This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 1

Internet of Drones
Mirmojtaba Gharibi, Raouf Boutaba, Fellow, IEEE, and Steven L. Waslander, Member, IEEE

Abstract—The Internet of Drones (IoD) is a layered network inexpensive, and convenient access to the goods and items
control architecture designed mainly for coordinating the access already in or near an urban area, including consumer goods,
of unmanned aerial vehicles to controlled airspace, and providing fast-food, medicine, and even on-demand groceries.
navigation services between locations referred to as nodes. The
IoD provides generic services for various drone applications such Despite a wave of drone package delivery prototype an-
as package delivery, traffic surveillance, search and rescue and nouncements (e.g. Matternet [4], Amazon’s prime air [5],
more. In this paper, we present a conceptual model of how such Google’s project wing [6], and DHL’s Parcelcopter [7]), to
an architecture can be organized and we specify the features that the best of our knowledge, there are not any rigorous pub-
an IoD system based on our architecture should implement. For lication concerning the architecture of a drone-specific air
doing so, we extract key concepts from three existing large scale
networks, namely the air traffic control network, the cellular traffic management system as the technology is still in its
network, and the Internet and explore their connections to our infancy. One good starting point is NASA’s Unmanned Aerial
novel architecture for drone traffic management. A simulation System Traffic Management (UTM) project [8]–[10], which
platform for IoD is being implemented which can be accessed organized a symposium to begin preparations of a solution
from www.IoDnet.org in the future. for low altitude traffic management to be proposed to the
Index Terms—Layered architecture, Internet of Drones (IoD), FAA. Related to this effort, both Amazon [11], [12] and
Internet, cellular network, air traffic control (ATC), low altitude Google [13] have published white papers which explore some
air traffic management, unmanned aerial vehicle (UAV).
of the strategies for managing the airspace and coordinating
aerial vehicles through onboard system requirements such
I. I NTRODUCTION as ADS-B and Vehicle to Vehicle (V2V) communication.
However, our contribution is to approach the drone airspace
HE INTERNET OF DRONES1 is an architecture de-
T signed for providing coordinated access to controlled
airspace for unmanned aerial vehicles (UAVs), often referred
management problem by providing a universal architecture and
a vocabulary of concepts to describe the IoD. In the future,
different IoD systems can be developed based on it with their
to as drones. With the on-going miniaturization of sensors and
set of protocols and implementations of the features required
processors and ubiquitous wireless connectivity, drones are
by our IoD architecture. We suggest a possible operational
finding many new uses in enhancing our way of life. There are
model based on our architecture and we discuss the desired
many applications for drone technology, ranging from the on-
goals of the architecture and also the benefits that it provides
demand package delivery, to traffic and wild life surveillance,
as well as the subtleties that have to be addressed for any IoD
inspection of infrastructure, search and rescue, agriculture, and
system.
cinematography. All these applications share a common need
for both navigation and airspace management. In this work, Shortly after we published our preprint [14], authors in [15]
we lay the architecture for generic services that can provide published a preprint exploring some of the ideas pertaining
such a foundation for all current and future applications. to a UAV traffic network, called uNet. In uNet, instead of
using a free-flight mode, similar to our architecture as will be
Among these applications, aerial package delivery will most
explained in the section architecture, the airspace is divided
urgently require a robust airspace allocation architecture, as it
into predefined routes. The authors argue that this provides
could result in many thousands of daily flights in the same
for less reliance on advanced sense and avoid technologies
geographic area, with many potential conflicts between drones
and the ease of assigning conflict-free routes to the drones
navigating along similar or intersecting routes. The benefit to
using the existing techniques. They consider use of sector-
the global logistics network is clear, as drones could usher
level uNets (sNets) where the traffic in each sector is under the
in a new era of on-demand delivery, and has been shown
authority of that particular uNet. We have a similar construct
to be cost-competitive relative to ground-based delivery as
in our architecture with different zones where each zone is
well [1], although longer haul transport clearly benefits from
under authority of one or multiple Zone Service Provider
bundling onto larger transport vehicles. Amazon states that
(ZSP). However, one difference is that in our architecture,
about 83% of their packages weigh below 2.5 kg [2], a
more than one ZSP can participate in managing the same zone.
reasonable maximum payload for today’s drones. Similarly,
Furthermore, we take a systematic approach in defining the
the average weight of packages delivered by Fedex is less than
layers of the architecture as well as the features that have to
5kg [3]. In our opinion, this model can provide on-demand,
be implemented for each layers.
M. Gharibi and R. Boutaba are with D. Cheriton School of Computer
Science, University of Waterloo, Waterloo, ON, N2L 3G1, Canada. Contacts:
[email protected] and [email protected] respectively. II. R ELEVANT N ETWORKS
S. L. Waslander is with the Mechanical and Mechatronics Department,
University of Waterloo, ON, N2L 3G1, Canada, [email protected] For designing the architecture of the IoD, we study three
1 Project’s homepage: www.IoDnet.org distinct large scale network structures; namely air traffic

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 2

control (ATC), cellular network, and the Internet. Each of technology for navigation and localizing other aircraft in their
these networks achieves some of the goals or functionalities proximity. ADS-B uses GPS for navigation and broadcasts
we desire for the IoD. In each case, however, their conceptual aircraft position periodically. Use of ADS-B Out (broadcaster
architecture falls short of providing a thorough solution to only with no receiver) within specific portions of airspace is
the unique challenges of IoD. Hence, the importance of mandated by 2020 [17]. Difficulty with aircraft localization
studying these systems is twofold. First, they have valuable has been a great problem in aviation, forcing most of the
lessons about how a scalable and fault tolerant network can be air traffic through certain preferred airways (analogous to the
engineered. Second, their differences guide us to IoD’s specific highways on the ground). However, use of ADS-B provides
challenges which have not been tackled before and are in need more efficient direct routing within NextGen which allows
of innovative solutions. We describe these structures through flying in a straight line from the departure to destination airport
a discussion of goals and functionality that are relevant to IoD (also known as free flight) by providing better situational
and the differences with IoD that need to be addressed in our awareness regarding the congestion. Unlike Internet where if
architecture. some part of a network exceeds its capacity, it conveniently
drops new transferring packets, this is not possible in ATC.
A. Air Traffic Control Network Therefore, all ARTCCs and sectors and airports must remain
within their capacity which makes in advance reservation nec-
ATC has strong relevance to IoD as efficiently utilizing the essary. Flight plans are submitted to a central entity called Air
airspace and maintaining collision free navigation is an integral Traffic Control System Command Center (ATCSCC) where
part of any IoD architecture. The functioning of ATC follows according to predicted loads, a delay is assigned to each flight
similar procedures around the globe. We briefly summarize to ensure the network will not be oversubscribed. Pilots will
the components of ATC in the United States. The Federal receive partial or complete clearance. Once airborne, with
Aviation Administration (FAA) is in charge of regulations and the unfolding of how the actual flights progress, additional
air safety, and has partitioned the United States’ airspace into delays are assigned to the flights. The idea is to apply these
24 areas each managed by one of the 24 Air Route Traffic delays as early as possible in the flight or before takeoff,
Control Centers (ARTCC) (Fig. 1). There are bilateral letters rather than near the end where the maneuver space and fuel
of agreement between any two adjacent ARTCCs on how capacity are limited. These delays can be achieved by ground
aircraft must transition from one ARTCC to another. Similarly, hold, lowering the cruising speed or by standard holding
within each ARTCC, the airspace is partitioned into between patterns. These assigned delays are communicated to the sector
20 to 80 sectors and each sector is exclusively managed by one controllers so they know how long they must keep the aircraft
controller and the aircraft transitions between sectors are done in their designated sector. Interested readers are referred to
according to facility directives. The main driver in designating [18] for a full treatment of air traffic control systems.
the boundaries of ARTCCs as well as the sectors within
There are certain differences between IoD and ATC. As
each ARTCC is to distribute the load in an equitable way.
the number of drones scales up to the thousands sharing the
As it is evident in Fig. 1, the high volume of flights in the
limited airspace at any time, use of a centralized entity like
densely populated east coast translates into a higher number
ATCSCC for load prediction and assignment is not possible.
of ARTCCs than the central United States.
Hence, one has to look at decentralized solutions. With that
volume of flights, separation must be autonomously done by
ZSE the drones and it is not wise to rely on human interventions
ZMP ZBW for safety management, in contrast to NextGen. The limited
ZLC
ZAU ZOB
ZNY
airspace of the urban environment can only accommodate
ZOA
ZDV drones that have minimum performance requirement which,
ZID
ZKC ZDC
ZLA
depending on the situation, can be stringent such as a require-
ZAB ZME
ZFW ZTL
ment to execute holding patterns in a small area (ideally hover
ZJX as in the case of vertical take off and landing (VTOL) aircraft)
ZHU
and ability to easily land when necessary. This opens up
ZMA
many possibilities within IoD for handling congestion which
is not available to the ATC system. Free flight, although a step
Fig. 1. ARTCCs in the contiguous United States (recreated from FAA [16]).
The zones cover the airspace above and slightly beyond the contiguous United
forward for ATC, is only partially implementable within IoD
States. due to limited urban airspace, obstacles such as buildings and
birds and high level of congestion anticipated. In other words,
Traditionally, the main role of air traffic controllers was to the airspace must be highly regulated to ensure smooth air
keep a prescribed separation between all aircraft. However, traffic flow is achieved.
within the next generation of ATC (NextGen) – a new system
with the motivation to address the lack of scalability of the
current system, pilots are more autonomous and as a result in B. Cellular Network
charge of their own separation and controllers intervene only In the cellular network, the coverage area is partitioned
when necessary. This is possible due to pilots being equipped into most commonly hexagonal cells forming a honeycomb
with Automatic Dependent Surveillance-Broadcast (ADS-B) pattern. The communication signals in each cell are sent

