajectory and imagery datasets available for all users that can
creased coverage and connectivity, high throughput aerial monitoring, be helpful in simulating unmanned aerial monitoring platforms. These
and improved signals and location data. The idea is to allow U.S. datasets include Blackbird [272], KAUST [273], Mid-Air [274] and
researchers to test new ways of increasing wireless speed and capacity UZH-FPV [275]. We observe that each dataset offers UAV motions
in an experimental infrastructure, where nodes are mobile with the with different characteristics. The most appropriate dataset should
ability to transmit and receiving radio/video waves from user devices be selected based on the specific test requirements and conditions.
while moving on demand. This can be convenient under disaster relief There exist many UAV datasets containing UAV imagery, as well as
circumstances, in which existing cellular networks may be damaged. aircraft trajectory datasets. However, they do not exactly reflect UAV
Moreover, it is expected to make an impact in time-sensitive deliv- motion patterns, which is the main feature in path planning and routing
eries, smart agriculture, autonomous driving, and accident control scenarios. Therefore, there is a need to produce more relevant UAV
applications [267]. trajectory datasets. Some methods use real data to train deep learning
Another remote experimentation Platform is the Open Wireless algorithms for trajectory design and recognition [276,277]. Having
Data-driven Experimental Research (POWDER) that operates as a highly access to accurate trajectory datasets will allow researchers to develop
flexible, remotely accessible, end-to-end software defined platform better solutions for real-world scenarios.
supporting a broad range of wireless and mobile related research [268].
This NSF-funded center has many features and capabilities including a 10. Future trends and remaining challenges
massive MIMO base station and Software-Defined Radios (SDRs) that
advances competitors in scale, realism, diversity, flexibility, and access. Despite the recent advances in developing ML-powered networking
More advanced routing protocols which require function virtualiza- protocols for UAV networks, there still are challenges and issues that
tion and SDR technology can be tested in this environment before would be the center of attention for coming years. From the commu-
implementing in UAV networks. nication perspective, most technical challenges arise from the limited
We can also find experimental platforms that act as a ground control payload, processing power, and structure-free and highly dynamic na-
station operating artificial UAVs. These platforms can be used to test ture of unmanned aerial platforms [278–282]. This section summarizes
communication protocols developed for UAV networks in simulated part of these challenges and the future outlook of UAV networking
environments. Depending on the test scenario and objectives, we can technology from different perspectives, including AI integration, energy
select from a list of available software packages. These platforms in- efficiency, security, regulations, etc.
clude QGround-Control [253], Mission Planner [254], MAVProxy [255]
and UGCS [256], as presented in Table 6. 10.1. AI integration
Some experimental platforms use a graphical interface for user
convenience, and some are command-line-based to provide more flexi- Using AI to accelerate networking is the dominant research trend,
bility. We analyze the inclusion of MAVLink and the tools’ compatibility as discussed in this paper. AI is shown to offer superior performance
with pilot software. We found that PX4 Autopilot, Ardupilot, ROS, for communication, control, and operation of autonomous UAVs un-
DJI Pilot, Innoflight, Micropilot, Mikrokopter, Microdrones, and Parrot der different networking scenarios [25–27]. Moreover, it can opti-
are open-source autopilot systems capable of controlling autonomous mize network management and reduce their complexity [283]. We
vehicles, with a variety of aircraft operation scenarios, such as aerial believe that the current networking designs have not yet fully uti-
mapping, surveying, and more applications. lized the power of AI-based solutions, and further research is on the
It is worth mentioning that some of the mentioned experimental way to integrate AI and networking paradigms by using more ad-
platforms offer AI capabilities for networking research. For instance, vanced ML algorithms [67], RL-based decision-making [284], and deep
NS-3 has NS3-GYM [269], and NS3-AI [270] extension modules that learning [285] for different aspects of networking, including sensing,
enable applying AI to network simulations in NS-3. The key idea is to scheduling, routing, spectrum sharing, path planning, and resource
provide a high-efficiency solution to allow data interaction between NS- allocation.
3 and other AI frameworks and encourage the use of AI in networking However, it is worth mentioning that the use of AI-based methods
research. Also, AERPAW and POWDER contain AI embedded into hard- can bring up new challenges, and needs to be studied carefully. For
ware for superior detection, tracking, and classification of UAVs [271], instance, AI methods can help to predict the future locations of the
as well as for spectrum-maximizing resource allocations. network nodes and potential link losses, and hence improve the power

17
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

consumption by avoiding the transmission of hopeless packets that 10.4. Energy efficiency
will be lost during the transmission. However, the use of AI methods
can add to the computation complexity and CPU power use in power- UAVs are highly constrained in payload and battery lifetime. The
constrained UAVs. Therefore, the right choice of the routing protocol available energy should be optimally used for navigation, sensing, ac-
should be based on the application-specific requirements and design tuation, transmission, and data processing [290]. Therefore, developing
constraints. Then, as a future solution, we need to study if the effect energy-efficient networking to prolong mission time and extend cover-
of including AI techniques into our model is worth the complexity that age area is usually considered a top priority in UAV networks [291,
will be added, which could be affecting other important aspects such 292].
as the network lifetime and power consumption. A parallel research direction to solve energy issues of UAVs is
developing new battery technologies, such as hydrogen fuel cells [293],
and enhanced lithium-ion batteries [294]. Also, Radio Frequency (RF)
10.2. Connectivity transmission can be used for Wireless Power Transfer (WPT) [295],
which provides controllable and sustainable energy supply for UAVs
Maintaining connectivity for UAV networks remains an important [296,297]. Further research is required to improve the use of WPT
issue considering the limited communication range of commercial by reducing the distance between charging stations and UAVs, the
UAVs, which is typically limited to a few miles, while the coverage area random energy arrivals, and the scalable nature of UAVs [298,299].
in some applications like forest fire monitoring, disaster management, Another research direction is considering mmWave communications for
and wildlife monitoring can scale to hundreds of miles [286–288]. UAV networks, energy beamforming, and the optimized placement of
Connectivity loss can cause packet drop, frequent link re-establishment, wireless charging stations. Integration with 5G, and 6G wireless systems
shortened link lifetime, prolonged delays, and ultimately disrupt the are also research topics related to energy optimization. A few routing
mission and compromise the user QoE. Recent methods use ML meth- protocols consider energy as a routing criterion. Developing multi-
ods to predict network topology change ahead of time and avoid objective and constrained optimization methods for routing protocols
connectivity loss by learning-based routing [105,220,222,224]. An- to enhance routing efficiency, while maintaining maximal connectivity
other approach is including the link remaining time or path lifetime and minimizing energy consumption is an important future direction.
into the routing criteria [37]. Other methods try to identify and resolve Further research is required to develop energy-aware routing protocols.
Some recent works offer storing data and postponing calculations to the
the connectivity loss by methods such as placement of new UAVs and
future to reduce power consumption while flying [300].
link re-establishment mechanisms [219].
Another potential solution can be using energy-efficient ML and
One of the potential future trends would be integrating online
DL algorithms for networking, such as energy-efficient convolutional
path planning methods with networking algorithms to realize topology
neural networks [301]. However, new challenges come into place when
control with minimal connectivity issues. This approach is feasible in
we use the mentioned techniques. First, it is misbelieved that reducing
many real-world scenarios where the coverage area is defined, but there
the energy consumption of the algorithms does not necessarily lead
is a high degree of freedom in UAV’s motion paths (e.g., search and
to a reduction of the overall energy consumption. Second, in some
rescue scenarios, regular forest monitoring, etc.). Also, connectivity is
scenarios, measuring the energy consumption becomes redundant since
dependent on the interference caused by objects and environmental energy and time are correlated, and time is already measured. Third,
factors. Therefore, using more advanced AI methods to predict the it might be hard to measure the energy consumption, making it time
influence of network nodes, environment, and surrounding objects on consuming and impractical [302]. Despite a few scenarios where these
networking quality, can mitigate connectivity issues. In some appli- statements are true, we can find that reducing the energy consumption
cations with sparsely distributed nodes, intermittent connectivity is of the algorithms used will impact positively the overall energy con-
unavoidable. Also, connectivity can be caused by UAVs that ran out sumption and also, measuring energy consumption can offer a unique
of battery. Developing AI methods to predict and accommodate such overview, compared to time consumed. Last, there are some solutions
conditions is another potential research direction [34]. that can model the energy consumption of different algorithms [303],
as for example Alphabet’s DeepMind.
10.3. Routing
10.5. Spectrum management

In Section 8, we studied AI-enabled routing protocols, which use ML

Enabling high-rate, low-latency, and ultra-reliable wireless commu-
algorithms to predict network topology directly (e.g., predictive routing
nications in UAV networks is a necessity for future applications. Cur-
methods [32,105]) or indirectly (e.g., RL-based methods [242,243])
rently, UAVs use different communication protocols, including WiFi,
and use it for the route selection process. However, there still exist
LTE, LoRA, and 5G for A2 A and A2G communications. In recent
challenges to be addressed.
years, progress has been made in obtaining additional dedicated radio-
Most routing protocols consider a fully connected network, whereas frequency spectrum (5030–5091 MHz) for drone operations [279]. In
in reality, link breakages exist, causing routing protocols to fail [289]. addition to the usual ways of power management, interference control,
Also, node mobility in routing protocols is mostly developed for and spectrum-efficient networks, the use of different spectrum sensing,
tested in 2D spaces, whereas UAVs move in 3D spaces. The third spectrum sharing, and spectrum leasing is considered to extend the
research challenge is developing vision-based target tracking meth- service area of UAVs [101,102,304,305], especially in unexpected and
ods to predict network topology, noting that future UAVs will be harsh conditions such as disaster management.
equipped with Graphics/Tensor Processing Units (GPU/TPUs), capable In a different line of research, some models offer using fiber optic
of running deep learning methods for video processing. Developing communications [306], laser [307] and LiFi [308] to provide a faster
probabilistic and priority-based routing protocols to prioritize packets and more efficient way for transmitting large amounts of data over
with critical and confidential content is another potential research long distances to cover the increasing demand for bandwidth. These
direction. Finally, extending RL-based method to accommodate non- methods would alleviate the spectrum scarcity issue. However, more
linear motions and implementing light-weight routing protocols for research is expected to solve optical communications’ specific issues,
miniaturized UAVs are two important remaining challenges. including sensitivity to interference and adaptive antenna steering.

18
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Table 8
Remaining issues and future directions.
Open issues Problems Future directions
∙ Use ML, RL and DL techniques for different aspects of networking,
∙ Current networking designs have not yet fully utilized the
including sensing, scheduling, routing, spectrum sharing, path planning,
power of AI-based solutions.
∙ AI integration and resource allocation.
∙ The use of AI methods can add to the computation
∙ Study if AI improves performance and does not impact negatively
complexity and CPU power use in power-constrained UAVs.
network lifetime and power consumption aspects.
∙ Use ML methods to predict network topology change ahead of time
and avoid connectivity loss by using learning-based routing.
∙ Link failures bring limited network lifetime.
∙ Consider techniques that enhance link lifetime by adding link remaining
∙ Communication links between UAVs are extremely vulnera-
time or path lifetime into routing criteria.
∙ Connectivity ble.
∙ Resolve the connectivity loss by methods such as placement of new
∙ Connectivity is dependent on the interference caused by
UAVs and link re-establishment mechanisms.
objects and environmental factors.
∙ Integrate online path planning methods with networking algorithms to
realize topology control with minimal connectivity issues.
∙ Design a routing protocol that considers intermittent connectivity into
∙ Most routing protocols consider a fully connected network, the routing selection process, as well as 3-dimensional space.
whereas in reality, link breakages happen. ∙ Design a topology-predictive routing to accommodate non-linear mo-
∙ Node mobility in routing protocols is mostly developed for tions.
∙ Routing and tested in 2D spaces, whereas UAVs move in 3D spaces. ∙ Develop vision-based target tracking methods for GPU-powered UAVs
∙ Non-linear motions are not considered in current routing to predict network topology.
protocols. ∙ Develop a light-weight AI-powered routing appropriate for miniaturized
∙ Miniaturized UAVs have limited battery lifetime. UAVs.

∙ Develop energy-efficient networking to prolong mission time and extend

coverage area as a top priority.
∙ Develop new battery technologies, such as hydrogen fuel cells and
enhanced lithium-ion batteries.
∙ UAVs are highly constrained in payload.
∙ Use RF transmission for WPT.
∙ Battery designs are limited in energy optimization trends.
∙ Energy efficiency ∙ Consider mmWave communications, energy beamforming and place-
∙ Only a small portion of routing protocols considers energy
ment optimization of wireless charging stations.
or power as a routing criteria.
∙ Examine energy-aware routing future directions to incorporate energy
level in the decision-making criteria to extend path lifetime.
∙ Use energy-efficient ML and DL algorithms for networking, such as
energy-efficient convolutional neural networks.

