Control of Communication Networks For Teams of UAVs Handbook of Unmanned Aerial Vehicles

Control of Communication Networks
for Teams of UAVs 66

Andrew Kopeikin, Sameera S. Ponda, and Jonathan P. How
Contents
66.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1620
66.2 Design Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1622
66.3 Communications and Networking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1625
66.3.1 System and Network Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1625
66.3.2 Wireless Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1626
66.3.3 Network Topology and Consensus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1629
66.3.4 Information Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1630
66.4 Coordination Under Communication Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1632
66.4.1 Information Consensus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1632
66.4.2 Effects on Cooperative Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1634
66.4.3 Effects on Cooperative Decision Making . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1636
66.5 Motion Planning for Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1638
66.5.1 Graph Theoretic Methods and Potential Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1638
66.5.2 Reactive and Adaptive Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1640
66.5.3 Implementation Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1641
66.6 Communication Relays for Network Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1643
66.6.1 Controlling Relay Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1643
66.6.2 Deploying Relays to Support a Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1645
66.6.3 Controlling Communications Through Task Allocation . . . . . . . . . . . . . . . . . . . . . . . . . 1647
66.7 Concluding Remarks and Remaining Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1648
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1650
Abstract
The communication network is a fundamental component of a multi-UAV
system. It enables exchanges in command and control messages and allows
remotely sensed mission data to be sent to processing centers. Proper control
of the network, the topic of this chapter, is critical for the system to function
A. Kopeikin () • S.S. Ponda • J.P. How

Department of Aeronautics and Astronautics, Aerospace Controls Laboratory Massachusetts
Institute of Technology, Cambridge, MA, USA
e-mail: kopeikin@MIT.EDU; sponda@alum.mit.edu; jhow@mit.edu
K.P. Valavanis, G.J. Vachtsevanos (eds.), Handbook of Unmanned Aerial Vehicles, 1619
DOI 10.1007/978-90-481-9707-1 19,
© Springer Science+Business Media Dordrecht 2015
1620 A. Kopeikin et al.
properly. Following a brief overview, Sect. 66.2 lists some of the main challenges
in network communication control. Next, Sect. 66.3 provides background dis-
cussion in wireless networking including system architecture, wireless channel
performance, topology models, and information routing. Bounds and limitations
in cooperative control and decision making because of network limitations
are provided in Sect. 66.4. Next, the chapter provides several network control
strategies for multi-UAV systems. First, Sect. 66.5 presents motion planning
methods to control the topology, and second, Sect. 66.6 describes relay deploy-
ment techniques to extend the performance of the network. Finally, Sect. 66.7
provides some closing remarks and directs the reader to further literature on this
topic.
66.1 Overview
Network and communication systems such as the Internet, social media, cellphones,
and countless other wireless systems have revolutionized human way of life.
These same technologies enable teams of multiple human and unmanned agents
to share situational awareness and planning strategies to cooperate and improve
their performance in executing a set of mission tasks. As such, the communication
network is a fundamental component of a multiple unmanned aerial vehicle
(UAV) system. Controlling it is often just as important as controlling the vehicles
because inadequate communications can significantly degrade team performance.
Even though ad hoc networks have been actively studied in the past decades for
military (Leiner et al. 1996), home (Negus et al. 2000), and metropolitan (Ritter et al.
2001) network applications, they have only recently started to generate significant
interest for cooperative control of multiple robotic air, ground, and sea vehicles.
There are several general architectures to deploy multiple UAVs as a team. UAVs
may be connected only to a central node, such as a human-operated control station,
which receives information, computes a plan of action, and sends instructions
specific to each vehicle (centralized control). Alternately, UAVs may be connected
directly to each other, exchange information, and each formulate their own plan
without a designated leader (decentralized control) (Stipanović et al. 2004; Choi
et al. 2009). Hybrid centralized and decentralized solutions also exist, and UAVs
may also cooperate with other manned and unmanned space, air, ground, and sea
agents over the network in a joint effort (Chandler et al. 2001; Chandler and Pachter
2001; Marshall et al. 2011) as illustrated in Fig. 66.1. In all cases, team coordination
requires agents to exchange their state information, observations of the world, and
control decisions such as task allocation or motion planning. Furthermore, if the
UAVs are being used to gather information, as is often the case, the collected data
may need to be communicated to a designated point for analysis, possibly in real
time (Office of the Secretary of Defense 2009).
Because UAVs are highly mobile vehicles, information is most commonly
exchanged across the network using wireless communication. Signals containing
encoded messages travel between transmitting and receiving radio modules over
wireless channels. The quality of the channel is fundamentally based on the strength
66 Control of Communication Networks for Teams of UAVs 1621
UAV Comm. Relay Manned