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 3

to and received from the mobile users by a dedicated base holds exclusive rights to certain bandwidths in the frequency
station. Each base station uses a certain frequency which is spectrum. There are at least two reasons why a central design
different from the near base stations’ frequencies to minimize does not seem a good choice for IoD. Firstly, the tasks of
the interference. The range of signal for each base station IoD are computationally intensive. Hence we have to offload
determines the size of each cell. Each base station can only it to many autonomous systems which coordinate with each
carry a certain amount of calls over its frequency channel. As other. This way we reduce the complexity of the problem while
such, the main driver in determining the size of each cell is the settling for a less optimized solution. Secondly, as mentioned
expected number of mobile users in the region (Fig. 2). Hence in the cellular network, a portion of the frequency spectrum
the densely populated downtown areas can have many smaller is allocated exclusively to a corporation which means it has
cells whereas in the rural areas, fewer cells with higher range total control over its use. However, in our design for the
are used. Each of the base stations are connected to a central IoD, each portion of the airspace must be shared by all the
entity called Mobile Telecommunications Switching Office companies serving the same airspace and hence the amount of
(MTSO). The MTSO is in charge of periodic localization resources available to each company is less predictable. This
of the mobile units and assigning a base station to them. means flight planning is a more involved task in a trade-off
Furthermore, it assigns channels to each call and performs the for a more efficient service provider market. We believe the
task of handoff or handover which is basically the transfer of exclusive right to the portions of the spectrum has made the
responsibility for a moving mobile unit from one base station entry of new competitors to the cellular market quite difficult,
to the other base station as it enters a new partition. We will effectively resulting in a market with only a limited number
later use the same word in the context of IoD. See [19], [20], of providers.
and [21] for a comprehensive treatment of the cellular network.
C. Internet
In the Internet, the goal is to connect networks of computers
together, so all the computers on the world-wide network can
communicate. The Internet has a layered architecture consist-
ing of five layers as shown in Fig. 3. Layering makes it easier
to solve the problem that the Internet addresses by separating
concerns. Each of these layers is to be thought as a service and
upper layers use the services of lower layers. For example, the
link layer is concerned solely with the transfer of data on a
single communication link or between two adjacent nodes and
Fig. 2. An illustration of cellular networks and base stations. the physical layer is concerned with the physical means for
transferring signals through various mediums, such as air (in
Compared to the best effort philosophy of the Internet, in case of WiFi) or Ethernet cables. The Internet layer, relying
telecommunication, the philosophy is that a call must not be on the connectivity provided by the link layer is concerned
admitted if there are not enough resources to sustain it until mainly with routing or forwarding data packets between any
its completion. Hence, the handoff process poses a unique two nodes potentially on two different local networks through
challenge as it is not known whether admitting a call in a the use of standard global addressing as a best effort service
cell will result in later termination as the mobile unit enters a rather than a reliable one. This is achieved by routers which
new cell due to a lack of available channels in the new cell. locally make a decision about forwarding the data packets
Since the base stations usually belong to one corporation, the they receive to one of the immediately connected networks.
MTSO centrally makes decisions whether to allow access to Utilizing the universal unreliable connectivity provided by
a user in an effort to minimize the probability of a dropped Internet layer, the transport layer is concerned with tasks
call. As we will see, a similar problem exists for drones in such as the reliability of transmission and congestion control.
IoD. It is much less expensive to hold a drone on the ground Finally the application layer, uses this global and (if needs be)
than to allow it to takeoff and later ground it (order it to land) reliable connectivity for various applications like Web, Email,
or hold it (order to hover or execute holding patterns) due to VoIP, Remote Login, etc. Such a decentralized and deliberately
a lack of resources. Hence, IoD has a design philosophy that simple architecture has made the Internet a unique engineering
is similar to that of cellular telecommunications networks. feat in that it scaled by many orders of magnitude. Readers
There are still various differences between IoD and cellular can refer to [22] for a comprehensive treatment of the subject
networks in an abstract level. A subtle difference is that of the Internet and to [23] and [24] for discussions of the
in the case of cellular network, the MTSO does not know philosophical guidelines in its design.
which cell will be the next cell a mobile unit will enter after There are similarities and differences between the Internet
admitting the call in the first place. But in IoD the source and IoD. Routing is a task performed by both networks.
and destination is known to a greater extent for a trip by However, the time scale on which the Internet operates is
the drones which will allow a more optimized utilization of much smaller. In the case of IoD, the longer computation time
the network resources. Another difference is the central role can allow for the calculation of more optimal routes. Thus, a
MTSO plays which is in part possible because each company possibility is to adopt the routing protocols and adjust them

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 4

Application will make on-demand package delivery as well as other drone

Transport applications possible.
Internet
Link IV. A RCHITECTURE
Physical In this section, we explain our architecture in more detail.
Fig. 3. Layers in the architecture of the Internet The purpose of our architecture is to provide extensible generic
services to a diverse range of applications, namely navigation
service between any two nodes in an efficient and coordinated
accordingly. Another difference is that in the Internet, packets manner as well as other common or future services such as
that overload the system can be conveniently dropped since location aware communication. A need for navigation is the
it is buffered and resending it is cheap. In the IoD case, it common denominator for drone applications. Serving this need
is not possible to drop drones since they are physical objects will enable these applications to build on top of the services
and the only option is to remove them from the airspace by provided by the architecture. Furthermore, drones are mobile
ordering them to land and providing resources to them to yet tasks are local. In case a pool of worker drones rather than
execute a landing order which is an expensive task. Thus some individual drones are responsible for performing these task,
kind of reservation has to be enforced to ensure the system only the local drones (i.e. those near the task location) should
operates within its capacity to remain economical and viable. be notified. Hence, providing a mechanism for location aware
Whereas ATC is not a system that scales well, the Internet communication is another common need of the applications
is designed and shown to scale well and with the expected as well as other services for which the need will become
proliferations of drones, IoD has to be an architecture that can apparent in the future. Two important concepts to distinguish
scale. Using the Internet’s design guidelines that has afforded in our paper are that of an IoD architecture and an IoD system:
it such scalability, such as a decentralized design or providing an architecture gives abstract design and feature requirements
generic services with the least amount of assumption about that need to be implemented by any system that is based
the users of the services is monumental in IoD. on that architecture whereas a system gives concrete proto-
cols (interfaces and algorithms) that implement the features
III. O UR CONTRIBUTION required by the architecture. Hence, it is possible to have
many IoD systems all based on the same core architecture
Our core contribution is formulating a complex and mul-
each with their own advantages and disadvantages. Obviously
tifaceted problem and showing how in an abstract level, it
in any engineering project, not all architectures are viable.
is related to the vast amount of existing literature on the
Accordingly, at least one working IoD system must implement
three existing networks, namely air traffic control, cellular
an IoD architecture to prove it is viable.
network, and the Internet. We have crafted a blueprint for
the implementation of an IoD system based on our IoD
architecture. By comparing the challenges that IoD and each A. Structure
of the three named networks address in an abstract way, we To describe our architecture, first we need to introduce a
have established relationships between existing solutions to set of concepts and explain how they are related to each
the specific problems of IoD, hence creating well formulated other in our architecture. Words with special meanings for
open problems for the research community in a diverse range our architecture are italicized and they will form a vocabulary
of fields. For instance, on an abstract level all four networks for discussing it.
have to route physical objects or data. We have uncovered Airspace is the resource that is utilized by the drones. In
this connection and others such as congestion control, admis- our architecture, the airspace is structured similar to the roads
sion control, and addressing schemes. We have explained the network in the cities. Drones are only allowed inside the
existing strategies and made it clear what prevents a straight following three: airways playing a similar role to the roads,
forward adoption of them for IoD on some of these matters. intersections formed by at least two airways, and nodes which
As mentioned before, although there have been numerous are the points of interest reachable through an alternating
announcements in the media on drone applications such as sequence of airways and intersections. Each of these three has
package delivery prototypes, there are not any publication concrete geometric shape and is guaranteed to be collision
on the architecture for these systems. The FAA’s move to free from static structures. Movement of drones inside the
address integration of drones in the national airspace [25], airways and intersections is regulated (for example drones
in response to a mandate by the US House of Representatives must move only in the designated direction(s) of an airway
[26] reiterates that IoD is a timely architecture that addresses or intersection) whereas inside the nodes, drones are in the
important questions in this arena. Although there is significant free flight mode (Fig. 4). The airspace is partitioned into zones
excitement in the industry, to this date, this topic has not and hence each zone contains its airways, intersections, and
received much attention in the academic community. IoD nodes. Adjacent zones are reachable from each other through
serves as a first step for bringing these important issues to the inbound and outbound gates which are the intersections at
forefront of academic endeavours and provides the academic the border but they are special in that they belong to both
community with well-defined problems to tackle. Our hope is zones. No airway is allowed to cross the border between two
that an implementation of IoD in the next three to five years zones, unless it is segmented into two airways with a gate at