∙ Spectrum unavailability can cause the loss of command and ∙ Spectrum sharing and spectrum leasing techniques using advanced AI-
control of the aircraft. based solutions.
∙ Spectrum management
∙ Spectrum remains vulnerable to unintentional or intentional ∙ Optimal communications including sensitivity to interference and
interference. adaptive antenna steering can optimize data acquisition objectives.

∙ UAVs are usually subject to different security attacks.

∙ Aerial monitoring systems may exchange imagery with ∙ Use hardware-driven security keys for UAV authentication and enable
people’s private information, which requires higher protection the non-repudiation feature.
levels. ∙ ML methods can be used to detect and eliminate jamming, or provide
∙ Security and user privacy
∙ Conventional PKI-based asymmetric security solutions are not an additional reference for positioning verification.
feasible due to the lack of central authority to issue digital ∙ Develop secure routing schemes that alleviate security issues while
signatures. finding optimal paths.
∙ Jamming attacks can also disrupt UAV missions.

∙ Develop certification standards and air traffic requirements for UAV

∙ Lack of or ambiguity of regulations and standards for UAV
operations that are universal.
∙ Operational regulations operations, characteristics, safety requirements, secrecy and
∙ Use AI software for regulation compliance to ensure safety of confiden-
privacy considerations, and allowed airspace.
tial information, risks mitigation and instant response to new regulatory
requirements.

10.6. Security and user privacy authentication and enabling the non-repudiation feature is recently pro-
posed as a potential future direction [313]. Jamming attacks can also
Developing secure and privacy-preserving networking methods is disrupt UAV missions, especially when they rely in GPS positioning.
another key challenge for UAV networks. UAVs are usually subject Using alternative localization methods [314], as well as, designing a
to different security attacks, including physical hijacking, jamming secure handover mechanism which is resistant to both passive and ac-
attacks, cyber-attacks, man-in-the-middle attack, intruding malicious tive attacks [315], can solve this issue. Also, machine learning methods
nodes, channel interception, and denial of service, especially when fly- can be used to detect and eliminate jamming, or provide an additional
ing over adversary territory [43]. Also, aerial monitoring systems may reference for positioning verification [316]. Another emerging research
exchange imagery with people’s private information, which requires trend is developing secure routing schemes that alleviate security issues
higher protection levels. while finding optimal paths [317].
A key challenge in structure-less UAV networks is that using con-
ventional PKI-based asymmetric security solutions is not feasible due 10.7. Operational regulations
to the lack of central authority to issue digital signatures. Therefore,
methods based on distributed certificate [309], key pre-distribution Another hindrance to the widespread use of drone technology is the
algorithms [310,311], and blockchain security [312] are under inves- lack of or ambiguity of sufficient regulations and standards for UAV
tigation. Also, the idea of using hardware-driven security keys for UAV operations, characteristics, safety requirements, secrecy and privacy

19
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

considerations, and allowed airspace. In the US, the FAA is responsible Acknowledgments
for developing certification standards and air traffic requirements for
drones. For instance, flying drones above class G airspace and in The work of Jacob Chakareski has been supported in part by the
autonomous modes require special permits from the FAA that can take NSF under awards CCF-1528030, ECCS-1711592, CNS-1836909, CNS-
a long time. Also, international coordination would help develop global 1821875, and CNS-1836909
regulations, noting that different territories follow different standards
and regulations. For instance, currently, there exist three different References
regions, including (i) region 1 that covers Europe, Africa, and parts
of the Middle East, (ii) region 2 that covers America, and (iii) region [1] E.N. Barmpounakis, E.I. Vlahogianni, J.C. Golias, Unmanned Aerial Aircraft
3 that covers Asia and the Pacific, which have different frequency Systems for transportation engineering: Current practice and future chal-
lenges, Int. J. Transp. Sci. Technol. 5 (3) (2016) 111–122, http://dx.doi.org/
bands for UAV operations. One more potential solution would be
10.1016/j.ijtst.2017.02.001, http://www.sciencedirect.com/science/article/pii/
using AI software for regulation compliance, which ensures increased S2046043016300533, Unmanned Aerial Vehicles and Remote Sensing.
safety of confidential information, risks mitigation and instant response [2] K. Kanistras, G. Martins, M.J. Rutherford, K.P. Valavanis, A survey of unmanned
to new regulatory requirements. This way, we could find AI-enabled aerial vehicles (UAVs) for traffic monitoring, in: 2013 International Conference
drones authorized for use in different regions, following the operational on Unmanned Aircraft Systems (ICUAS), 2013, pp. 221–234, http://dx.doi.org/
10.1109/ICUAS.2013.6564694.
regulations at each specific area.
[3] Z. Zaheer, A. Usmani, E. Khan, M.A. Qadeer, Aerial surveillance system using
UAV, in: 2016 Thirteenth International Conference on Wireless and Optical
10.8. Summary of challenges Communications Networks (WOCN), 2016, pp. 1–7, http://dx.doi.org/10.1109/
WOCN.2016.7759885.
We identified the need for (i) advanced AI methods to precisely [4] D. Bein, W. Bein, A. Karki, B.B. Madan, Optimizing border patrol operations
predict networking conditions and environmental factors, (ii) integra- using unmanned aerial vehicles, in: 2015 12th International Conference on
Information Technology - New Generations, 2015, pp. 479–484, http://dx.doi.
tive networking and topology control for extended connectivity, (iii)
org/10.1109/ITNG.2015.83.
vision-based tracking methods for GPU-powered UAVs, (iv) topology- [5] S. Waharte, N. Trigoni, Supporting Search and Rescue Operations with UAVs,
predictive routing to accommodate non-linear motions, (v) developing in: 2010 International Conference on Emerging Security Technologies, 2010,
light-weight AI-powered routing appropriate for miniaturized UAVs, pp. 142–147.
(vi) distributed structure-free extension of asymmetric security proto- [6] M. Erdelj, E. Natalizio, K.R. Chowdhury, I.F. Akyildiz, Help from the sky:
leveraging uavs for disaster management, IEEE Pervasive Comput. 16 (1) (2017)
cols based on key pre-distribution, hardware-driven keys, blockchain,
24–32.
and distributed certificate methods, (vii) ML methods to recognize and [7] M. Messous, S. Senouci, H. Sedjelmaci, Network connectivity and area coverage
combat jamming attacks, (viii) energy-efficient low-power and low- for UAV fleet mobility model with energy constraint, in: 2016 IEEE Wirel.
complexity networking, (ix) AI-based spectrum sharing and leasing Commun. and Networking Conference, 2016, pp. 1–6.
policies, security-aware mission planning, and (xi) universal regulation [8] J. Chakareski, S. Naqvi, N. Mastronarde, J. Xu, F. Afghah, A. Razi, An energy
efficient framework for uav-assisted millimeter wave 5g heterogeneous cellular
and guidelines for UAV operation, as top key issues that are worthy of
networks, IEEE Trans. Green Commun. Netw. 3 (1) (2019) 37–44, http://dx.
investigation by the research community. A more detailed summary of doi.org/10.1109/TGCN.2019.2892141.
the remaining challenges and future trends is included in Table 8. [9] M.R. Brust, B.M. Strimbu, A networked swarm model for UAV deployment
in the assessment of forest environments, in: 2015 IEEE Tenth International
11. Conclusion Conference on Intelligent Sensors, Sensor Networks and Information Processing
(ISSNIP), 2015, pp. 1–6, http://dx.doi.org/10.1109/ISSNIP.2015.7106967.
The dominant trend of UAV systems is departing from single-drone [10] J. Chakareski, Drone networks for virtual human teleportation, in: Proc. ACM
Workshop on Micro Aerial Vehicle Networks, Systems, and Applications, Niagra
systems to networked autonomous drones to accomplish complicated Falls, NY, USA, 2017, pp. 21–26.
tasks at lower cost and time. The idea of using AI-based network- [11] J. Chakareski, Uav-iot for next generation virtual reality, IEEE Trans. Image
ing and control protocols to accommodate dynamic situations has Process. 28 (12) (2019) 5977–5990.
recently gained a lot of attention, thanks to the advances in AI and [12] M. Khan, J. Chakareski, S. Gupta, RF-FSO Dual-Path UAV Network for High
deep learning methods powered by fast and affordable computational Fidelity Multi-Viewpoint Scalable 360-degree Video Streaming, in: Proc. IEEE
Int’l Workshop on Multimedia Signal Processing, Tampere, Finland, 2020, pp.
platforms. Here, we focused on the most recent AI-based methods
1–6.
that aim to model and predict network topology to facilitate more [13] LOCUST: Autonomous, Swarming UAVs fly into the future, ONR, 2015,
efficient information flow for aerial networks. This important class of http://www.onr.navy.mil/Media-Center/Press-Releases/2015/LOCUST-low-
routing protocols is overlooked or not received the deserved emphasis cost-UAV-swarm-ONR.aspx.
in previous survey papers. [14] S. Siebert, J. Teizer, Mobile 3d mapping for surveying earthwork using an
unmanned aerial vehicle (uav), 2013.
In this paper, we reviewed AI-based routing protocols designed
[15] J. Plaza, Using drones to monitor volcano activity and save lives. Commercial
for UAV networks, highlighting the benefits and costs of each type, UAV News, 2018, https://www.commercialuavnews.com/public-safety/using-
along with available testing and implementation tools, relations to drones-to-monitor-volcano-activity-and-save-lives.
mobility models and networking protocols, and connection to UAV [16] Brain-controlled drones are here: What’s coming in the next five years?
swarming. These methods include the direct use of ML methods for ScienceDaily, 2017, https://www.sciencedaily.com/releases/2017/09/
170929230433.htm.
topology prediction, as well as learning-by-experience approaches. We
[17] Electrical And computer engineering, CASA adds drone detection to its early
also reported substantial improvements of the most recent methods in warning monitoring of severe weather | Electrical and Computer Engineering |
terms of connectivity control, successful packet delivery rate, transmis- UMass Amherst, 2018, https://ece.umass.edu/news/casa-adds-drone-detection-
sion delay, and throughput for AI-based routing protocols, compared its-early-warning-monitoring-severe-weather.
to conventional methods. We also provided our perspective on the [18] Y.A. Pederi, H.S. Cheporniuk, Unmanned Aerial Vehicles and new technological
methods of monitoring and crop protection in precision agriculture, in: 2015
future trends for AI-based networking, which mainly center around
IEEE International Conference Actual Problems of Unmanned Aerial Vehicles
developing method for more accurate environment perception, vision- Developments (APUAVD), 2015, pp. 298–301.
based tracking, light-weight DL algorithms appropriate for on-board [19] A. Cherubini, A. Papini, R. Vertechy, M. Fontana, Airborne wind energy
processing, AI-based spectrum sharing, and security-aware networking. systems: A review of the technologies, Renew. Sustain. Energy Rev. 51
(2015) 1461–1476, http://dx.doi.org/10.1016/j.rser.2015.07.053, http://www.
Declaration of competing interest sciencedirect.com/science/article/pii/S1364032115007005.
[20] A.A. Paranjape, S. Chung, K. Kim, D.H. Shim, Robotic herding of a flock of birds
using an unmanned aerial vehicle, IEEE Trans. Robot. 34 (4) (2018) 901–915,
The authors declare that they have no known competing finan- http://dx.doi.org/10.1109/TRO.2018.2853610.
cial interests or personal relationships that could have appeared to [21] Amazon Prime Air, Amazon, https://www.amazon.com/Amazon-Prime-Air/b?
influence the work reported in this paper. ie=UTF8&node=8037720011.