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 5

Intersection
Airway

Node
Zone 3 Zone 1
Airway Intersection Gate

Zone 2

Node

Fig. 5. The zone graph for zone 1 is shown. Intersections, nodes, and gates
are shown with circles and marked accordingly. They constitute the vertices.
Fig. 4. An illustration of airways, intersections, and nodes Airways are shown with arrows and they are the edges of the graph. Most
likely, there are many gates between any two zones, but for simplicity we
show only two.

the border joining the airways. The graph that is formed by

treating both nodes and intersections (which include gates) as Zone 3 Inbound Outbound Zone 1
gate gate
the vertices and airways as the directed edges is called the zone
graph (Fig. 5). A path in the zone graph is called a pathway.
We use the word element to refer to airways, intersections, and
nodes. To be reachable, every element has a global address
similar to how hosts have a global address on the Internet. If
we take the gates as the vertices and connect co-zone gates Transit
with directed edges called transits, we call the resulting graph Zone 4 Zone 2
the interzone graph. Inside each zone, the cost of traveling
between any pair of gates is called the transit cost where the
cost can be time, distance, etc. (Fig 6). A path in this graph is Fig. 6. The interzone graph for the zones 1-4 is shown. Gates are the vertices
called a route. For the zone graph, we use the word progress of the graph. Transits as edges are representatives of the possibility of a trip
from the inbound gate to the outbound gate for the drones. Transit cost can be
within an airway or intersection to state how far the drone any cost function associated with the trip between two gates, such as average
has progressed the element according to some progress metric trip time. Between any two zones, there can be many gates, but for the sake
(e.g. distance from the beginning of an airway). In the zone or of simplicity we show only two.
interzone level, the vertices and edges contain meta data e.g.
in the form of components and attributes as in an XML tag
which provide data about the particular vertex or edge. Among 1) In each zone, any of the ZSPs provides navigation in-
the meta data is the minimum performance required from formation between any two elements in their designated
any drone that wishes to travel along the particular element, zone to the requesting drones. The license to operate a
such as drone range limitations, landing restrictions, and other specific zone is granted by higher authorities. They estab-
physical constraints. Meta data may also contain more detailed lish and enforce the governing laws regarding the airways,
information about a particular element; for example, the meta intersections and public nodes such as maximum allowed
data at a node representing a park can have a map of the drone capacity or density in them. Our architecture is not
park which a drone could use upon entry to the node. A concerned with how ZSPs are realized, but it is worth
portion of airspace is either public or private. All elements in mentioning that implementing a ZSP merely as software
public and private airspace are considered public and private seems conceivable. We call an organization that offers
respectively. For private elements, the access rules for drones ZSPs an Internet of Drone Service Provider (IoDSP).
is specified as meta data, such as which drones are allowed Adjacent ZSPs co-manage the gates and coordinate with
access to them. At the lowest level of abstraction, we deal each other on handoff; that is when a drone crosses the
with points in the airspace. The points are uniquely identified border and the responsibility has to be transferred to a
using the coordinate system of (latitude, longitude, altitude). new ZSP. Furthermore, ZSP can order a drone to land
For instance, an airway’s geometry is understood using points. or hold its position by hovering or executing holding
A path through points is called a trajectory. Beware that we patterns and we call these actions grounding and holding
do not use the term trajectory in the same way it is used in respectively. Fig. 7 presents a schematic of the Greater
robotics research where it means a time dependent path. Toronto Area (GTA) in Canada together with the ZSPs
deployed in the zones by four IoDSPs.
2) Drones in IoD are the autonomous aerial vehicles which
B. Components are capable of collision free navigation along a planned
Our architecture comprises of two groups of components: route between two nodes and have various performance
Zone Service Providers (ZSP) and drones. All ZSPs and drones characteristics, such as their range, whether they are
are connected to the cloud, so communication between any two capable of VTOL and hovering, etc. They broadcast
components is possible. information about their position and their future path

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 6

Application
Service
Toronto
End to End (E2E)
Node to Node (N2N)
Airspace
Mississauga
Fig. 8. Layers in the architecture of IoD

We tackle each of these tasks in a separate layer. The reason

this seems to be a good way of tackling navigation is that
Lake Ontario
having a single giant system with its map and airspace access
Hamilton
mechanism is computationally complex and unsustainable, if
10km
not impossible. By dividing the problem into smaller sub-
problems, each of them becomes more tractable. Therefore,
Fig. 7. A schematic of the Greater Toronto Area and the zones served by
the ZSPs deployed by four different IoDSPs each colored differently. Handoff we trade a more optimal solution for a more tractable solution.
occurs at the boundaries of the zones. As mentioned, there is more needed than just navigation. For
example, for a package delivery task requested by a grocery
store inside some zone, only the drones that are near the
which will be used by all ZSPs, not only the particular
store (say in the same zone) should be notified, not all the
one serving the drone. Regardless of how ZSPs and
drones in the realm of IoD. Hence, ZSPs must meet these
drones are implemented, they shall interact with each
zone-specific demands through a service layer that is used by
other through standard protocols. For instance, this allows
all applications. The service layer is extensible to meet the
that two competing firms have two different implemen-
needs for future services as they will become apparent by the
tations for their ZSPs and still different drones with
common needs of applications.
different implementations are able to communicate with
Our architecture consists of five layers as shown in Fig. 8.
the ZSPs through the standard protocols. Drones are
Drones have functions that fall in all the layers while ZSPs
required to assume fully autonomous operation beyond
only have functions that fall under the airspace layer up to
line of sight operation, be equipped with sense and
the service layer. In a strictly layered architecture, each layer
avoid technology and be capable of emergency landing.
provides services that are used by the layer directly above it.
Furthermore, specialized airworthiness certification must
As is the case with the Internet (see [22, pp. xvi,xx,xxi,33-
be considered to establish reliability levels for drones
36,87,147]), our architecture is a relaxed layered architecture
that are comparable to those of commercial aircraft when
where upper layers can access lower layers and not just the
operating over inhabited areas.
layers directly below them. In effect, layering provides an
effective way for logical organization of the architecture and
C. Layers its easy communication to other engineers and should not
Similar to the Internet, we propose a layered architecture for be treated as a never to be broken rule. The lower layers
IoD. Layering provides many benefits such as the separation are not aware of the specifications of the higher layers. The
of concerns, scalability, maintainability of the code base, and interactions between the layers shall be through standard
flexibility of modifying a layer with minimal changes needed interfaces. The protocols then are defined between the same
to the other layers. The fundamental goal that the architecture layers of two components.
is concerned with is to enable drones to perform various We describe each layer in terms of the features it is required
applications by providing common generic services for all to implement to comply with our architecture. This means
applications. Consequently, the architecture has two goals. that any IoD system must implement those features and define
Firstly, it is to provide guidance to a drone from a source specific protocols and interfaces that make access to those
node to a target node and coordinate all drones’ access to the features possible. We use capital letters as our convention for
airspace as a service to the drone. Secondly, it is to make the name of the features.
available an extensible platform for other common current or 1) Airspace Layer: The airspace layer is required to
future services that are needed by applications such as delivery implement the following features along with the needed
of messages that are intended for a pool of worker drones for protocols and interfaces for using these features.
an application in a specific zone (an example message is a list
of local task requests). MAP: ZSP is required to hold geometric representation of
The navigation can be reduced to three sub-tasks. Firstly, the elements in the zone graph; i.e. the airways, intersections,
the drone will have to traverse a path on the interzone graph and nodes.
from the source zone to the destination zone. Secondly, to
traverse within each zone, the drone must traverse a path on the AIRSPACE BROADCAST AND TRACK: Drones have
airways and intersections of the zone graph. Lastly, a trajectory to broadcast periodically their three dimensional coordinates
of points must be chosen which the drone has to follow to stay and their future trajectories. It is conceivable these data are
inside the boundaries of the airways, intersections, and nodes. needed for path planning in this layer and indirectly in other