20
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

[22] UPS Flight Forward™ Drone Delivery, UPS, https://www.ups.com/us/en/ [49] B. Knight, A guide to military drones, DW.COM, 2017, https://www.dw.com/
services/shipping-services/flight-forward-drones.page. en/a-guide-to-military-drones/a-39441185.
[23] L. Gupta, R. Jain, G. Vaszkun, Survey of important issues in UAV [50] A. Sharma, Everything you need to know about industrial grade drones,
communication networks, IEEE Commun. Surv. Tutor. 18 (2) (2016) 2020, https://jungleworks.com/everything-you-need-to-know-about-industrial-
1123–1152. grade-drones/.
[24] Unmanned Aerial Vehicle (UAV) Market. Market Research Firm, [51] Flyability, Commercial Drones: Industries that Use Drones, Deliverables, and
https://www.marketsandmarkets.com/Market-Reports/unmanned-aerial- Our List of the Top Models on the Market, Flyability, https://www.flyability.
vehicles-uav-market-662.html. com/commercial-drones.
[25] K.-I. Kim, K.-H. Kim, M. Imran, P. Khan, E. Tovar, F. Ali, UAV-Enabled [52] FAA, Recreational flyers & modeler community-based organizations, 2020,
healthcare architecture: Issues and challenges, Future Gener. Comput. Syst. https://www.faa.gov/uas/recreational_fliers/.
(2019) http://dx.doi.org/10.1016/j.future.2019.01.028. [53] R. Yin, W. Li, Z.-q. Wang, X.-x. Xu, The application of artificial intelligence
[26] M. Chen, H. Wang, S. Mehrotra, V.C.M. Leung, I. Humar, Intelligent networks technology in UAV, in: 2020 5th International Conference on Information
assisted by cognitive computing and machine learning, IEEE Netw. 33 (3) Science, Computer Technology and Transportation (ISCTT), 2020, pp. 238–241,
(2019) 6–8, http://dx.doi.org/10.1109/MNET.2019.8726065. http://dx.doi.org/10.1109/ISCTT51595.2020.00048.
[27] P. Bithas, E. Michailidis, N. Nomikos, D. Vouyioukas, A. Kanatas, A survey on [54] R. Hof, AI-On-A-Chip Soon will make phones, drones and more a lot
machine-learning techniques for UAV-Based communications, Sensors 19 (2019) smarter, Forbes Magazine, 2016, https://www.forbes.com/sites/roberthof/
http://dx.doi.org/10.3390/s19235170. 2016/05/07/ai-on-a-chip-soon-will-make-phones-drones-and-more-a-lot-
[28] B. .S, Study of ad hoc networks with reference to MANET, VANET, FANET, Int. smarter/?sh=75b2d2ba7eef.
J. of Adv. Res. Comput. Sci. Softw. Eng. 7 (2017) 390, http://dx.doi.org/10. [55] T. Hwang, Computational power and the social impact of artificial intelligence,
23956/ijarcsse/V7I7/0159. 2018, arXiv:1803.08971.
[29] H. Lashari, H.M. Ali, A. Laghari, Uav communication networks issues: A review, [56] C. Dilmegani, AI Chips in 2021: Guide to cost-efficient AI training & infer-
Arch. Comput. Methods Eng. (2020) http://dx.doi.org/10.1007/s11831-020- ence, AIMultiple, 2021, https://research.aimultiple.com/ai-chip/#what-are-its-
09418-0. components.
[30] T. Izydorczyk, G. Berardinelli, P. Mogensen, M.M. Ginard, J. Wigard, I.Z. [57] K. Freund, AI hardware: Harder than it looks, Forbes Magazine, 2019,
Kovács, Achieving high UAV uplink throughput by using beamforming on https://www.forbes.com/sites/moorinsights/2019/10/07/ai-hardware-harder-
board, IEEE Access 8 (2020) 82528–82538, http://dx.doi.org/10.1109/ACCESS. than-it-looks/?sh=23dc41e5471f.
2020.2991658. [58] Freund, Intel shows off its ai chips and chops, forbes, 2018, https:
[31] W. Yuan, C. Liu, F. Liu, S. Li, D.W.K. Ng, Learning-based predictive beamform- //www.forbes.com/sites/moorinsights/2018/06/01/intel-shows-off-its-ai-
ing for UAV communications with jittering, IEEE Wirel. Commun. Lett. 9 (11) chips-and-chops/?sh=272064ad6643.
(2020) 1970–1974, http://dx.doi.org/10.1109/LWC.2020.3009951. [59] I. Insider, Drone market outlook in 2021: industry growth trends,
[32] M. Khaledi, A. Rovira-Sugranes, F. Afghah, A. Razi, On greedy routing in
market stats and forecast, Business Insider, 2021, https://www.
dynamic UAV networks, in: IEEE International Conference on Sensing, Com-
businessinsider.com/drone-industry-analysis-market-trends-growth-forecasts#:
munication and Networking (SECON 2018), 2018, http://dx.doi.org/10.1109/
~:text=InsiderIntelligencedefinesenterprisedrones,annualgrowthrate(CAGR).
WiSEE.2017.8124890.
[60] L. Schroth, Drone manufacturer market shares: DJI leads the way, Drone In-
[33] O. Sami Oubbati, M. Atiquzzaman, T. Ahamed Ahanger, A. Ibrahim, Softwariza-
dustry Insights, 2021, https://droneii.com/drone-manufacturer-market-shares-
tion of UAV networks: A survey of applications and future trends, IEEE Access
dji-leads-the-way-in-the-us.
8 (2020) 98073–98125.
[61] PrecisionHawk, Geospatial Data Analytics for the Enterprise https://www.
[34] D. Shumeye Lakew, U. Sa’ad, N. Dao, W. Na, S. Cho, Routing in flying ad hoc
precisionhawk.com/.
networks: A comprehensive survey, IEEE Commun. Surv. Tutor. 22 (2) (2020)
[62] FortemTechnologies, DroneHunter: Net Gun Drone Capture: Products,
1071–1120.
Fortem Technologies, 2021, https://fortemtech.com/products/
[35] Q. Sang, H. Wu, L. Xing, P. Xie, Review and comparison of emerging routing
dronehunter/?gclid=Cj0KCQjwl9GCBhDvARIsAFunhskjqqROz4pMBmxT_
protocols in flying Ad Hoc networks, Symmetry 12 (2020) 971, http://dx.doi.
bxGxRK5KMyievKr4lH3kejZtH_gLS1qQt0X68kaAsEfEALw_wcB.
org/10.3390/sym12060971.
[63] UAVCoach, Top 100 drone companies to watch in 2020, UAV Coach, 2021,
[36] B. Alzahrani, O.S. Oubbati, A. Barnawi, M. Atiquzzaman, D. Alghazzawi, UAV
https://uavcoach.com/drone-companies/.
Assistance paradigm: State-of-the-art in applications and challenges, J. Netw.
[64] C.B. Insights, How drones will impact society: From fighting war to forecasting
Comput. Appl. (2020) http://dx.doi.org/10.1016/j.jnca.2020.102706.
weather, UAVs change everything, CB Insights Research, 2020, https://www.
[37] R.A. Nazib, S. Moh, Routing protocols for unmanned aerial vehicle-aided
cbinsights.com/research/drone-impact-society-uav/.
vehicular Ad Hoc networks: A survey, IEEE Access 8 (2020) 77535–77560.
[65] G.D. Maayan, How do ai-based drones work? Medium, Heartbeat, 2020, https:
[38] M.Y. Arafat, S. Moh, Routing protocols for unmanned aerial vehicle networks:
//heartbeat.fritz.ai/how-ai-based-drones-work-a94f20e62695.
A survey, IEEE Access 7 (2019) 99694–99720, http://dx.doi.org/10.1109/
ACCESS.2019.2930813. [66] M.-A. Lahmeri, M.A. Kishk, M.-S. Alouini, Artificial intelligence for UAV-enabled
[39] O.S. Oubbati, M. Atiquzzaman, P. Lorenz, M.H. Tareque, M.S. Hossain, Rout- wireless networks: A survey, IEEE Open J. Commun. Soc. 2 (2021) 1015–1040,
ing in flying Ad Hoc networks: Survey, constraints, and future challenge http://dx.doi.org/10.1109/OJCOMS.2021.3075201.
perspectives, IEEE Access 7 (2019) 81057–81105. [67] Q. Zhang, M. Mozaffari, W. Saad, M. Bennis, M. Debbah, Machine learning
[40] M. Khan, K.-L. Yau, R. Md. Noor, M. Imran, Routing schemes in FANETs: for predictive on-demand deployment of Uavs for wireless communications, in:
A survey, Sensors (Basel, Switzerland) 20 (2019) http://dx.doi.org/10.3390/ 2018 IEEE Global Communications Conference (GLOBECOM), 2018, pp. 1–6,
s20010038. http://dx.doi.org/10.1109/GLOCOM.2018.8647209.
[41] J. Jiang, G. Han, Routing protocols for unmanned aerial vehicles, IEEE [68] J. Chen, U. Yatnalli, D. Gesbert, Learning radio maps for UAV-aided wireless
Commun. Mag. 56 (1) (2018) 58–63. networks: A segmented regression approach, in: 2017 IEEE International Con-
[42] A. Awang, K. Husain, N. Kamel, S. Aïssa, Routing in vehicular Ad-hoc networks: ference on Communications (ICC), 2017, pp. 1–6, http://dx.doi.org/10.1109/
A survey on single- and cross-layer design techniques, and perspectives, IEEE ICC.2017.7997333.
Access 5 (2017) 9497–9517. [69] J. Kim, C. Park, J. Ahn, Y. Ko, J. Park, J.C. Gallagher, Real-time UAV sound
[43] M. Jean aime, M.-S. Mahmoud, N. Larrieu, Survey on UAANET routing protocols detection and analysis system, in: 2017 IEEE Sensors Applications Symposium
and network security challenges, Ad-Hoc Sens. Wirel. Netw. 37 (2017). (SAS), 2017, pp. 1–5, http://dx.doi.org/10.1109/SAS.2017.7894058.
[44] O.S. Oubbati, A. Lakas, F. Zhou, M. Günes, M.B. Yagoubi, A survey on position- [70] Y. Zhang, J. Wen, G. Yang, Z. He, X. Luo, Air-to-air path loss prediction based
based routing protocols for flying Ad hoc networks (FANETs), Veh. Commun. on machine learning methods in urban environments, Wirel. Commun. Mob.
10 (2017) 29–56. Comput. 2018 (2018) 1–9, http://dx.doi.org/10.1155/2018/8489326.
[45] C. Suthaputchakun, Z. Sun, Routing protocol in intervehicle communication [71] W. Xia, S. Rangan, M. Mezzavillla, A. Lozano, G. Geraci, V. Semkin, G.
systems: A survey, IEEE Commun. Mag. 49 (2011) 150–156, http://dx.doi.org/ Loianno, Generative neural network channel modeling for millimeter-wave UAV
10.1109/MCOM.2011.6094020. communication, 2020, arXiv:2012.09133.
[46] J. Brown, Types of military drones: The best technology available today, My [72] M. Chen, W. Saad, C. Yin, Deep learning for 360◦ content transmission in
Drone Lab, 2020, https://www.mydronelab.com/blog/types-of-military-drones. UAV-enabled virtual reality, in: ICC 2019 - 2019 IEEE International Conference
html. on Communications (ICC), 2019, pp. 1–6, http://dx.doi.org/10.1109/ICC.2019.
[47] Milsource, 4 Military sensor technologies that drones are transporting to the 8761472.
commercial market, MilSource - Military Ethernet Switch and Router Prod- [73] S. Wang, J. Chen, Z. Zhang, G. Wang, Y. Tan, Y. Zheng, Construction of a virtual
ucts, 2019, https://militaryethernet.com/4-military-sensor-technologies-drones- reality platform for UAV deep learning, in: 2017 Chinese Automation Congress
transporting-commercial-market/. (CAC), 2017, pp. 3912–3916, http://dx.doi.org/10.1109/CAC.2017.8243463.
[48] J.R. Wilson, The future of artificial intelligence and quantum computing, [74] K. Mukadam, A. Sinh, R. Karani, Detection of landing areas for unmanned
Military & Aerospace Electronics, 2020, https://www.militaryaerospace.com/ aerial vehicles, in: 2016 International Conference on Computing Communication
computers/article/14182330/future-of-artificial-intelligence-and-quantum- Control and automation (ICCUBEA), 2016, pp. 1–5, http://dx.doi.org/10.1109/
computing. ICCUBEA.2016.7860044.