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 7

layers for calculating the progress. example to landing sites which will be used in case the drone
cannot continue on its path, such as unexpected fuel shortage
PLAN TRAJECTORY: ZSP has to provide trajectories or when grounding by ZSP is necessary. ZSP has to take
to be followed by the drone, so it stays inside the boundaries into account the performance characteristics of the drones
of airways, intersections, and nodes of the planned pathway. among other things when allocating a pathway to a drone by
verifying the drone meets the minimum performance rating
AIRSPACE PRECISE CONTROL: We envision a possible for the paths. Also, various meta data for each element in
need for ZSP to request specific maneuvers from a drone the path can be disclosed to the drones such as the weather
such as holding, moving to a new point, or landing at a forecasts.
point. This seems to be a reasonable feature to expect from a
universal architecture. REFUEL: A path to a fuel station node (fuel station
can be third party depending on the preference of the drone)
COLLISION AVOIDANCE: In case of dynamic objects should be provided by ZSP to a drone that needs to refuel.
such as other drones or birds obstructing the airways or ZSP must direct the drones to fuel station nodes that are
intersections, the drone must avoid colliding with them by compatible with them. For example, drones can run on
overruling the trajectory. The drone must communicate with electricity, gas or even hydrogen (in case of fuel cell). When
other drones in proximity through standard protocols for a drone asks for refuelling, ZSP will give a pathway to the
coordinated maneuvers for avoiding collision. proper fuel station accordingly.

WEATHER CONDITION: ZSP must provide the drones with N2N PRECISE CONTROL: It must be possible for
the weather conditions such as wind speed and temperature, ZSP to command the drone to hold or to move to an element,
so drones can successfully take these data into account at the or to land at a node.
time of executing a trajectory.
EMERGENCY: When a drone faces a software or hardware
2) Node to Node layer: The features required for the node failure, if it is capable enough, it has to broadcast an SOS
to node layer is as follows. message to ZSP which must make arrangements such as
broadcasting relevant information to all the drones, so other
ZONE GRAPH: ZSP keeps an up to date zone graph drones change their pathway or hold in their current element.
that is augmented with the information broadcast from all the Furthermore, ZSP must detect when a drone abruptly stops
drones such as the current airway, intersection, or node of the broadcasting message and issue the emergency procedures.
drones and their future paths as well as their progress within
an airway or intersection. ZSP knows how many drones are CONGESTION NOTIFICATION: Upon request, ZSP
inside an element and roughly how they are spaced out in an must provide congestion report between any two elements
airway or intersection. In the zone graph, the meta data for inside the zone.
elements are stored too, such as the minimum performance
requirement which is also a function of the weather report and 3) End to End layer: The end to end layer must implement
changes in time. Furthermore, ZSP must provide protocols the following features.
for obtaining the information in the zone graph (e.g. for
viewing). Also, it must provide protocols for updating the INTERZONE GRAPH: ZSP must store the partial interzone
map, such as identifying certain airways, intersection, or even graph at the very least. That is, it must have information at
the complete zone as no fly areas. Also, ZSP must provide least about the the gates and transit costs in its zone (the other
protocols for integrating weather reports. end of spectrum is to have complete knowledge about the
interzone graph). This gives a partial or local knowledge of
N2N BROADCAST AND TRACK: Drones are required to the interzone graph which must be learned through different
broadcast their current element, their progress within it in means such as interaction with other zones, or input from
case of airways or intersections, and their future path, and administrator. The graph is augmented with the data broadcast
their estimated fuel time left periodically in a way that is from drones, so it is known which drones are inside the
accessible to all ZSPs. zone and of them which are inside the gates and which are
transiting between gates. A protocol must be implemented
PLAN PATHWAY AND CONTINGENCY: A path on for obtaining the data stored in the interzone graph (e.g. for
the zone graph must be provided by ZSP to a drone that viewing the graph).
requests a path between any two elements as source and
(intermediate) destination in the same zone. The path consists ROUTING: Any two adjacent zones are likely to have
of a sequence of airways and intersections that have to several gates connecting them. The ZSPs have to provide
be navigated for the drone to travel from the source to drones with one next intermediate gate. The transit cost can
destination. The path does not have to be complete and be used to provide a shorter route.
a partial path for getting closer to the destination is also
acceptable. A contingency path must also be provided for HANDOFF: Drones must be able to switch to the new

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 8

ZSP when entering a new adjacent zone. ZSP must be able private drones. Since airspace is a public space, all drones are
to handle the incoming and outgoing drones. required to be registered with the government for a license
to operate. Interestingly, at the time of writing this paper
EXPLICIT CONGESTION NOTIFICATION: ZSP has (December 16, 2015), FAA published an interim final rule (for
to give explicit congestion notification on any of its gates a definition, see [27]) that mandates owners of drones with a
and transits to at least the ZSPs in the adjacent zones. The weight between 250 grams to 25 kg to register it with the U.S.
algorithm to determine a gate or a transit is congested is up Department of Transportation [28].
to the implementation by the particular ZSP. The map of where zones are located and the public airways,
intersections, and nodes inside each zone is created by the
4) Service layer: The service layer is an extensible layer municipalities in consultation with FAA as it is the ultimate
that currently has the following mandatory feature and can aviation authority. Drone operations must be confined to inside
be extended to add more services in the future as needs arise. of these elements and this must be enforced by the police.
Furthermore, unauthorized entry of any drone to the private
ZONE BROADCAST : The main role of the service airways, intersections, and nodes is considered trespassing. Ar-
layer is to provide a common platform where zone-related eas that do not fall into any zones are considered unregulated.
messages can be broadcast to the drones. For instance, a task Private airspace can be defined in various ways. For exam-
request that needs to be performed by a drone in a particular ple, it can be the airspace directly above a private property
zone can be broadcast to all the drones in that zone through and below some elevation level. The municipalities set the
service layer in ZSP. A particular task can be grocery pick up boundaries of the private airspaces. As noted before, within
in a zone. Through encapsulation, the service layer does not private airspace, private elements are located. The owner of
understand the content of the message. However, applications the private airspace, if inclined, has to design his/her own
by relying on the service layer for receiving the message will map of these elements, according to the constraints set by
make sense of it. municipalities. The map is submitted to the municipality for
the purpose of integration with the city’s map along with
5) Application layer: There is no feature requirement for consents for releasing the map to one, two or even all IoDSPs.
the application layer. These are the applications that will Therefore, one possibility is that a private node be served
be written in the future to use the architecture. The point exclusively by a single IoDSP, a model similar to how a host
of having a general airspace navigation and control service is connected to the Internet using only one ISP. At the same
along with other services as is provided by the four layers of time any drone company can serve the node so that all the
airspace, N2N, E2E and service is that many application we drones are potentially available to the customer resulting in
can conceive of will use these services as a foundation. So by faster service time. Alternatively, all IoDSPs could provide
providing it once, we enable the whole range of applications the same services to every nodes and differentiate themselves
simultaneously, rather than providing a dedicated service to through better implementations of protocols.
each application. The non-exclusive license for IoDSPs to provide their
services within each zone is granted by the municipalities.
Airways, intersections, and nodes have to be used according
D. Cross-cutting features
to the policies set by the municipalities such as the maximum
Any feature listed here cannot be addressed by one single drone capacity or density. IoDSPs are obliged to provide
layer and needs to be implemented in several layers. service to all drones without discrimination (For example, an
IoDSP cannot deny service to a drone in retaliation to the
SECURITY: There are a variety of threats that must be drone using a different IoDSP in the previous zone). These
safeguarded against, among them are authentication of drones policies must be enforced by the municipalities and the police.
and ZSPs and other components outside the IoD system, More than one IoDSP can operate within the same zone. There
jamming of the broadcast messages, clogging the airspace, is no lower or upper limit on the number of zones within which
and hacking of the drones or ZSPs. a company can operate. Any IoDSP can serve any node, as
long as the private owner of the airspace has pre-authorized its
V. O PERATION MODEL access to the map through municipalities as mentioned above.
Our architecture can lend itself to various operation models. And finally, during handoff (i.e. when the drone enters a new
We discuss one seemingly reasonable model here and in zone and the responsibility must be transferred to a new ZSP),
remainder of the paper we assume that we have adopted this drones can choose any ZSP in the new zone.
model.
B. Interactions with outside
The interactions with outside are orthogonal to our archi-
A. Model tecture. To decide what entities will use the IoD system and
Public is the owner of the most of airspace. There are what protocols will be used is a design choice to be made at
two groups of drone owners. The first group are companies the time of implementation. However we mention some of the
operating fleets of drones and offering various services such as entities that in all likelihood will interact with the system to
logistics to users. The second group are individuals with their give a real world picture of how an IoD system might operate.