21
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

[75] N. Imanberdiyev, C. Fu, E. Kayacan, I.-M. Chen, Autonomous navigation [101] A. Shamsoshoara, F. Afghah, A. Razi, S. Mousavi, J. Ashdown, K. Turk,
of UAV by using real-time model-based reinforcement learning, in: 2016 An autonomous spectrum management scheme for unmanned aerial vehicle
14th International Conference on Control, Automation, Robotics and Vision networks in disaster relief operations, IEEE Access 8 (2020) 58064–58079.
(ICARCV), 2016, pp. 1–6, http://dx.doi.org/10.1109/ICARCV.2016.7838739. [102] A. Shamsoshoara, M. Khaledi, F. Afghah, A. Razi, J. Ashdown, Distributed
[76] X. Liu, Y. Liu, Y. Chen, Reinforcement learning in multiple-UAV networks: cooperative spectrum sharing in UAV networks using multi-agent reinforcement
Deployment and movement design, 2019, arXiv:1904.05242. learning, in: 2019 16th IEEE Annual Consumer Communications & Networking
[77] G. Faraci, A. Raciti, S.A. Rizzo, G. Schembra, Green wireless power Conference (CCNC), IEEE, 2019, pp. 1–6.
transfer system for a drone fleet managed by reinforcement learning in [103] A. Jalali, L. Schiff, Beam forming and pointing in a network of unmanned aerial
smart industry, Appl. Energy 259 (2020) 114204, http://dx.doi.org/10. vehicles (UAVs) for broadband access, Google Patents, US Patent 9,712,228,
1016/j.apenergy.2019.114204, https://www.sciencedirect.com/science/article/ 2017.
pii/S0306261919318914. [104] A. Muralidharan, Y. Mostofi, Energy optimal distributed beamforming using
[78] N.I. Mowla, N.H. Tran, I. Doh, K. Chae, Federated learning-based cognitive unmanned vehicles, IEEE Trans. Control Netw. Syst. 5 (4) (2017) 1529–1540.
detection of jamming attack in flying Ad-Hoc network, IEEE Access 8 (2020) [105] A. Rovira-Sugranes, A. Razi, Predictive routing for dynamic UAV networks,
4338–4350, http://dx.doi.org/10.1109/ACCESS.2019.2962873. in: 2017 IEEE International Conference on Wireless for Space and Extreme
[79] Y. Liu, J. Nie, X. Li, S.H. Ahmed, W.Y.B. Lim, C. Miao, Federated learning in Environments (WiSEE), 2017, pp. 43–47, http://dx.doi.org/10.1109/WiSEE.
the sky: Aerial-Ground air quality sensing framework with UAV swarms, IEEE 2017.8124890.
Internet Things J. (2020) http://dx.doi.org/10.1109/JIOT.2020.3021006, 1–1. [106] A. Rovira-Sugranes, F. Afghah, A. Razi, Optimized compression policy for
[80] A. Razi, C. Wang, F. Almaraghi, Q. Huang, Y. Zhang, H. Lu, A. Rovira-Sugranes, flying Ad hoc networks, in: 2019 16th IEEE Annual Consumer Communications
Predictive routing for wireless networks: Robotics-based test and evaluation Networking Conference (CCNC), 2019, pp. 1–2, http://dx.doi.org/10.1109/
platform, in: 2018 IEEE 8th Annual Computing and Communication Workshop CCNC.2019.8651761.
and Conference (CCWC), 2018, pp. 993–999. [107] A. Rovira-Sugranes, A. Razi, Optimizing the age of information for blockchain
[81] W. Shi, J. Cao, Q. Zhang, Y. Li, L. Xu, Edge computing: Vision and challenges, technology with applications to IoT sensors, IEEE Commun. Lett. 24 (1) (2020)
IEEE Internet Things J. 3 (5) (2016) 637–646, http://dx.doi.org/10.1109/JIOT. 183–187, http://dx.doi.org/10.1109/LCOMM.2019.2949557.
2016.2579198. [108] M.C. Jeffrey, S. Subramanian, C. Yan, J. Emer, D. Sanchez, A scalable archi-
[82] A.V. Dastjerdi, R. Buyya, Fog computing: Helping the internet of things realize tecture for ordered parallelism, in: 2015 48th Annual IEEE/ACM International
its potential, Computer 49 (8) (2016) 112–116, http://dx.doi.org/10.1109/MC. Symposium on Microarchitecture (MICRO), 2015, pp. 228–241, http://dx.doi.
2016.245. org/10.1145/2830772.2830777.
[83] G. Baldini, S. Karanasios, D. Allen, F. Vergari, Survey of wireless communication [109] Z. Xiaoning, Analysis of military application of UAV swarm technology, in:
technologies for public safety, IEEE Commun. Surv. Tutor. 16 (2) (2014) 2020 3rd International Conference on Unmanned Systems (ICUS), 2020, pp.
619–641, http://dx.doi.org/10.1109/SURV.2013.082713.00034. 1200–1204, http://dx.doi.org/10.1109/ICUS50048.2020.9274974.
[110] J. Peng, Z. Zhang, Q. Wu, B. Zhang, Anti-jamming communications in
[84] V. Sharma, I. You, G. Pau, M. Collotta, J. Deok Lim, J. Kim, LoRaWAN-Based
UAV swarms: A reinforcement learning approach, IEEE Access 7 (2019)
energy-efficient surveillance by drones for intelligent transportation systems,
180532–180543, http://dx.doi.org/10.1109/ACCESS.2019.2958328.
Energies 11 (2018) http://dx.doi.org/10.3390/en11030573.
[111] G. Bansal, B. Sikdar, S-MAPS: Scalable mutual authentication protocol for
[85] V. Delafontaine, F. Schiano, G. Cocco, A. Rusu, D. Floreano, Drone-aided
dynamic UAV swarms, IEEE Trans. Veh. Technol. 70 (11) (2021) 12088–12100,
localization in LoRa IoT networks, in: 2020 IEEE International Conference
http://dx.doi.org/10.1109/TVT.2021.3116163.
on Robotics and Automation (ICRA), 2020, pp. 286–292, http://dx.doi.org/10.
[112] Bansal and Sikdar, Location aware clustering: Scalable authentication protocol
1109/ICRA40945.2020.9196869.
for UAV swarms, IEEE Netw. Lett. 3 (4) (2021) 177–180, http://dx.doi.org/10.
[86] O. Saraereh, A. Alsaraira, I. Khan, P. Uthansakul, Performance evaluation of
1109/LNET.2021.3116073.
UAV-enabled LoRa networks for disaster management applications, Sensors 20
[113] L. Ruetten, P.A. Regis, D. Feil-Seifer, S. Sengupta, Area-optimized UAV swarm
(2020) 1–18, http://dx.doi.org/10.3390/s20082396.
network for search and rescue operations, in: 2020 10th Annual Computing
[87] S. Chandrasekharan, K. Gomez, A. Al-Hourani, S. Kandeepan, T. Rasheed, L.
and Communication Workshop and Conference (CCWC), 2020, pp. 0613–0618,
Goratti, L. Reynaud, D. Grace, I. Bucaille, T. Wirth, S. Allsopp, Designing and
http://dx.doi.org/10.1109/CCWC47524.2020.9031197.
implementing future aerial communication networks, IEEE Commun. Mag. 54
[114] M. Campion, P. Ranganathan, S. Faruque, A review and future directions of UAV
(5) (2016) 26–34, http://dx.doi.org/10.1109/MCOM.2016.7470932.
swarm communication architectures, in: 2018 IEEE International Conference on
[88] Different Wi-Fi protocols and data rates, Intel, https://www.intel.com/content/
Electro/Information Technology (EIT), 2018, pp. 0903–0908, http://dx.doi.org/
www/us/en/support/articles/000005725/network-and-io/wireless.html.
10.1109/EIT.2018.8500274.
[89] xFold SPY, xFoldRig, http://www.xfoldrig.com/xfold-spy/.
[115] A. Tahir, J. Böling, M.-H. Haghbayan, H.T. Toivonen, J. Plosila, Swarms
[90] X. Lin, V. Yajnanarayana, S.D. Muruganathan, S. Gao, H. Asplund, H. Maatta-
of unmanned aerial vehicles — A survey, J. Ind. Inf. Integr. 16
nen, M.B. A, S. Euler, Y.E. Wang, The sky is not the limit: LTE for unmanned
(2019) 100106, http://dx.doi.org/10.1016/j.jii.2019.100106, https://www.
aerial vehicles, CoRR, abs/1707.07534, 2017, arXiv:1707.07534.
sciencedirect.com/science/article/pii/S2452414X18300086.
[91] Y. Zeng, J. Lyu, R. Zhang, Cellular-connected UAV: Potentials, challenges and [116] X. Chen, J. Tang, S. Lao, Review of unmanned aerial vehicle swarm communica-
promising technologies, CoRR, abs/1804.02217, 2018, arXiv:1804.02217. tion architectures and routing protocols, Appl. Sci. 10 (10) (2020) http://dx.doi.
[92] T. Lagkas, V. Argyriou, S. Bibi, P. Sarigiannidis, UAV IoT Framework views org/10.3390/app10103661, https://www.mdpi.com/2076-3417/10/10/3661.
and challenges: towards protecting drones as ‘‘things’’, Sensors 18 (11) (2018) [117] I. Bekmezci, O.K. Sahingoz, S. Temel, Flying Ad-Hoc networks (FANETs):
4015. A survey, Ad Hoc Netw. 11 (3) (2013) 1254–1270, http://dx.doi.org/10.
[93] S. Sekander, H. Tabassum, E. Hossain, Multi-tier drone architecture for 5G/B5G 1016/j.adhoc.2012.12.004, http://www.sciencedirect.com/science/article/pii/
cellular networks: Challenges, trends, and prospects, IEEE Commun. Mag. 56 S1570870512002193.
(3) (2018) 96–103. [118] A. Sivakumar, C.-Y. Tan, UAV swarm coordination using cooperative control for
[94] S.A.R. Naqvi, S.A. Hassan, H. Pervaiz, Q. Ni, Drone-aided communication as a establishing a wireless communications backbone, in: 9th International Joint
key enabler for 5G and resilient public safety networks, IEEE Commun. Mag. Conference on Autonomous Agents and Multiagent Systems 2010, 2010, pp.
56 (1) (2018) 36–42. 1157–1164.
[95] M.Y. Selim, A.E. Kamal, Post-disaster 4G/5G network rehabilitation us- [119] D.F. Vránics, M. Palik, B. Zs, Electronic administration of unmanned aviation
ing drones: Solving battery and backhaul issues, in: 2018 IEEE Globecom with public key infrastructure PKI), Secur. Future 3 (4) (2019) 152–155.
Workshops (GC Wkshps), IEEE, 2018, pp. 1–6. [120] O.K. Sahingoz, Networking models in flying ad-hoc networks (fanets): concepts
[96] W. Saad, M. Bennis, M. Chen, A vision of 6G wireless systems: Applications, and challenges, J. Intell. Robot. Syst. 74 (1-2) (2013) 513–527, http://dx.doi.
trends, technologies, and open research problems, IEEE Netw. 34 (3) (2020) org/10.1007/s10846-013-9959-7.
134–142, http://dx.doi.org/10.1109/MNET.001.1900287. [121] Y. Zhou, J. Li, L. Lamont, C. Rabbath, Modeling of packet dropout for UAV
[97] M. Mozaffari, X. Lin, S. Hayes, Towards 6G with connected sky: UAVs and wireless communications, in: 2012 International Conference on Computing,
beyond, CoRR, abs/2103.01143, 2021, arXiv:2103.01143. Networking and Communications (ICNC), 2012, pp. 677–682, http://dx.doi.org/
[98] S. Aggarwal, N. Kumar, S. Tanwar, Blockchain-envisioned UAV communication 10.1109/ICCNC.2012.61675080.
using 6g networks: open issues, use cases, and future directions, IEEE Inter- [122] H. Shariatmadari, R. Ratasuk, S. Iraji, A. Laya, T. Taleb, R. Jäntti, A. Ghosh,
net Things J. 8 (7) (2021) 5416–5441, http://dx.doi.org/10.1109/JIOT.2020. Machine-type communications: current status and future perspectives toward
3020819. 5G systems, IEEE Commun. Mag. 53 (9) (2015) 10–17, http://dx.doi.org/10.
[99] U. Challita, A. Ferdowsi, M. Chen, W. Saad, Machine learning for wireless 1109/MCOM.2015.7263367.
connectivity and security of cellular-connected UAVs, IEEE Wirel. Commun. [123] F. Boccardi, R.W. Heath, A. Lozano, T.L. Marzetta, P. Popovski, Five disruptive
26 (1) (2019) 28–35, http://dx.doi.org/10.1109/MWC.2018.1800155. technology directions for 5G, IEEE Commun. Mag. 52 (2) (2014) 74–80, http:
[100] J. Park, Y. Kim, J. Seok, Prediction of information propagation in a drone //dx.doi.org/10.1109/MCOM.2014.6736746.
network by using machine learning, in: 2016 International Conference on [124] M. Agiwal, A. Roy, N. Saxena, Next generation 5G wireless networks: A
Information and Communication Technology Convergence (ICTC), 2016, pp. comprehensive survey, IEEE Commun. Surv. Tutor. 18 (3) (2016) 1617–1655,
147–149, http://dx.doi.org/10.1109/ICTC.2016.7763456. http://dx.doi.org/10.1109/COMST.2016.2532458.

22
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

[125] P. Demestichas, A. Georgakopoulos, D. Karvounas, K. Tsagkaris, V. Stavroulaki, [151] S. Atev, O. Masoud, N. Papanikolopoulos, Learning traffic patterns at inter-
J. Lu, C. Xiong, J. Yao, 5G On the Horizon: Key challenges for the radio-access sections by spectral clustering of motion trajectories, in: Intelligent Robots
network, IEEE Veh. Technol. Mag. 8 (3) (2013) 47–53, http://dx.doi.org/10. and Systems, 2006 IEEE/RSJ International Conference on, IEEE, 2006, pp.
1109/MVT.2013.2269187. 4851–4856.
[126] G. Beni, J. Wang, Swarm intelligence in cellular robotic systems, in: P. Dario, [152] J.-G. Lee, J. Han, X. Li, A unifying framework of mining trajectory patterns
G. Sandini, P. Aebischer (Eds.), Robots and Biological Systems: Towards a New of various temporal tightness, IEEE Trans. Knowl. Data Eng. 27 (6) (2015)
Bionics?, Springer Berlin Heidelberg, Berlin, Heidelberg, 1993, pp. 703–712. 1478–1490.
[127] S. Awasthi, B. Balusamy, V. Porkodi, Artificial intelligence supervised swarm [153] C. Larsen, M. Zawodniok, S. Jagannathan, Route aware predictive congestion
uavs for reconnaissance, in: U. Batra, N.R. Roy, B. Panda (Eds.), Data Science control protocol for wireless sensor networks, in: Intelligent Control, 2007. ISIC
and Analytics, Springer Singapore, Singapore, 2020, pp. 375–388. 2007. IEEE 22nd International Symposium on, IEEE, 2007, pp. 13–18.
[128] J. Johnson, Artificial intelligence, drone swarming and escalation risks in future [154] H. Tan, Y. Wu, B. Shen, P.J. Jin, B. Ran, Short-term traffic prediction based
warfare, RUSI J. 165 (2) (2020) 26–36, http://dx.doi.org/10.1080/03071847. on dynamic tensor completion, IEEE Trans. Intell. Transp. Syst. 17 (8) (2016)
2020.1752026. 2123–2133.
[129] J. Kusyk, M.U. Uyar, K. Ma, J.J. Wu, W. Ruan, D.K. Guha, G. Bertoli, J. [155] P.P. Choi, M. Hebert, Learning and predicting moving object trajectory: a
Boksiner, AI Based flight control for autonomous UAV swarms, in: 2018 piecewise trajectory segment approach, Robotics Inst. (2006) 337.
International Conference on Computational Science and Computational Intel- [156] G. Aoude, J. Joseph, N. Roy, J. How, Mobile agent trajectory prediction using
ligence (CSCI), 2018, pp. 1155–1160, http://dx.doi.org/10.1109/CSCI46756. Bayesian nonparametric reachability trees, in: Infotech@ Aerospace 2011, 2011,
2018.00223. p. 1512.
[130] J. Kusyk, M.U. Uyar, K. Ma, E. Samoylov, R. Valdez, J. Plishka, S.E. Hoque, [157] A. Razi, Bayesian signal recovery under measurement matrix uncertainty:
G. Bertoli, J. Boksiner, Artificial intelligence and game theory controlled Performance analysis, IEEE Access 7 (2019) 102356–102365.
autonomous UAV swarms, Evol. Intell. (2020) 1–18. [158] G. Lee, R. Mallipeddi, M. Lee, Identification of moving vehicle trajectory
[131] T. Camp, J. Boleng, V. Davies, A survey of mobility models for ad hoc using manifold learning, in: International Conference on Neural Information
network research, Wireless communications and mobile computing 2 (5) (2002) Processing, Springer, 2012, pp. 188–195.
483–502. [159] M.S. Grewal, Kalman filtering, Springer, 2011.
[132] J. Broch, D.A. Maltz, D.B. Johnson, Y.-C. Hu, J. Jetcheva, A performance com- [160] A. Razi, Optimal measurement policy for linear measurement systems with
parison of multi-hop wireless ad hoc network routing protocols, in: Proceedings applications to UAV network topology prediction, IEEE Trans. Veh. Technol.
of the 4th annual ACM/IEEE international conference on Mobile computing and 69 (2) (2019) 1970–1981.
networking, ACM, 1998, pp. 85–97. [161] K. Lee, S. Hong, S.J. Kim, I. Rhee, S. Chong, Slaw: a new mobility model for
[133] W. Wang, X. Guan, B. Wang, Y. Wang, A novel mobility model based on human walks, in: INFOCOM 2009, IEEE, IEEE, 2009, pp. 855–863.
[162] Z.X. Ming, Z. Yue, Y. Fan, A. Vasilakos, Interference-based topology control
semi-random circular movement in mobile ad hoc networks, Inf. Sci. 180
algorithm for delay-constrained mobile ad hoc networks, in mobile computing,
(3) (2010) 399–413, http://dx.doi.org/10.1016/j.ins.2009.10.001, http://www.
IEEE Trans. 14 (4) (2015) 742–754.
sciencedirect.com/science/article/pii/S0020025509004265.
[163] T.-E. Lu, K.-T. Feng, et al., Predictive mobility and location-aware routing
[134] J. Yoon, M. Liu, B. Noble, Sound mobility models, in: Proceedings of the 9th
protocol in mobile ad hoc networks, in: GLOBECOM’05: IEEE Global Telecom-
annual international conference on Mobile computing and networking, ACM,
munications Conference, Vols 1-6: DISCOVERY PAST AND FUTURE, 2005, pp.
2003, pp. 205–216.
899–903.
[135] M.C. Gonzalez, C.A. Hidalgo, A.-L. Barabasi, Understanding individual human
[164] R. Groenevelt, E. Altman, P. Nain, Relaying in mobile ad hoc networks: the
mobility patterns, Nature 453 (7196) (2008) 779–782.
brownian motion mobility model, Wirel. Netw. 12 (5) (2006) 561–571.
[136] I. Rhee, M. Shin, S. Hong, K. Lee, S.J. Kim, S. Chong, On the levy-walk nature of
[165] M. Owen, R.W. Beard, T.W. McLain, Implementing dubins airplane paths on
human mobility, IEEE/ACM transactions on networking 19 (3) (2011) 630–643.
fixed-wing uavs, in: Handbook of Unmanned Aerial Vehicles, Springer, 2015,
[137] L. Cao, M. Grabchak, Smoothly truncated levy walks: toward a realistic
pp. 1677–1701.
mobility model, in: 2014 IEEE 33rd International Performance Computing and
[166] J. Wiest, M. Höffken, U. Kreßel, K. Dietmayer, Probabilistic trajectory prediction
Communications Conference (IPCCC), IEEE, 2014, pp. 1–8.
with gaussian mixture models, in: Intelligent Vehicles Symposium (IV), 2012
[138] A. Orsino, A. Ometov, G. Fodor, D. Moltchanov, L. Militano, S. Andreev, O.N.
IEEE, IEEE, 2012, pp. 141–146.
Yilmaz, T. Tirronen, J. Torsner, G. Araniti, et al., Effects of heterogeneous
[167] K. Chandrashekar, M.R. Dekhordi, J.S. Baras, Providing full connectivity in
mobility on d2d-and drone-assisted mission-critical mtc in 5g, IEEE Commun.
large ad-hoc networks by dynamic placement of aerial platforms, in: IEEE
Mag. 55 (2) (2017) 79–87.
MILCOM 2004. Military Communications Conference, 2004, 3, IEEE, 2004, pp.
[139] K. Kumari, B. Sah, S. Maakar, A survey: different mobility model for fanet,
1429–1436.
International Journal of Advanced Research in Computer Science and Software
[168] W.T.L. Teacy, J. Nie, S. McClean, G. Parr, Maintaining connectivity in UAV
Engineering 5 (6) (2015).
swarm sensing, in: 2010 IEEE Globecom Workshops, 2010, pp. 1771–1776,
[140] I. Bekmezci, M. Ermis, S. Kaplan, Connected multi UAV task planning for flying
http://dx.doi.org/10.1109/GLOCOMW.2010.5700246.
ad hoc networks, in: Communications and Networking (BlackSeaCom), 2014 [169] M. Rosalie, M.R. Brust, G. Danoy, S. Chaumette, P. Bouvry, Coverage Opti-
IEEE International Black Sea Conference on, IEEE, 2014, pp. 28–32. mization with Connectivity Preservation for UAV Swarms Applying Chaotic
[141] S.-J. Lee, M. Gerla, C.-C. Chiang, On-demand multicast routing protocol, in: Dynamics, in: 2017 IEEE International Conference on Autonomic Computing
Wireless Communications and Networking Conference, 1999. WCNC. 1999 (ICAC), 2017, pp. 113–118.
IEEE, 3, IEEE, 1999, pp. 1298–1302. [170] R. Costa, D. Rosario, E. Cerqueira, A. Santos, Enhanced connectivity for robust
[142] O. Bouachir, A. Abrassart, F. Garcia, N. Larrieu, A mobility model for UAV ad multimedia transmission in UAV networks, in: 2014 IFIP Wireless Days (WD),
hoc network, in: 2014 International Conference on Unmanned Aircraft Systems 2014, pp. 1–6.
(ICUAS), 2014, pp. 383–388, http://dx.doi.org/10.1109/ICUAS.2014.6842277. [171] S. Alsamhi, O. Ma, M.S. Ansari, Convergence of machine learning and robotics
[143] E. Kuiper, S. Nadjm-Tehrani, Mobility models for UAV group reconnaissance communication in collaborative assembly: mobility, connectivity and future
applications, in: Wireless and Mobile Communications, 2006. ICWMC’06. perspectives, J. Intell. Robot. Syst. (2019) 1–26.
International Conference on, IEEE, 2006, 33–33. [172] E.W. Dijkstra, A note on two problems in connexion with graphs, Numer. Math.
[144] E. Kieffer, G. Danoy, P. Bouvry, A. Nagih, Hybrid mobility model with 1 (1) (1959) 269–271, http://dx.doi.org/10.1007/BF01386390.
pheromones for UAV detection task, in: 2016 IEEE Symposium Series on [173] L. Ying, S. Shakkottai, A. Reddy, S. Liu, On combining shortest-path and back-
Computational Intelligence (SSCI), 2016, pp. 1–8, http://dx.doi.org/10.1109/ pressure routing over multihop wireless networks, IEEE/ACM Trans. Netw. 19
SSCI.2016.7850104. (3) (2011) 841–854, http://dx.doi.org/10.1109/TNET.2010.2094204.
[145] Vehicular mobility trace of the city of Cologne, Germany, Institute of Trans- [174] O.K. Sahingoz, Networking models in flying Ad-Hoc networks (FANETs):
portation Systems at the German Aerospace Center (ITS-DLR), 2016, http: Concepts and challenges, J. Intell. Robot. Syst. 74 (1) (2014) 513–527,
//kolntrace.project.citi-lab.fr/. 10.1007/s10846-013-9959-7.
[146] S. Uppoor, O. Trullols-Cruces, M. Fiore, J.M. Barcelo-Ordinas, Generation and [175] T.H. Clausen, P. Jacquet, Optimized link state routing protocol (OLSRP), in:
analysis of a large-scale urban vehicular mobility dataset, IEEE Trans. Mob. The Internet Engineering Task Force, MANET working Group, vol. 3626, 2003.
Comput. 13 (5) (2014) 1061–1075. [176] A. Alshabtat, L. Dong, J. Li, F. Yang, Low latency routing algorithm for
[147] B.A. Boghossian, S.A. Velastin, Image processing system for pedestrian moni- unmanned aerial vehicles Ad-hoc networks, in: Int. J. Elect. Comput. Eng., 6
toring using neural classification of normal motion patterns, Meas. Control 32 (2011) 48–54.
(9) (1999) 261–264. [177] S.Y. Dong, Optimization of OLSR routing protocol in UAV ad HOC network,
[148] O. Ossama, H.M.O. Mokhtar, M.E. El-Sharkawi, An extended k-means technique in: 2016 13th International Computer Conference on Wavelet Active Media
for clustering moving objects, Egypt. Inform. J. 12 (1) (2011) 45–51. Technology and Information Processing (ICCWAMTIP), 2016, pp. 90–94.
[149] M. Bennewitz, W. Burgard, G. Cielniak, S. Thrun, Learning motion patterns of [178] N. Coutinho, R. Matos, C. Marques, A. Reis, S. Sargento, J. Chakareski, A.
people for compliant robot motion, Int. J. Robot. Res. 24 (1) (2005) 31–48. Kassler, Dynamic dual-reinforcement-learning routing strategies for quality of
[150] J. Harri, F. Filali, C. Bonnet, Mobility models for vehicular ad hoc networks: a experience-aware wireless mesh networking, Elsevier Comput. Netw. 88 (2015)
survey and taxonomy, IEEE Commun. Surv. Tutor. 11 (4) (2009). 269–285.