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 9

Individual Drone Owner NWS FAA VI. D ISCUSSION AND F UTURE W ORK
A. Goals, principles, and benefits of our design
In our design of the architecture, we have encouraged
principles of openness, modularity, and interoperability. To
Zone 1 Zone 2 achieve this, we require drones to broadcast their information
using standard open protocols to communicate with ZSPs
or other drones. Similarly we require interaction between
Zone 3
all ZSPs through standard protocols. We believe it is not
common or reasonable to expect competing ZSPs share traffic
information and other statistics. Hence, by requiring drones to
broadcast their current position and future path, we give all
ZSPs in the same zone a chance to manage the traffic and
have the big picture of the zone.
One immediate benefit of openness, modularity, and inter-
operability is the lower overall cost for creating the navigation
network. Similarly, there will be lower initial investment and
Store Drone Company Fuel Station lower complexity for the new IoDSPs or drone companies
Fig. 9. Drones and ZSPs are components inside the boundaries of the IoD
to enter the market. This leads to the organic growth of the
system as depicted by the box. Outside components such as fuel stations, network by lowering the barriers to entry and both new and
private or corporate drone owners, governmental organizations such as weather existing firms will benefit from the network effect. A key
services or FAA interact with drones or ZSPs through standard protocols. Solid
lines show some of the possible interactions.
consequence of these principles is that through innovation,
companies will compete in their implementations while they
coordinate on the standards. In our architecture, we have tried
An example IoD system implements protocols between the to require a minimal set of functionalities from IoD systems.
ZSPs and US National Weather Services (NWS) to disable Our intention is that this leaves the door open for introduction
and enable parts of the network in an automated way. The US of innovative protocols and algorithms rather than the ones
Federal Aviation Administration (FAA) might declare a no-fly forced by us.
zone which is communicated through another protocol to the One important aspect in our design is scalability and sur-
ZSPs. Other important entities are possibly third party fuel vivability. From the experience of the Internet, survivability
stations. There can be well-defined protocols for negotiating is a prerequisite for scalability. In an expansive IoD system,
between drones and third party fuel stations (if the drone opts failures will be common place and the IoD system must
for using them) with the possible role of ZSPs for brokering gracefully survive them. The IoD architecture comprises of
the messages. Also, for direct messages between a fuel station many autonomous subsystems that interact with each other
and drone at the time of docking, machine to machine (M2M) only locally. This makes it possible to contain failures as
protocols can be used due to the low latency that is required opposed to have them ripple through the entire system and
for the task. Since both drones and ZSPs are connected to make it unstable. For example, the design of our architecture
the cloud, users, companies owning the drones, administrators, promotes that only a small portion of a drone’s trip be reserved
retailers like grocery stores, etc., can communicate with them at any time by a local ZSP, since no ZSP has the authority to
through standard protocols like http (Fig. 9). reserve a path beyond its zone. On the contrary, if the entire
trip was reserved and for some reason the drone could not meet
its reservation, this would affect the whole system. Further-
C. Strategies for deployment more, by relying on autonomous subsystems, the complexity
A particularly attractive deployment strategy is the use of becomes manageable as the size of computational problems
already deployed cellular networks. As explained earlier, in that need to be solved will be substantially smaller.
the cellular network, each provider partitions its coverage area An important goal in our design is to provide generic ser-
into cells and places base stations in each of the zones. Since vices which can serve many diverse applications. Furthermore,
these base stations are already deployed, the physical space is applications that are not even conceived today are more likely
available and they are capable of running the ZSP software. to build on top of generic services than highly specialized
Therefore, they seem well positioned to implement ZSPs ones.
and provide wide network coverage for IoD. This strategy The concept of collision free network of airways and
becomes even more interesting considering that drones have intersections let us circumvent the high cost of 3D mapping
to use mobile communication which is basically what the of the terrain and the buildings in a city. Basically, instead of
base stations provide. This means that ZSPs will provide not guaranteeing a general statement that every possible trajectory
only navigation, but the main communication channel for the in an area is free of obstacles (or even worse not promise that
drones. Since base stations are connected to the cloud, ZSPs but require drones to avoid it as the only safety measure),
can communicate with each other or with other outside entities we guarantee one example trajectory in the area that is free of
over the cloud. obstacles (similar to the road networks). This is a substantially

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 10

easier task, as mathematicians can attest to when proving a such as e.g. in Asynchronous Transfer Mode (ATM) network
theorem versus providing an example for which the theorem technologies (see [31] for instance) or metaphorically we start
holds. In addition, this enables a higher control over where moving while asking (from the ZSPs) for direction. The latter
drones can and cannot be (such as near airports) which is is the approach the Internet takes and we believe this is the
important for safety and security, and noise control (such as superior approach for IoD systems. The main reason is that
near hospitals or residential units). Law enforcement will be since drones take a long time to complete their trips, reserving
possible when drones only operate through predicted routes, the entire path for them is a wasteful use of the airspace
as a course violation (such as trespassing in the private as precise prediction of the future position of a drone in a
airspace) will be easily detected. Furthermore, this results in a complex system like IoD is not possible. As a result, in the
more predictable traffic model which provides more organized IoD systems that we advocate, there is no guarantee a drone
data for planning, traffic management, and scheduling. It is will complete its trip without being occasionally grounded by
worth mentioning that in ATC, a standard model for avoiding ZSPs a few times along its path, due to a lack of enough
airborne collisions is the vertical separation of traffic according airspace. However, grounding a drone is expensive in terms
to direction of flights. In IoD, the model is that each airway and of energy consumption, travel delay and waste of airspace.
intersection has a specific direction of flight which is meant The situation is somewhat similar to the cellular network. In
to achieve the same purpose. both, it takes a long time for a cellphone user or a drone to
Due to the extreme scarcity of urban airspace and safety enter a new cell or zone, respectively. The idea in the cellular
critical nature of drone operations, we believe it is necessary network is that it is best to not admit a call, if it has to be
to highly regulate its use through a model such as collision dropped later. A similar policy in an IoD system is useful as
free network of airways and intersections as alluded to above. it is more expensive to ground a drone than to not let it get
As such, notions like free flight as described in the context airborne in the first place. This subject is studied extensively
of NextGen are unlikely to be practical at least in the urban in the area of call admission control (CAC) (See [32] for a
environment. At the same time we do not take the freedom survey). However, there are three major differences:
away completely as free flight is possible inside the nodes
1) In the cellular network, the scheme needed for reserving
by default with the extra flexibility for having collision free
resources is simpler. Basically one has to ensure that
maps and other information if needed as the meta-data for the
future cells have enough capacity to admit the mobile
nodes. Nodes can have any geometries which means they have
unit. However, in the case of reserving the zone-graph’s
no restriction on the size. Once the drone enters a node, the
elements for the drone, there is more than one way a
ZSP no longer provides a specific trajectory as in the case
drone can travel between any two nodes and hence there
of airways or intersections. ZSP will only provide the meta
is more complexity in deciding whether enough resources
data about that particular node to the drone. Hence a farming
are set aside for a particular drone or not.
drone is in the free flight mode inside a node representing
2) On the other hand, in the IoD setting, if nothing unex-
a farm. If on the other hand, ZSP’s help with navigation is
pected happens, the path a drone will take can be par-
needed for a node representing a large national park, the node
tially or completely known (depending on the particular
must be divided into multiple nodes each connected through
implementation of IoD system) whereas in the cellular
a network of airways and intersections, but again inside the
network it is often not known to which adjacent cell the
newly formed nodes, free flight is the mode of operation.
mobile unit will enter next. Hence, in the IoD, there is
Each drone in an IoD system will be capable of performing
less uncertainty over the path.
one or more applications. In the Internet a user demanding a
3) The CAC decisions are made centrally, partially moti-
service accesses a specific host on the network and interacts
vated by the fact that in the cellular network, the adjacent
with a specific application through unique IP and port num-
cells mostly belong to the same company. However, Dis-
bers. However, in the IoD the dominant model is not to make
tributed Call Admission Control (DCAC) is a possibility,
requests to a specific drone directly, rather a pool of drones
as shown by the seminal papers [33] and [34], where
be ready to accept these tasks broadcast by the ZSP’s service
reservations must be made not only in the current cell of
layer. This is analogous to the position of the information-
a mobile user, but also to a less extent in the neighbouring
centric networking line of research for the design of the future
cells and the cells beyond to accommodate the mobile
Internet (see [29] and [30] for recent surveys). The position
unit as it enters the new cells. If such a reservation is
is that users are mainly interested in the information rather
deemed possible after the base stations communicated
than the host to host connectivity. In our case, we advocate
with each other, then the call will be admitted. Referred to
that users are mainly interested in a service such as package
as the shadow clusters concept, it is similar to a quantum
delivery or power lines inspection, not the particular drone
wave function which maps the probability of finding an
that performs it or the particular path within which the drone
electron in any region in the space where electron is
travels to perform the task.
analogous to the mobile unit.
It is conceivable that a similar idea for IoD inspired by DCAC
B. Routing can provide a routing algorithm that grounds very few drones
A fundamental question about how we implement routing is while utilizing the airspace in an efficient way, by not reserving
whether we reserve the entire path before the start of the trip the entire path from the source node to the destination node