23
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

[179] C.E. Perkins, P. Bhagwat, Highly dynamic destination-sequenced distance-vector [204] R. Shirani, M. St-Hilaire, T. Kunz, Y. Zhou, J. Li, L. Lamont, Combined reactive-
routing (DSDV) for mobile computers, in: Proceedings of the Conference on geographic routing for unmanned aeronautical Ad-hoc networks, in: 2012
Communications Architectures, Protocols and Applications, in: SIGCOMM ’94, 8th International Wireless Communications and Mobile Computing Conference
ACM, New York, NY, USA, 1994, pp. 234–244, http://dx.doi.org/10.1145/ (IWCMC), 2012, pp. 820–826.
190314.190336. [205] Y. Li, M. St-Hilaire, T. Kunz, Enhancements to reduce the overhead of the
[180] J. Chroboczek, The babel routing protocol, IETF RFC 6126, 2011, https://tools. reactive-greedy-reactive routing protocol for unmanned aeronautical Ad-Hoc
ietf.org/html/rfc6126. networks, in: 2012 8th International Conference on Wireless Communications,
[181] M. Al-Ghazal, A. El-Sayed, H. Kelash, Routing optimlzation using genetic Networking and Mobile Computing, 2012, pp. 1–4.
algorithm in Ad Hoc networks, in: 2007 IEEE International Symposium on [206] D. Rosario, Z. Zhao, A. Santos, T. Braun, E. Cerqueira, A beaconless
Signal Processing and Information Technology, 2007, pp. 497–503. opportunistic routing based on a cross-layer approach for efficient video
[182] S.A.N. Shaha, V. Pai, U.K. Shenoy, Comparison of wireless routing protocols dissemination in mobile multimedia IoT applications, Comput. Commun. 45
over FTP traffic in mobile and non mobile nodes, in: 2017 International (2014) 21–31, http://dx.doi.org/10.1016/j.comcom.2014.04.002, http://www.
Conference on Intelligent Computing, Instrumentation and Control Technologies sciencedirect.com/science/article/pii/S0140366414001388.
(ICICICT), 2017, pp. 349–353. [207] S. Temel, I. Bekmezci, LODMAC: Location Oriented directional MAC pro-
[183] B. Bellur, R.G. Ogier, A reliable, efficient topology broadcast protocol for tocol for FANETs, Comput. Netw. 83 (2015) 76–84, http://dx.doi.org/10.
dynamic networks, in: INFOCOM ’99. Eighteenth Annual Joint Conference of 1016/j.comnet.2015.03.001, http://www.sciencedirect.com/science/article/pii/
the IEEE Computer and Communications Societies. Proceedings, vol. 1, IEEE, S138912861500081X.
1999, pp. 178–186 vol.1. [208] S. Biswas, R. Morris, ExOR: Opportunistic routing in multi-hop wireless
[184] A. Neumann, C. Aichele, M. Lindner, S. Wunderlich, Ietf better ap- networks[C], in: Proc. of the ACM SIGCOMM 2005, 2005, pp. 133–143.
proach to mobile Ad hoc networking (B.A.T.M.A.N.), IETF Internet Draft [209] Y. Ko, N.H. Vaidya, Location-aided routing (LAR) in mobile Ad Hoc net-
draft-wunderlich-openmesh-manet-routing-00, 2008, https://tools.ietf.org/html/ works, Wireless Netw. 6 (4) (2000) 307–321, http://dx.doi.org/10.1023/A:
draft-wunderlichopenmesh-manet-routing-00. 1019106118419.
[185] D. B. Johnson, D. Maltz, Dynamic source routing in ad hoc wireless networks, [210] J. Mingliang, L. Jinyang, T.Y. Chiang, Cluster based routing protocol(CBRP)
Mobile Comput. 353 (1999). functional specification, 1998, https://tools.ietf.org/id/draft-ietf-manet-cbrp-
[186] S. Murthy, J.J. Garcia-Luna-Aceves, An efficient routing protocol for wireless spec-00.txt.
networks, Mobile Netw. Appl. 1 (2) (1996) 183–197, http://dx.doi.org/10. [211] K. Devarajan, V. Padmathilagam, An enhanced cluster gateway switch routing
1007/BF01193336. protocol (ecgsr) for congestion control using aodv algorithm in manet, Int.
[187] Y. Wang, J. Liu, A backup multipath routing protocol for ad hoc networks J. Comput. Appl. Technol. 123 (3) (2015) 37–42, http://dx.doi.org/10.5120/
with dynamic topology, in: 2012 3rd International Conference on System ijca2015905277.
Science, Engineering Design and Manufacturing Informatization, vol. 1, 2012, [212] K. Obraczka, K. Viswanath, G. Tsudik, Flooding for reliable multicast in multi-
pp. 237–241. hop ad hoc networks, Wireless Netw. 7 (6) (2001) 627–634, http://dx.doi.org/
[188] P.M. Carthy, D. Grigoras, Multipath associativity based routing, in: Second 10.1023/A:1012323519059.
Annual Conference on Wireless On-demand Network Systems and Services,
[213] Y. Sasson, D. Cavin, A. Schiper, Probabilistic broadcast for flooding in wireless
2005, pp. 60–69, http://dx.doi.org/10.1109/WONS.2005.24.
mobile ad hoc networks, in: 2003 IEEE Wirel. Commun. and Networking, 2003.
[189] R. Dube, C.D. Rais, K.-Y. Wang, S.K. Tripathi, Signal stability-based adaptive WCNC 2003, vol. 2, 2003, pp. 1124–1130, vol.2.
routing (SSA) for ad hoc mobile networks, IEEE Pers. Commun. 4 (1) (1997)
[214] H. Tian, H. Shen, T. Matsuzawa, RandomWalk routing for wireless sensor net-
36–45.
works, in: Sixth International Conference on Parallel and Distributed Computing
[190] S.S. Naik, A.U. Bapat, Message priority based routing protocol in MANETs, in:
Applications and Technologies (PDCAT’05), 2005, pp. 196–200.
2015 International Conference on Pervasive Computing (ICPC), 2015, pp. 1–5.
[215] X. Li, X. Tao, N. Li, Energy-efficient cooperative MIMO-based random walk
[191] Y.-H. Wang, C.-F. Chao, Dynamic backup routes routing protocol for mo-
routing for wireless sensor networks, IEEE Commun. Lett. 20 (11) (2016)
bile ad hoc networks, Inf. Sci. 176 (2) (2006) 161–185, http://dx.doi.
2280–2283.
org/10.1016/j.ins.2004.09.016, http://www.sciencedirect.com/science/article/
[216] Z. Zheng, A.K. Sangaiah, T. Wang, Adaptive communication protocols in flying
pii/S0020025504003275.
Ad Hoc network, IEEE Commun. Mag. 56 (1) (2018) 136–142, http://dx.doi.
[192] D. Dimitrova, M. Brogle, T. Braun, G. Heijenk, N. Meratnia, Joint ERCIM
org/10.1109/MCOM.2017.1700323.
eMobility and MobiSense Workshop, University of Bern, 2012.
[217] S. Rosati, K. Kruzelecki, L. Traynard, B.R. Mobile, Speed-aware routing for UAV
[193] J. Hope Forsmann, R. Hiromoto, J. Svoboda, A time-slotted on-demand routing
ad-hoc networks, in: 2013 IEEE Globecom Workshops (GC Wkshps), 2013, pp.
protocol for mobile ad hoc unmanned vehicle systems - art. no. 65611P, 2007.
1367–1373.
[194] Z.J. Haas, M.R. Pearlman, P. Samar, The zone routing protocol (ZRP) for
[218] S. Rosati, K. Kruzelecki, G. Heitz, D. Floreano, B. Rimoldi, Dynamic routing for
ad hoc networks, IETF Internet Draft draft-ietf-manet-zonezrp-04, 2002, https:
flying Ad Hoc networks, IEEE Trans. Veh. Technol. 65 (3) (2016) 1690–1700,
//www.ietf.org/proceedings/55/I-D/draft-ietf-manet-zone-zrp-04.txt.
http://dx.doi.org/10.1109/TVT.2015.2414819.
[195] V. Park, S. Corson, Temporally-ordered routing algorithm (TORA), IETF Internet
[219] L. Lin, Q. Sun, J. Li, F. Yang, A novel geographic position mobility oriented
Draft draft-ietf-manet-tora-spec-04, 2001, http://www.ietf.org/proceedings/52/
routing strategy for UAVs, J. Comput. Inf. Syst. 8 (2012) 709–716.
I-D/draftietf-manet-tora-spec-04.txt.
[196] B. Karp, H.T. Kung, Gpsr: greedy perimeter stateless routing for wireless [220] C. Zang, S. Zang, Mobility prediction clustering algorithm for UAV networking,
networks, in: Proceedings of the 6th Annual International Conference on Mobile in: 2011 IEEE GLOBECOM Workshops (GC Wkshps), 2011, pp. 1158–1161,
Computing and Networking, in: MobiCom ’00, ACM, New York, NY, USA, 2000, http://dx.doi.org/10.1109/GLOCOMW.2011.6162360.
pp. 243–254, http://dx.doi.org/10.1145/345910.345953. [221] G. Gankhuyag, A.P. Shrestha, S. Yoo, Robust and reliable predictive routing
[197] C. Liu, J. Wu, Efficient geometric routing in three dimensional Ad Hoc networks, strategy for flying ad-hoc networks, IEEE Access 5 (2017) 643–654.
in: IEEE INFOCOM 2009, 2009, pp. 2751–2755. [222] Y. Li, M. St-Hilaire, T. Kunz, Improving routing in networks of uavs via scoped
[198] R. Flury, R. Wattenhofer, Randomized 3D geographic routing, in: IEEE flooding and mobility prediction, in: 2012 IFIP Wireless Days, 2012, pp. 1–6,
INFOCOM 2008 - The 27th Conference on Computer Communications, 2008. http://dx.doi.org/10.1109/WD.2012.6402827.
[199] Y. Li, S. Xie, Y. Yu, Analysis of greedy forwarding in vehicular Ad Hoc networks, [223] W. Jung, J. Yim, Y. Ko, QGeo: Q-Learning-Based Geographic Ad Hoc routing
in: 2011 International Conference on System science, Engineering design and protocol for unmanned robotic networks, IEEE Commun. Lett. 21 (10) (2017)
Manufacturing informatization, vol. 2, 2011, pp. 344–347. 2258–2261, http://dx.doi.org/10.1109/LCOMM.2017.2656879.
[200] S. Li, H. Gao, D. Wu, An energy-balanced routing protocol with greedy [224] B. Sliwa, C. Schüler, M. Patchou, C. Wietfeld, PARRoT: Predictive Ad-hoc
forwarding for WSNs in cropland, in: 2016 IEEE International Conference on Routing fueled by reinforcement learning and trajectory knowledge, 2020,
Electronic Information and Communication Technology (ICEICT), 2016, pp. 1–7. arXiv:2012.05490.
[201] J. Zhou, Y. Chen, B. Leong, P.S. Sundaramoorthy, Practical 3D geographic [225] C. He, S. Liu, S. Han, A fuzzy logic reinforcement learning-based routing
routing for wireless sensor networks, in: Proceedings of the 8th ACM Conference algorithm for flying Ad Hoc networks, in: 2020 International Conference on
on Embedded Networked Sensor Systems, in: SenSys ’10, ACM, New York, NY, Computing, Networking and Communications (ICNC), 2020, pp. 987–991, http:
USA, 2010, pp. 337–350, http://dx.doi.org/10.1145/1869983.1870016. //dx.doi.org/10.1109/ICNC47757.2020.9049705.
[202] D. Rosario, Z. Zhao, T. Braun, E. Cerqueira, A. Santos, I. Alyafawi, Opportunistic [226] R.S. Sutton, A.G. Barto, Reinforcement learning: An introduction, IEEE Trans.
routing for multi-flow video dissemination over flying Ad-Hoc networks, in: Neural Netw. 16 (1988) 285–286.
Proceeding of IEEE International Symposium on a World of Wireless, Mobile [227] A. Rovira-Sugranes, F. Afghah, J. Qu, A. Razi, Fully-echoed Q-routing with
and Multimedia Networks 2014, 2014, pp. 1–6, http://dx.doi.org/10.1109/ simulated annealing inference for flying Adhoc networks, IEEE Trans. Netw.
WoWMoM.2014.6918947. Sci. Eng. (2021) http://dx.doi.org/10.1109/TNSE.2021.3085514.
[203] W. Qingwen, L. Gang, L. Zhi, Q. Qian, An adaptive forwarding protocol for [228] J.A. Boyan, M.L. Littman, Packet routing in dynamically changing networks:
three dimensional flying Ad Hoc networks, in: 2015 IEEE 5th International a reinforcement learning approach, in: Int. Conf. on Neural Info. Processing
Conference on Electronics Information and Emergency Communication, 2015, Systems, Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1993, pp.
pp. 142–145, http://dx.doi.org/10.1109/ICEIEC.2015.7284506. 671–678.