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 11

for them. Otherwise, prediction errors will ripple through the A mechanism that exists in the Internet literature to avoid
whole system and make it unstable [23]. As noted above, the overloading a link is a token bucket scheme (for example see
routing task will be harder in IoD because of the complex [36]) in which tokens simply represent resources and each
structure of the resources, but the lower uncertainty over party is given a token, only if there is a token left. However,
the drone’s path can be useful. Developing such a routing it is not clear how such a mechanism would work for an IoD
algorithm is an important contribution to IoD. system as there are more than one ZSPs which can grant access
to the same element, and being competitors, it is reasonable
to assume they will not share information.
C. Congestion control
An IoD system must achieve fairness in allocation of the
With the possibility of thousands of drones at flight at any airspace. However, fairness is a subjective term and can lead
point in time in an urban environment, a main purpose of the to different designs depending on how the fairness is defined.
IoD architecture is to coordinate access to the airspace. It is Should we give more priority to the faster drones at the
instructive to first discuss how the congestion control in the expense of slower ones, since they use the airspace for a
Internet works. The goal in the Internet is to ensure efficient shorter period or should we allocate the airspace to each drone
and fair use of bandwidth. There is no central mechanism that in an equitable way? In IoD, similar to the Internet, related
in the short run allocates bandwidth to each of the hosts, i.e. to the question of fairness is a design that takes into account
the end nodes which are the users of the network. Rather, hosts the Quality of Service (QoS), i.e. the network performance
allocate a fair and efficient amount of bandwidth to themselves according to various metrics. The interesting fact is that not
in a participatory fashion. They do this by probing the network all the applications have the same needs. For example, a drone
and refraining to add more loads to it if they realize the that surveys the traffic has to stay aloft for extended time where
network is in a congested state. This is done by analyzing short interruptions are not necessarily important whereas a
the amount of time it takes for the delivery acknowledgment drone that delivers a package needs the airspace for a short
(ACK) to be received by the sender (if ever in case of a period of time and has to minimize its delivery time to meet
dropped packet). To probe the network in a decentralized customers’ demands.
way, the network is driven toward congestion which creates
delayed or lost packets which results in delayed or unsent D. Communication signalling
ACKs respectively. From this, the sender realizes it must slow
Since drones are wireless and ZSPs have to broadcast, there
down in sending more packet until the network becomes less
will be a high amount of communication signalling which can
congested. This is an implicit way of inferring congestion.
flood the allocated frequency channels. IoD protocols must
Today, some of the routers in the middle of the Internet are
be designed with respect to the channel capacities as well as
capable of sending Explicit Congestion Notification (ECN)
the number of drones and ZSP that will use the channel. If
[35] by looking at the number of packets that they have in the
a high signalling overhead is inevitable for the functioning of
queue that are not yet sent. This is a helpful feature, because
IoD, then communication channels must also be treated as a
running a network in a congested mode is not efficient;
resource similar to how airways and intersection are treated.
something that we have to do when the network does not
Therefore, for the purpose of reserving the airspace for the
provide feedback, just to be able to implicitly infer congestion.
drone, communication channels should be reserved as well
The congestion status must be known in IoD within each
and if any of these resources are not available a reservation
of ZSPs for two reasons. First, running a congested airspace
should be deemed not possible.
translates into grounding or holding which are both expensive
operations. Second, in the Internet, the ACKs happen on the
orders of few hundreds of milliseconds. This fast feedback E. Addressing schemes
loop allows implicit congestion probing as a viable option. In Similar to the zone graph elements, drones are in need
IoD, probing directly with drones (i.e. by seeing if drones get of global addressing. Whereas airways, intersections, and
stuck in the congestion or not) is orders of magnitude slower. nodes as well as ZSPs are stationary, drones are mobile.
Because of especially high cost of congestion for the IoD, we Hierarchical addressing schemes similar to telephone numbers
believe we have to require a feature in the IoD architecture or IP addresses can prove useful for the zone graph elements
for explicit congestion notifications to at least the neighbouring or ZSPs. However, a particular shortcoming of the current
zones in the E2E level. This is not needed in the N2N level, Internet is that when IP was designed, it was assumed that it
as any ZSP has complete knowledge of the congestion on all will work with stationary units. However, with the proliferation
the airways, intersections, and nodes due to the broadcasts of mobile devices, that assumption is no longer valid. It
from the drones. A major difference with ATC is that there seemed reasonable at the time the Internet was designed to
is no central controller for the whole network (ATCSCC) that have IP address serve two purposes; i.e. identification and
regulates the load on the whole network while each ARTCC localization. Identification is achieved by requiring every host
only ensures separation, which would have a negative effect to have a unique IP address. Localization is achieved by
on scalability. separating the IP addresses into a network portion and a host
In the design of a congestion control algorithm, it is an open portion where each network can be part of a larger network;
research question how to achieve a fair and efficient allocation an idea referred to as subnetting. This design choice results in
of the airspace while not overloading any of the elements. poor performance when the hosts are mobile [22]. Therefore,

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 12

any addressing scheme for drones should perhaps separate iterations of the architecture. To implement an IoD system
these two functions in some form as is the case with most there are many non-programming questions that have to be
solutions to mobility on the Internet including Mobile IP and answered, such as the questions discussed about routing and
IPv6. A particularly interesting choice would be geographical congestion control. This will be the main area of our focus
addressing [37] where each drone is assigned an evolving in the future works. Building IoD is a great undertaking
address according to its current geographical position. For which needs the participation of the research community at
instance, this can provide a finer control over which drones large. By presenting the architecture in the current stage,
to dispatch for a local task in a zone. useful protocols can be discussed and designed by the research
community which can be validated once a simulation as well
F. Drones and minimum performance as a physical platform for IoD is ready. Furthermore, the
design of the IoD architecture itself can benefit from the work
In an IoD system, an important ability that might be man- of the researchers working on diverse range of networks from
dated by authorities in high traffic areas such as lower altitude air traffic control to cellular to the Internet who will apply
in the urban airspace is the VTOL ability which enables easier their knowledge to IoD.
grounding or holding by hovering. This can mean that most
of the urban airways, intersections, and nodes in the lower
altitude may require VTOL whereas in higher urban altitude I. Economics of IoD
it may not be required. This is because VTOL drones are From an economic point of view, the operation model and
highly versatile and can perform tasks in an environment the protocols of the system must provide enough incentives to
with very little airspace available to them. Most commercial the stakeholders to pursue the desired actions. It is interesting
aircraft are each equipped with on-board systems like Traffic to study related questions through the lens of game theory and
Alert and Collision Avoidance System (TCAS) which are mechanism design.
designed to decrease the chance of mid-air collision (see [38]
for instance). It is a complex system and just to avoid collision
between two aircraft, thousands of lines of code are needed. J. Legislation
In our case of lower altitude urban airspace, it is reasonable Another major topic is to provide a legal framework for the
to assume that in often congested area with thousands of IoD. One of the main barriers in utilizing the drones today
drones in flight, to avoid mid-air-collision, aircraft must be is lack of legislation that properly address the technology.
able to hover and move vertically to regulate the traffic, similar This is manifested in the recent Public Law 112-95 titled
to the road network and cars which can stop. Drones are “FAA modernization and reform act of 2012” [26] enacted by
ultimately responsible for avoiding collisions mid-air and a US House of Representatives. In the Public Law 112-95, the
TCAS like system for drones without hovering abilities is a secretary of Transportation is mandated among other things
major challenge for more than two drones. firstly to develop a comprehensive plan for expediting the
integration of civil Unmanned Aircraft Systems (UAS) into the
G. Security national airspace system, and secondly create a 5 year roadmap
Security is not a topic that can be addressed by any single for their introduction. Thirdly, specifically for small unmanned
layer. A major challenge in the Internet today is that security is aircraft system, a rulemaking was required that would expedite
mostly provided by the application layer and there is a lack of the start of their civilian operation in national airspace system.
in-place security mechanism in the lower layers. The Internet In response, in 2013, FAA along with other governmental
has been exploited for its security vulnerabilities which have agencies jointly published a comprehensive plan for inte-
led some network researchers to consider the security as one of gration of UAS into national airspace system [25]. In this
the main goals in the next architectures for the Internet [39]– document, UAS national goals and objectives are described.
[41]. Arguably, damages from malicious users are more severe One of the goals is to make civil visual-line-of-sight operation
in the case of IoD compared to the Internet and security must of small UAS a routine by 2015. Initially this will be outside
be one of the core issues that any architecture for IoD should of class B and C airspace and above urban areas. In accordance
address. Given the experience from the Internet, we required with this goal, in February 2015, FAA published a notice
in our architecture that security be implemented across all the of proposed rulemaking [42] that addresses introduction of
layers, as it is a cross-cutting concern. small UAS (i.e. weighing less than 25kg) into national airspace
system. Various safety measures have been proposed such as
visual-line-of-sight operations. Flights are restricted to day
H. Validation and technical implementation time at a maximum altitude of 152.4m above the ground. Small
This paper presents a conceptual architecture and the grand UAS cannot operate in class A airspace. However, operation
technical contribution is to instantiate at least one system within class B,C,D, and E airspace is possible with permission
based on it to validate and demonstrate that our architecture from ATC. Furthermore, operation in class G airspace does not
can work in practice. This entails designing protocol suites require a permission from ATC. As mentioned before, for UAS
and interfaces between the layers and implementing the layers that weighs more than 250 grams, the owner has to register it
with the required features. Any inconsistency or inefficiency with FAA [28] for outdoor operation. Another goal set forth
revealed at the time of implementation can be used for later in the comprehensive plan is to make routine operation of

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 13

UAS possible in the national airspace by 2015 for the public work that is only viable if these new classes are introduced.
organizations and by 2020 for the civilians. According to [43], the FAA provides a transparent process
To comply with the public law [26], FAA has published a for setting regulations which encourages comments from the
roadmap for integration of UAS in the national airspace system public as well as other feedback mechanisms for avoiding
[43]. Currently for UAS to access the airspace, Certificates of onerous regulations. In the process of crafting new legisla-
Waiver or Authorization (COA) are needed for public oper- tion, the FAA has been soliciting feedback from the UAS
ation and certain airworthiness certificates for experimental community with one example being creation of the Advisory
civil application as mandated in [26]. Initially FAA plans to and Rulemaking Committees(ARC) for UAS comprising of
accommodate UAS in the near-term (next 5 years), then it members from industry and academia [44]. We are optimistic
transitions into the period of integration (5-10 years) in the that the stakeholders will influence the process in a way that
mid-term and in the long- term (more than 10 years) it is new airspace classes are created for UAS rather than what we
expected that requirements from UAS will evolve based on the believe is fitting a fundamentally new technology into a frame
safety requirements from all type of aircraft and is consistent that was designed for a different technology.
with the timeline for NextGen vision. FAA asserts that for
UAS to be allowed access to the national airspace, they must VII. C ONCLUSION
be able to apply and be accepted for standard airworthiness Many drone applications can benefit from a unified frame-
certificate. work that coordinates their access to the airspace and helps
A challenging goal for the FAA is to integrate UAS without them navigate to the points of interest where they have to
segregating various types of aircraft. Two important required perform a task. Any architecture poised to provide this service
technologies according to the FAA’s roadmap is Sense and must be scalable and be able to provide it to thousands of
Avoid (SAA) and Control and Communication (C2). The drones, which will share the congested and limited urban
SAA is expected to ensure self-separation and at a later airspace.
stage collision avoidance which needs to be interoperable with In this paper, we laid out the conceptual foundation for
other collision avoidance systems as well as compatible with such an architecture by developing a vocabulary of concepts
ATC separation services. According to the FAA, third party- for describing the architecture and identifying the relevant
communication service providers are used frequently today components of it as well as deciding on the boundaries of
and it is a routine task for FAA to effectively monitor their the architecture. Furthermore, we designed a structure for the
performance. The choice of the right type of third party C2 airspace and provided strategies for utilizing that structure in
providers is dependent on the choice of UAS architecture. At the airspace. Our design makes it possible to provide generic
International Telecommunication Unions World Radiocommu- services that can be used by many applications. To effectively
nication Conference in 2012, an agreement was reached to tackle the problem of “how to enable drones to perform
dedicate a part of frequency spectrum for exclusive use by tasks”, we divided the overall required functionality of the
UAS. This paves the way for the operation of UAS across architecture into logical layers. The main sub-problem was the
international borders and protects UAS from interference from airspace navigation and coordination for various applications
other devices. [43] as addressed in the first three layers of IoD. We addressed
According to the roadmap [43], the FAA asserts that unless other common services that are needed by applications such as
new classes of airspace are specifically created for UAS, for location aware communication in an extensible service layer.
them to be accepted for integration in the national airspace In IoD architecture, we describe the features that are required
system, they must satisfy the following requirements from to be implemented in each of these layers by IoD systems.
FAA (with notable exception of line-of-sight small UAS). In Furthermore, we suggested an operation model that identifies
addition to airworthiness certificates alluded to above, any the role of private and public organizations in the governance
UAS must register and execute an IFR flight plan (see [18] for of IoD. Additionally, we explored and discussed some of the
a definition) and be equipped with ADS-B (Out); i.e. the ADS difficulties that have to be addressed for an effective IoD
broadcasting component. Furthermore, they have to meet the system. In all of this, we used and referred to the wealth
minimum performance and equipage requirement of the area of knowledge acquired from three large scale networks, the
where the operation takes place. Additionally, each UAS must cellular network, air traffic control, and the Internet. Finally,
have a flight crew including a pilot-in-command who is only we discussed the differences and future works that can benefit
in charge of only one UAV and fully autonomous operations from the solutions from the vast existing literature on these
will not be allowed. Also, minimum required separation must three subjects.
be met in the controlled airspace and ATC will be in charge
for separation services for the applicable airspace classes for ACKNOWLEDGMENT
manned and unmanned aircraft. We would like to thank Yashar Ganjali for fruitful discus-
In our opinion, the ban on the fully autonomous operation sions on the subject.
set forth by the FAA in their roadmap takes away the major
benefits of any drone architecture, including IoD. Fortunately, R EFERENCES
the FAA does not rule out the introduction of new classes
[1] R. D’Andrea, “Guest editorial can drones deliver?” Automation Science
specifically designated for UAS in their roadmap as mentioned and Engineering, IEEE Transactions on, vol. 11, no. 3, pp. 647–648,
above. Certainly, IoD in its current form is a theoretical frame- 2014.