24
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

[229] S. Khodayari, M.J. Yazdanpanah, Network routing based on reinforcement [256] S.P.H. Engineering / UgCS, Leading drone control software, UgCS, https://www.
learning in dynamically changing networks, in: ICTAI’05, 2005, 5 pp.–366. ugcs.com/.
[230] S.P.M. Choi, D.-Y. Yeung, Predictive q-routing: a memory-based reinforce- [257] A.I. Hentati, L. Krichen, M. Fourati, L.C. Fourati, Simulation tools, environments
ment learning approach to adaptive traffic control, in: Conf. on Neural Info. and frameworks for UAV systems performance analysis, in: 2018 14th Interna-
Processing Systems, MIT Press, Cambridge, MA, USA, 1995, pp. 945–951. tional Wireless Communications Mobile Computing Conference (IWCMC), 2018,
[231] S. Kumar, R. Mukkulainen, Confidence based dual reinforcement q-routing: an pp. 1495–1500, http://dx.doi.org/10.1109/IWCMC.2018.8450505.
adaptive online network routing algorithm, in: IJCAI’99, Morgan Kaufmann [258] C. Coopmans, M. Podhradský, N.V. Hoffer, Software- and hardware-in-the-loop
Publishers Inc., San Francisco, CA, USA, 1999, pp. 758–763. verification of flight dynamics model and flight control simulation of a fixed-
[232] F. Tekiner, Z. Ghassemlooy, T.R. Srikanth, Comparison of the Q-routing and wing unmanned aerial vehicle, in: 2015 Workshop on Research, Education
shortest path routing algorithm, 2004. and Development of Unmanned Aerial Systems (RED-UAS), 2015, pp. 115–122,
[233] N. Gupta, M. Kumar, A. Sharma, M.S. Gaur, V. Laxmi, M. Daneshtalab, M. http://dx.doi.org/10.1109/RED-UAS.2015.7440998.
Ebrahimi, Improved route selection approaches using Q-learning framework for [259] Microsoft, Microsoft AirSim: Open source simulator based on unreal engine for
2D NoCs, in: Int. Workshop on Many-core Embedded Systems, ACM, New York, autonomous vehicles from microsoft AI and research, 2017, https://github.com/
NY, USA, 2015, pp. 33–40. Microsoft/AirSim.
[234] Y. Shilova, M. Kavalerov, I. Bezukladnikov, Full echo Q-routing with adap- [260] J. Antunes, AirSim: A Simulator to help AI research for use in drones,
tive learning rates: A reinforcement learning approach to network routing, commercial UAV news, 2017, https://www.commercialuavnews.com/public-
in: IEEE Conf. EIConRusNW, 2016, pp. 341–344, http://dx.doi.org/10.1109/ safety/airsim-simulator-help-artificial-intelligence-research-use-drones.
[261] NS-3 Tutorial, 2020, https://www.nsnam.org/docs/release/3.32/tutorial/html/
EIConRusNW.2016.7448188.
index.html.
[235] M. Kavalerov, Y. Shilova, Y. Likhacheva, Adaptive q-routing with random
[262] OPNET Network Simulator, Opnet Projects, 2020, https://opnetprojects.com/
echo and route memory, in: Proceedings of the 20th Conference of Open
opnet-network-simulator/.
Innovations Association FRUCT, FRUCT Oy, Helsinki, Finland, Finland, 2017,
[263] M. Calvo-Fullana, D. Mox, A. Pyattaev, J. Fink, V. Kumar, A. Ribeiro, ROS-
pp. 20:138–20:145.
NetSim: A Framework for the integration of robotic and network simulators,
[236] D. Sharma, D. Kukreja, P. Aggarwal, M. Kaur, A. Sachan, Poisson’s probability-
IEEE Robotics Autom. Lett. 6 (2) (2021) 1120–1127, http://dx.doi.org/10.1109/
based q-routing techniques for message forwarding in opportunistic networks,
LRA.2021.3056347.
Int. J. Commun. Syst. 31 (2018) http://dx.doi.org/10.1002/dac.3593.
[264] S.M. Najafabadi, N. Mastronarde, M.J. Medley, J.D. Matyjas, UB-ANC: An Open
[237] F. Wang, R. Feng, H. Chen, Dynamic routing algorithm with q-learning for platform testbed for software-defined airborne networking and communications,
internet of things with delayed estimator, IOP Conf. Ser. Earth Environ. Sci. CoRR, abs/1509.08346, 2015, arXiv:1509.08346.
234 (2019) 012048. [265] V. Marojevic, I. Guvenc, R. Dutta, M.L. Sichitiu, B.A. Floyd, Advanced wireless
[238] T. Hendriks, M. Camelo, S. Latré, Q2-routing : a qos-aware q-routing algorithm for unmanned aerial systems: 5G standardization, research challenges, and
for wireless ad hoc networks, in: 2018 14th International Conference on AERPAW architecture, IEEE Veh. Technol. Mag. 15 (2) (2020) 22–30, http:
Wireless and Mobile Computing, Networking and Communications (WiMob), //dx.doi.org/10.1109/MVT.2020.2979494.
2018, pp. 108–115, http://dx.doi.org/10.1109/WiMOB.2018.8589161. [266] M.L. Sichitiu, I. Guvenc, R. Dutta, V. Marojevic, B. Floyd, AERPAW Emulation
[239] N. Lyu, G. Song, B. Yang, Y. Cheng, QNGPSR: A Q-Network Enhanced geo- overview, in: Proceedings of the 14th International Workshop on Wireless Net-
graphic Ad-Hoc routing protocol based on GPSR, in: 2018 IEEE 88th Vehicular work Testbeds, Experimental Evaluation & Characterization, in: WiNTECH’20,
Technology Conference (VTC-Fall), 2018, pp. 1–6, http://dx.doi.org/10.1109/ Association for Computing Machinery, New York, NY, USA, 2020, pp. 1–8,
VTCFall.2018.8690651. http://dx.doi.org/10.1145/3411276.3412188.
[240] J. LIU, Q. WANG, C. HE, Y. XU, ARdeep: Adaptive And reliable routing [267] NC State named a hot spot for 5G innovation, NC State News, 2019, https:
protocol for mobile robotic networks with deep reinforcement learning, in: //news.ncsu.edu/2019/12/5g-wireless-network-aerpaw/.
2020 IEEE 45th Conference on Local Comput. Netw. (LCN), 2020, pp. 465–468, [268] J. Breen, A. Buffmire, J. Duerig, K. Dutt, E. Eide, M. Hibler, D. Johnson, S.K.
http://dx.doi.org/10.1109/LCN48667.2020.9314848. Kasera, E. Lewis, D. Maas, A. Orange, N. Patwari, D. Reading, R. Ricci, D.
[241] Y. Chen, N. Lyu, G. Song, et al., A traffic-aware Q-network enhanced routing Schurig, L.B. Stoller, J. Van der Merwe, K. Webb, G. Wong, POWDER: Platform
protocol based on GPSR for unmanned aerial vehicle ad-hoc networks, Front. For open wireless data-driven experimental research, in: Proceedings of the 14th
Inform. Technol. Electron. Eng. 21 (2020) 1308–1320, http://dx.doi.org/10. International Workshop on Wireless Network Testbeds, Experimental Evaluation
1631/FITEE.1900401. and Characterization (WiNTECH), 2020, http://dx.doi.org/10.1145/3411276.
[242] J. Liu, Q. Wang, C. He, K. Jaffrès-Runser, Y. Xu, Z. Li, Y. Xu, 3412204.
QMR:Q-learning Based multi-objective optimization routing protocol for fly- [269] P. Gawłowicz, A. Zubow, Ns-3 meets openai gym: the playground for ma-
ing Ad Hoc networks, Comput. Commun. 150 (2020) 304–316, http: chine learning in networking research, in: ACM International Conference
//dx.doi.org/10.1016/j.comcom.2019.11.011, https://www.sciencedirect.com/ on Modeling, Analysis and Simulation of Wireless and Mobile Systems
science/article/pii/S0140366419308278. (MSWiM), 2019, http://www.tkn.tu-berlin.de/fileadmin/fg112/Papers/2019/
[243] Q. Yang, S. Jang, S. Yoo, Q-Learning-Based Fuzzy logic for multi-objective gawlowicz19_mswim.pdf.
routing algorithm in flying Ad Hoc networks, Wireless Pers. Commun. 113 [270] H. Yin, P. Liu, L. Zhang, L. Cao, Y. Gao, X. Hei, X.-J. Hei, NS3-AI: Enable
(2020) 115–138, http://dx.doi.org/10.1007/s11277-020-07181-w. Applying artificial intelligence to network simulation in ns-3, 2020.
[244] S. Jiang, Z. Huang, Y. Ji, Adaptive UAV-assisted geographic routing with Q- [271] AERPAW Equipment, AERPAW, https://aerpaw.org/aerpaw-equipment/.
Learning in VANET, IEEE Commun. Lett. (2020) http://dx.doi.org/10.1109/ [272] A. Antonini, W. Guerra, V. Murali, T. Sayre-McCord, S. Karaman, The blackbird
LCOMM.2020.3048250, 1–1. dataset: A large-scale dataset for UAV perception in aggressive flight, 2018,
[245] J. Wu, M. Fang, X. Li, Reinforcement learning based mobility adaptive routing arXiv:1810.01987.
for vehicular Ad-Hoc networks, 101 (4) (2018) 2143–2171 http://dx.doi.org/ [273] N. Smith, N. Moehrle, M. Goesele, W. Heidrich, UAV Pathplanning dataset
10.1007/s11277-018-5809-z. & benchmark, KAUST research repository, 2018, http://dx.doi.org/10.25781/
[246] O. Oubbati, N. CHAIB, A. Lakas, S. Bitam, P. Lorenz, U2rv: uav-assisted KAUST-SUHGA, http://hdl.handle.net/10754/630159.
[274] M.l.F. Vandroogenbroeck, Marc, Mid-Air Dataset, https://midair.ulg.ac.be/.
reactive routing protocol for vanets, Int. J. Commun. Syst. PP (2019) 1–13,
[275] J. Delmerico, T. Cieslewski, H. Rebecq, M. Faessler, D. Scaramuzza, Are we
http://dx.doi.org/10.1002/dac.4104.
ready for autonomous drone racing? The UZH-FPV drone racing dataset, in:
[247] Plane 11 Flight Simulator: More Powerful. Made Usable. X, 2020, https://www.
IEEE Int. Conf. Robot. Autom. (ICRA), 2019.
x-plane.com/.
[276] X. Pan, P. Desbarats, S. Chaumette, A deep learning based framework for
[248] A. Perry, The FlightGear Flight Simulator, 2004.
UAV trajectory pattern recognition, in: 2019 Ninth International Conference
[249] Osrf, Why Gazebo?, gazebo, http://gazebosim.org/.
on Image Processing Theory, Tools and Applications (IPTA), 2019, pp. 1–6,
[250] jMAVSim Simulation PX4 v1.9.0 Developer Guide, https://dev.px4.io/v1.9.0/ http://dx.doi.org/10.1109/IPTA.2019.8936099.
en/simulation/jmavsim.html. [277] X. Liu, Y. Liu, Y. Chen, L. Hanzo, Trajectory design and power control for multi-
[251] S. Shah, D. Dey, C. Lovett, A. Kapoor, AirSim: High-Fidelity Visual and UAV assisted wireless networks: A machine learning approach, IEEE Trans.
physical simulation for autonomous vehicles, CoRR, abs/1705.05065, 2017, Veh. Technol. 68 (8) (2019) 7957–7969, http://dx.doi.org/10.1109/TVT.2019.
arXiv:1705.05065. 2920284.
[252] M. Müller, V. Casser, J. Lahoud, N. Smith, B. Ghanem, UE4Sim: A Photo- [278] H. Wang, J. Wang, J. Chen, Y. Gong, G. Ding, Network-connected UAV com-
realistic simulator for computer vision applications, CoRR, abs/1708.05869, munications: potentials and challenges, CoRR, abs/1806.04583, 2018, arXiv:
2017, arXiv:1708.05869. 1806.04583.
[253] Overview - QGroundControl user guide, https://docs.qgroundcontrol.com/ [279] M. Cooney, Drones still face major communications challenges getting onto
master/en/index.html. US airspace, Network World, 2013, https://www.networkworld.com/article/
[254] Mission planner ground control station - Mission Planner documentation, https: 2224063.
//ardupilot.org/planner/docs/mission-planner-ground-control-station.html. [280] M. Mozaffari, W. Saad, M. Bennis, Y. Nam, M. Debbah, A tutorial on uavs for
[255] A.-M. Andreescu, M. Dima, A. Istrate, C. Muresan, C. Visoiu, P. Parvu, wireless networks: applications, challenges, and open problems, IEEE Commun.
Autonomous system for image geo-tagging and target recognition, 2014. Surv. Tutor. (2019) http://dx.doi.org/10.1109/COMST.2019.2902862.