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 14

[2] D. Gross. (2013) Amazon’s drone delivery: How would it work? [30] G. Xylomenos, C. N. Ververidis, V. Siris, N. Fotiou, C. Tsilopoulos,
[Online]. Available: http://www.cnn.com/2013/12/02/tech/innovation/ X. Vasilakos, K. V. Katsaros, G. C. Polyzos et al., “A survey of
amazon-drones-questions/ information-centric networking research,” Communications Surveys &
[3] FedEx Corporation. (2015) Q1 fiscal 2015 statistics. Tutorials, IEEE, vol. 16, no. 2, pp. 1024–1049, 2014.
[Online]. Available: http://investors.fedex.com/files/doc downloads/ [31] R. O. Onvural, Asynchronous transfer mode networks: performance
statistical/FedEx-Q1-FY15-Stat-Book v001 t195uu.pdf issues, 2nd ed. Norwood, MA: Artech House, 1994.
[4] A. Raptopoulos. (2013) No roads? there is a drone for that. [32] M. Ghaderi and R. Boutaba, “Call admission control in mobile cellu-
[Online]. Available: https://www.ted.com/talks/andreas raptopoulos lar networks: a comprehensive survey,” Wireless communications and
no roads there s a drone for that mobile computing, vol. 6, no. 1, pp. 69–94, 2006.
[5] Amazon.com Inc. (2013) Amazon prime air. [Online]. Available: [33] D. Levine, I. F. Akyildiz, M. Naghshineh et al., “A resource estimation
http://www.amazon.com/b?node=8037720011 and call admission algorithm for wireless multimedia networks using
[6] J. Stewart. (2014) Google tests drone deliveries in project wing trials. the shadow cluster concept,” Networking, IEEE/ACM Transactions on,
[Online]. Available: http://www.bbc.com/news/technology-28964260 vol. 5, no. 1, pp. 1–12, 1997.
[7] DHL Express. (2014) DHL parcelcopter launches ini- [34] M. Naghshineh and M. Schwartz, “Distributed call admission control
tial operations for research purposes. [Online]. Avail- in mobile/wireless networks,” Selected Areas in Communications, IEEE
able: http://www.dhl.com/en/press/releases/releases 2014/group/dhl Journal on, vol. 14, no. 4, pp. 711–717, 1996.
parcelcopter launches initial operations for research purposes.html [35] K. Ramakrishnan and S. Floyd, “A proposal to add explicit congestion
[8] NASA. (2015) NASA UTM 2015: The next era of aviation. [Online]. notification (ECN) to IP,” RFC 2481, 1999.
Available: http://utm.arc.nasa.gov/utm2015.shtml [36] S. Shenker and J. Wroclawski, “General characterization parameters for
[9] ——. (2015) UTM fact sheet. [Online]. Available: http://utm.arc.nasa. integrated service network elements,” RFC 2215, 1997.
gov/docs/UTM-Fact-Sheet.pdf [37] M. L. Sichitiu and M. Kihl, “Inter-vehicle communication systems: a
[10] P. Kopardekar. (2015) Safely enabling UAS operations in low- survey,” Communications Surveys & Tutorials, IEEE, vol. 10, no. 2, pp.
altitude airspace. [Online]. Available: http://utm.arc.nasa.gov/docs/ 88–105, 2008.
pk-final-utm2015.pdf [38] FAA. (2011) Introduction to TCAS II version 7.1. [On-
[11] Amazon.com Inc. (2015) Determining safe access with a line]. Available: http://www.faa.gov/documentLibrary/media/Advisory
best-equipped, best-served model for small unmanned aircraft Circular/TCAS%20II%20V7.1%20Intro%20booklet.pdf
systems. [Online]. Available: http://utm.arc.nasa.gov/docs/Amazon [39] A. Feldmann, “Internet clean-slate design: what and why?” ACM SIG-
Determining%20Safe%20Access%20with%20a%20Best-Equipped, COMM Computer Communication Review, vol. 37, no. 3, pp. 59–64,
%20Best-Served%20Model%20for%20sUAS[2].pdf 2007.
[12] ——. (2015) Revising the airspace model for the safe [40] J. Roberts, “The clean-slate approach to future Internet design: a survey
integration of small unmanned aircraft systems. [Online]. Available: of research initiatives,” Annals of telecommunications, vol. 64, no. 5,
http://utm.arc.nasa.gov/docs/Amazon Revising%20the%20Airspace% pp. 271–276, 2009.
20Model%20for%20the%20Safe%20Integration%20of%20sUAS[6].pdf [41] S. M. Bellovin, D. D. Clark, A. Perrig, and D. Song. (2006)
[13] Google Inc. (2015) Google UAS airspace sys- A clean-slate design for the next-generation secure internet.
tem overview. [Online]. Available: http://utm.arc.nasa.gov/docs/ [Online]. Available: http://groups.geni.net/geni/raw-attachment/wiki/
GoogleUASAirspaceSystemOverview5pager[1].pdf OldGPGDesignDocuments/GDD-05-05.pdf
[14] M. Gharibi, R. Boutaba, and S. L. Waslander, “Internet of drones,” [42] FAA. (2015) Operation and certification of
arXiv:1601.01289, 2016. small unmanned aircraft systems. [Online]. Avail-
[15] S. Devasia and A. Lee, “A scalable low-cost-uav traffic network (unet),” able: http://www.faa.gov/regulations policies/rulemaking/recently
arXiv:1601.01952v2, 2016. published/media/2120-AJ60 NPRM 2-15-2015 joint signature.pdf
[16] C. J. Kim. (2006) TMA integrated metrics assessment model. [43] ——. (2013) Integration of Civil Unmanned Aircraft Systems (UAS)
[Online]. Available: http://www.tc.faa.gov/LOGISTICS/GRANTS/pdf/ in the National Airspace System (NAS) roadmap. [Online]. Available:
2004/04-G-044.pdf/FINAL%20REPORT%2004-G-044.pdf http://www.faa.gov/uas/media/UAS Roadmap 2013.pdf
[17] US Government publishing office. (2010) Automatic dependent [44] ——. (2011) Unmanned aircraft systems aviation rulemaking
surveillance broadcast (ADS-B) out performance requirements to committee. [Online]. Available: http://www.faa.gov/regulations policies/
support air traffic control (ATC) service. [Online]. Available: rulemaking/committees/documents/media/UASARC-20110617.PDF
http://www.gpo.gov/fdsys/pkg/FR-2010-05-28/pdf/2010-12645.pdf
[18] M. Nolan, Fundamentals of air traffic control, 5th ed. Clifton Park,
NY: Delmar Cengage Learning, 2010.
[19] T. S. Rappaport et al., Wireless communications: principles and practice,
2nd ed. Upper Saddle River, NJ: Prentice Hall, 2002.
[20] A. Goldsmith, Wireless communications. Cambridge, United Kingdom:
Cambridge university press, 2005.
[21] W. Stallings, Wireless communications & networks, 2nd ed. New York
City, NY: Pearson Prentice Hall, 2005.
[22] L. L. Peterson and B. S. Davie, Computer Networks: A Systems
Approach, 5th ed. Burlington, MA: Morgan Kaufmann, 2011.
[23] R. Bush and D. Meyer, “Some Internet architectural guidelines and
philosophy,” RFC 3439, 2002.
[24] D. Clark, “The design philosophy of the DARPA Internet protocols,”
ACM SIGCOMM Computer Communication Review, vol. 18, no. 4, pp.
106–114, 1988.
[25] FAA. (2013) Unmanned Aircraft Systems (UAS) comprehensive
plan: A report on the nation’s UAS path forward. [Online].
Available: http://www.faa.gov/about/office org/headquarters offices/agi/
reports/media/UAS Comprehensive Plan.pdf
[26] H. of Representatives. (2012) FAA modernization and reform
act of 2012. [Online]. Available: http://www.gpo.gov/fdsys/pkg/
CRPT-112hrpt381/pdf/CRPT-112hrpt381.pdf
[27] Office of the Federal Register. (2011) A guide to the rulemaking
process. [Online]. Available: https://www.federalregister.gov/uploads/
2011/01/the rulemaking process.pdf
[28] FAA. (2015) Registration and marking requirements for small unmanned
aircraft. [Online]. Available: https://federalregister.gov/a/2015-31750
[29] B. Ahlgren, C. Dannewitz, C. Imbrenda, D. Kutscher, and B. Ohlman, “A
survey of information-centric networking,” Communications Magazine,
IEEE, vol. 50, no. 7, pp. 26–36, 2012.

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2016.2537208, IEEE Access

SUBMITTED TO IEEE ACCESS, JANUARY 2016 15

Mirmojtaba Gharibi received the B.ASc. in elec-

trical engineering from Sharif University of Tech-
nology. He completed his M.Math, and currently is
pursuing a PhD degree both in Computer Science at
University of Waterloo, Waterloo, Canada.

Raouf Boutaba received the M.Sc. and Ph.D. de-

grees in Computer Science from the University
Pierre & Marie Curie, Paris, in 1990 and 1994,
respectively. He is currently a professor of Computer
Science at the University of Waterloo. His research
interests include resource and service management
in networks and distributed systems. He is the found-
ing editor in chief of the IEEE Transactions on
Network and Service Management (20072010) and
on the editorial boards of several other journals. He
has received several best paper awards and other
recognitions such as the Premiers Research Excellence Award, the IEEE
Hal Sobol, Fred W. Ellersick, Joe LociCero, Dan Stokesbury, Salah Aidarous
Awards, and the McNaughton Gold Medal. He is a fellow of the IEEE, the
Engineering Institute of Canada, and the Canadian Academy of Engineering.

Steven L. Waslander received the B.Sc.E degree in

Applied Mathematics and Mechanical Engineering
from Queens University, Kingston, ON, Canada, in
1998, the M.S. in Aeronautics and Astronautics from
Stanford University in 2002 and the Ph.D. in Aero-
nautics and Astronautics from Stanford University
in 2007. He is currently an Associate Professor with
the Department of Mechanical and Mechatronics
Engineering, University of Waterloo, Waterloo, ON.
He is also the Director of the Waterloo Autonomous
Vehicles Laboratory. His main research interests
include perception, navigation and control of autonomous aerial rotorcraft
and ground rovers with a focus on simultaneous localization and mapping,
optimal motion planning and multi-robot coordination.

2169-3536 (c) 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.