25
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

[281] A. Fotouhi, H. Qiang, M. Ding, M. Hassan, L.G. Giordano, A. García-Rodríguez, [306] F. Delgado, J. Carvalho, T. Coelho, A.D. Santos, An optical fiber sensor and its
J. Yuan, Survey on UAV cellular communications: practical aspects, standardiza- application in UAVs for current measurements, Sensors 16 (11) (2016) 1800,
tion advancements, regulation, and security challenges, CoRR, abs/1809.01752, http://dx.doi.org/10.3390/s16111800.
2018, arXiv:1809.01752. [307] W. Griethe, M. Gregory, F. Heine, H. Kampfner, High speed laser
[282] H. Shakhatreh, A. Sawalmeh, A.I. Al-Fuqaha, Z. Dou, E. Almaita, I. Khalil, N.S. communications in UAV scenarios, 2011, http://dx.doi.org/10.1117/12.884274.
Othman, A. Khreishah, M. Guizani, Unmanned aerial vehicles: A survey on [308] B.O. Sadiq, A feasibility study of using light fidelity with multiple unmanned
civil applications and key research challenges, CoRR, abs/1805.00881, 2018, aerial vehicles for indoor collaborative and cooperative networking, CoRR,
arXiv:1805.00881. abs/1707.08627, 2017, arXiv:1707.08627.
[309] Jun Luo, J. Hubaux, P.T. Eugster, DICTATE: DIstributed CerTification Authority
[283] Y. Fu, S. Wang, C. Wang, X. Hong, S. McLaughlin, Artificial intelligence to
with probabilisTic frEshness for ad hoc networks, IEEE Trans. Dependable Secur.
manage network traffic of 5G wireless networks, IEEE Netw. 32 (6) (2018)
Comput. 2 (4) (2005) 311–323, http://dx.doi.org/10.1109/TDSC.2005.49.
58–64, http://dx.doi.org/10.1109/MNET.2018.1800115.
[310] L. Eschenauer, V.D. Gligor, A key-management scheme for distributed sensor
[284] L. Xiao, X. Lu, D. Xu, Y. Tang, L. Wang, W. Zhuang, UAV Relay in VANETs
networks, in: Proceedings of the 9th ACM Conference on Computer and Com-
against smart jamming with reinforcement learning, IEEE Trans. Veh. Technol.
munications Security, in: CCS ’02, Association for Computing Machinery, New
67 (5) (2018) 4087–4097, http://dx.doi.org/10.1109/TVT.2018.2789466.
York, NY, USA, 2002, pp. 41–47, http://dx.doi.org/10.1145/586110.586117.
[285] P. Fraga-Lamas, L. Ramos, V.M. Mondéjar-Guerra, T. Fernández-Caramés, A [311] M. Gharib, E. Emamjomeh-Zadeh, A. Norouzi-Fard, A. Movaghar, A novel
review on IoT deep learning UAV systems for autonomous obstacle detection probabilistic key management algorithm for large-scale MANETs, in: 2013 27th
and collision avoidance, Remote Sens. 11 (2019) 2144, http://dx.doi.org/10. International Conference on Advanced Information Networking and Applica-
3390/rs11182144. tions Workshops, 2013, pp. 349–356, http://dx.doi.org/10.1109/WAINA.2013.
[286] S. Islam, Q. Huang, F. Afghah, P. Fule, A. Razi, Fire frontline monitoring 107.
by enabling uav-based virtual reality with adaptive imaging rate, in: 2019 [312] I.J. Jensen, D.F. Selvaraj, P. Ranganathan, Blockchain technology for net-
53rd Asilomar Conference on Signals, Systems, and Computers, IEEE, 2019, worked swarms of unmanned aerial vehicles (UAVs), in: 2019 IEEE 20th
pp. 368–372. International Symposium on‘‘ A World of Wireless, Mobile and Multimedia
[287] M. Erdelj, E. Natalizio, UAV-Assisted disaster management: Applications and Networks’’(WoWMoM), IEEE, 2019, pp. 1–7.
open issues, in: 2016 international conference on computing, networking and [313] T. Alladi, N. Naren, G. Bansal, V. Chamola, M. Guizani, SecAuthUAV: A Novel
communications (ICNC), IEEE, 2016, pp. 1–5. authentication scheme for UAV-base station scenario, IEEE Trans. Veh. Technol.
[288] L.F. Gonzalez, G.A. Montes, E. Puig, S. Johnson, K. Mengersen, K.J. Gaston, (2020).
Unmanned aerial vehicles (uavs) and artificial intelligence revolutionizing [314] A. Couturier, M. Akhloufi, A review on absolute visual localization for UAV,
wildlife monitoring and conservation, Sensors 16 (1) (2016) 97. Robot. Auton. Syst. 135 (2021) 103666, http://dx.doi.org/10.1016/j.robot.
[289] M.Y. Arafat, S. Moh, Location-aided delay tolerant routing protocol in UAV 2020.103666.
networks for post-disaster operation, IEEE Access 6 (2018) 59891–59906, http: [315] G. Bansal, V. Chamola, Lightweight authentication protocol for inter base
//dx.doi.org/10.1109/ACCESS.2018.2875739. station communication in heterogeneous networks, in: IEEE INFOCOM 2020
- IEEE Conference on Comput. Commun. Workshops (INFOCOM WKSHPS),
[290] B. Uragun, Energy efficiency for unmanned aerial vehicles, in: 2011 10th
2020, pp. 871–876, http://dx.doi.org/10.1109/INFOCOMWKSHPS50562.2020.
International Conference on Machine Learning and Applications and Workshops,
9162714.
vol. 2, 2011, pp. 316–320, http://dx.doi.org/10.1109/ICMLA.2011.159.
[316] T. Erpek, Y.E. Sagduyu, Y. Shi, Deep learning for launching and mitigating
[291] H. Peng, A. Razi, F. Afghah, J.D. Ashdown, A unified framework for joint
wireless jamming attacks, IEEE Trans. Cogn. Commun. Netw. 5 (1) (2019) 2–14,
mobility prediction and object profiling of drones in UAV networks, CoRR,
http://dx.doi.org/10.1109/TCCN.2018.2884910.
abs/1808.00058, 2018, arXiv:1808.00058. [317] L. Abusalah, A. Khokhar, M. Guizani, A survey of secure mobile Ad Hoc
[292] J. Chen, U. Mitra, D. Gesbert, Optimal UAV relay placement for single user routing protocols, IEEE Commun. Surv. Tutor. 10 (4) (2008) 78–93, http:
capacity maximization over terrain with obstacles, in: 2019 IEEE 20th Inter- //dx.doi.org/10.1109/SURV.2008.080407.
national Workshop on Signal Processing Advances in Wireless Communications
(SPAWC), 2019, pp. 1–5, http://dx.doi.org/10.1109/SPAWC.2019.8815496.
[293] Z. Pan, L. An, C. Wen, Recent advances in fuel cells based propulsion systems
Arnau Rovira-Sugranes received his Ph.D. in Informatics
for unmanned aerial vehicles, Appl. Energy 240 (2019) 473–485, http://dx.doi.
and Computing from the School of Informatics, Computing
org/10.1016/j.apenergy.2019.02.079.
and Cyber Systems (SICCS) in Northern Arizona University
[294] E.F. Costa, D.A. Souza, V.P. Pinto, M.S. Araújo, A. Peixoto, E.D.C. Júnior,
(NAU), Flagstaff, AZ, USA, in August 2021. Earlier, he
Prediction of lithium-ion battery capacity in uavs, 2019,
graduated with a Master of Science in Informatics in 2019
[295] Q. Wu, G. Zhang, D.W.K. Ng, W. Chen, R. Schober, Generalized wireless- from Northern Arizona University and a Bachelor’s degree in
powered communications: When to activate wireless power transfer?, IEEE Industrial Electronics and Automation Engineering in 2016,
Trans. Veh. Technol. 68 (8) (2019) 8243–8248, http://dx.doi.org/10.1109/TVT. from Rovira i Virgili University. After graduating, he worked
2019.2924051. in an automotive seating and electrical systems company
[296] A. Kurs, A. Karalis, R. Moffatt, J. Joannopoulos, P. Fisher, M. Soljacic, Wireless before starting his Ph.D. program.
power transfer via strongly coupled magnetic resonances, Science (New York, His research interests include machine learning and data
N.Y.) 317 (2007) 83–86, http://dx.doi.org/10.1126/science.1143254. mining, communication and routing protocols and graph
[297] M. Khonji, M. Alshehhi, C.-M. Tseng, C.-K. Chau, Autonomous inductive theory for flying ad-hoc wireless networks (FANETs), which
charging system for battery-operated electric drones, pp. 322–327, 2017, http: produced peer-reviewed publications on predictive based
//dx.doi.org/10.1145/3077839.3078462. solutions for Internet of Things. Also, he has served as
[298] M. Nguyen, L.D. Nguyen, T.Q. Duong, H.D. Tuan, Real-time optimal resource an IEEE conference/journal paper reviewer and symposium
allocation for embedded UAV communication systems, IEEE Wirel. Commun. moderator.
Letters 8 (1) (2019) 225–228, http://dx.doi.org/10.1109/LWC.2018.2867775.
[299] S. Bi, R. Zhang, Placement optimization of energy and information access points
in wireless powered communication networks, IEEE Trans. Wirel. Commun. 15 Abolfazl Razi is an assistant professor in the School of
(3) (2016) 2351–2364, http://dx.doi.org/10.1109/TWC.2015.2503334. Informatics, Computing and Cyber Systems at Northern
Arizona University (NAU). He received his B.S., M.S. and
[300] A.M.R. Tolba, Trust-based distributed authentication method for collision attack
Ph.D. degrees, all in Electrical Engineering, respectively
avoidance in VANETs, IEEE Access 6 (2018) 62747–62755, http://dx.doi.org/
from Sharif University (1994–1998), Tehran Polytechnic
10.1109/ACCESS.2018.2875906.
(1999–2001), and University of Maine (2009–2013). Prior
[301] T.-J. Yang, Y.-H. Chen, V. Sze, Designing energy-efficient convolutional neural
to joining NAU, he held two postdoctoral positions in
networks using energy-aware pruning, in: Proceedings of the IEEE Conference
the field of machine learning and predictive modeling at
on Computer Vision and Pattern Recognition, 2017, pp. 5687–5695.
Duke University (2013–2014), and Case Western Reserve
[302] T. Johann, M. Dick, S. Naumann, E. Kern, How to measure energy-efficiency
University (2014–2015). He is the recipient of several
of software: Metrics and measurement results, in: 2012 First International
competitive awards including the Best Research of MCI in
Workshop on Green and Sustainable Software (GREENS), IEEE, 2012, pp. 51–54.
2008, Best Graduate Research Assistant of the Year Award
[303] E. García Martín, Energy efficiency in machine learning: A position paper,
from the College of Engineering, University of Maine in
in: 30th Annual Workshop of the Swedish Artificial Intelligence Society SAIS,
2011, and the Best Paper Award from the IEEE/CANEUS
Karlskrona, vol. 137, Linköping University Electronic Press, 2017, pp. 68–72.
Fly By Wireless Workshop in 2011. His current research
[304] C. Zhang, W. Zhang, Spectrum sharing for drone networks, IEEE J. Sel.
interests include smart connected communities, biomedical
Areas Commun. 35 (1) (2017) 136–144, http://dx.doi.org/10.1109/JSAC.2016.
signal processing, wireless networking, Internet of things,
2633040.
and predictive modeling.
[305] J. Kakar, V. Marojevic, Waveform and spectrum management for unmanned
aerial systems beyond 2025, CoRR, abs/1708.01664, 2017, arXiv:1708.01664.

26
A. Rovira-Sugranes et al. Ad Hoc Networks 130 (2022) 102790

Fatemeh Afghah is an Associate Professor with the School Jacob Chakareski is an Associate Professor in the College
of Informatics, Computing, and Cyber Systems, Northern of Computing at the New Jersey Institute for Technology
Arizona University (NAU), Flagstaff, AZ, USA, where she (NJIT), where he holds the Panasonic Chair of Sustain-
is the Director of Wireless Networking and Information ability and directs the Laboratory for AI-Enabed Virtual
Processing (WiNIP) Laboratory. Prior to joining NAU, she and Augmented Reality Immersive Communications and
was an Assistant Professor with the Electrical and Computer Networked Systems. His research interests span next gen-
Engineering Department, North Carolina A&T State Univer- eration virtual and augmented reality systems, UAV IoT
sity, Greensboro, NC, USA, from 2013 to 2015. Her research sensing and networking, fast reinforcement learning, 5G
interests include wireless communication networks, decision wireless edge computing and caching, ubiquitous immersive
making in multi-agent systems, radio spectrum management, communication, and societal applications. He received the
and artificial intelligence in healthcare. Her research has Adobe Data Science Faculty Research Award in 2017 and
been continually supported by NSF , AFRL, AFOSR, and 2018, the Swiss NSF Career Award Ambizione (2009), the
ABOR. AFOSR Faculty Fellowship in 2016 and 2017, and Best
She is the recipient of several awards including the Paper Awards at ICC 2017 and MMSys 2021. He completed
Air Force Office of Scientific Research Young Investigator his Ph.D. degree at Rice and Stanford in Electrical and
Award in 2019, NSF CAREER Award in 2020, NAU’s Most Computer Engineering, held research appointments with
Promising New Scholar Award in 2020, and NSF CRII Award Microsoft, HP Labs, and EPFL, and served on the advisory
in 2017. She is the author/co-author of over 80 peer- board of Frame, Inc. His research has been supported by the
reviewed publications and served as the organized and the NSF, NIH, AFOSR, Adobe, Tencent Research, NVIDIA, and
TPC chair for several international IEEE workshops in the Microsoft. For further info, please visit www.jakov.org.
field of UAV communications, including IEEE INFOCOM
Workshop on Wireless Sensor, Robot, and UAV Networks
(WiSRAN’19) and IEEE WOWMOM Workshop on Wireless
Networking, Planning, and Computing for UAV Swarms
(SwarmNet’20).

27