Aircraft
UAV Team
Data Analysis
Center
Cooperative Search
Mission
UGV Team
Fig. 66.1 A networked multi-UAV team performing a cooperative search mission. UAVs are seen
cooperating with ground vehicles, and a communication relay UAV connects the team to the base
as well as to other manned aircraft involved in the mission
of the signal at the receiver compared to noise and interference in the environment,
a value named the signal to interference and noise ratio (SINR) (Goldsmith 2005).
The quality of the link drives the probability that information transmitted will
successfully be received, and affects the rate at which information can be exchanged
over that channel.
One typical objective in controlling a multi-UAV team is to ensure that the
system remains in a state where the communication network can support cooperative
planning (or coordination) and data collection requirements. The network itself
can be controlled through its topology, which describes the set of interconnections
between nodes and how strong those connections are. There are several general
methods to control the topology in a team of mobile agents:
1. Wireless channels generally degrade with increasing distance and due to obsta-
cles in the line of sight. Cooperative motion planning can position the vehicles
in locations which improve data transfers (Goldsmith 2005).
2. The received signal can be improved by increasing transmission power (Santi
2005), changing the orientation or directing the transmitter (Ahmed et al. 2011),
or choosing the proper time and location to transmit (Yan and Mostofi 2010).
3. The topology can be controlled by actively choosing which links should be
active and which exchanges are better suited for multiple hop node-to-node
transfers (Zavlanos and Pappas 2008).
4. Additional mobile agents can be deployed as communication relays to extend
the performance of the underlying network (Zhu et al. 2009).
These control methods can be employed individually or in tandem to achieve
the performance required in the communication network. The design trade must
properly consider the benefits, costs, and implications in computation, communica-

tion, and vehicle resources associated with each of these methods.
The concept of multi-UAV teams is still fairly new operationally. As such, the
methods to control communication in these systems are actively evolving through
research studies (discussed in this chapter) and lessons learned from other types
of networks. Aerial networking itself has been employed for years in commercial
and military aircraft operations (Mulkerin and Zernic 2003). Modern generation
strike aircraft and aerial intelligence platforms commonly share mission parameters
among themselves via Tactical Data Links for real-time cooperation. These systems
however are limited in range, throughput, and interoperability and exhibit delays.
Because information sharing has become so crucial in warfare, large-scale UAVs
are being deployed as persistent communication relays to better connect aircraft and
ground nodes (Jonson et al. 2008). Similar concepts are under active investigation to
rapidly provide communications for large-scale disaster recovery scenarios similar
to Hurricane Katrina (Federal Communications Commission 2011) or first response
in challenging environments such as wild land fire fighting (Zajkowski et al. 2006).
It is paramount to understand how best to employ these systems to maximize
their network support. Many other multi-unmanned vehicle concepts are being
considered for military (U.S. Air Force Chief Scientist AF/ST 2010), exploration
(Anderson et al. 2003), and surveying (Leonard et al. 2010) applications which will
also implicitly require control of communications to operate.
The purpose of this chapter is to provide an overview of the different ele-
ments of network communication control for teams of UAVs. It describes the
challenges involved in controlling communications in such multi-agent systems
and provides technical fundamentals in wireless communications theory as support.
The chapter analyzes recent literature studies about the effects of limited interagent
communications on the performance of team cooperation. Furthermore, it reviews
recently developed techniques to control network communications in a team using
motion control, communication relays, and cooperative task allocation strategies.
The objective is to familiarize the reader with this relatively new and rapidly
evolving topic.
66.2 Design Challenges
There are numerous and often competing challenges in controlling network com-
munications in a multi-UAV system. The design must properly balance between
ensuring that the communication requirements are met without over-constraining
the system design and operation. Furthermore, even if designed and controlled
properly, network communications still exhibit limitations and uncertainties in
performance which impact the system.
The control of network communications follows the same feedback control
principles used for many other types of systems. The state, or topology of the
network, is sensed and used as feedback, with a model of its dynamics, to predict the
outcome of control inputs. Modeling and accurately predicting the performance of
wireless channels is a significant challenge in this process. In general, links degrade

with increasing internode distance due to path loss. Signal shadowing occurs due
to obstacles in the environment such as buildings, mountains, or even the platform
vehicle itself. Furthermore, multipath replicas of the signal reflecting and scattering
off these obstacles constructively or destructively interfere with each other at the
receiver (see Fig. 66.3). Path loss, shadowing, and multipath fading effects are
increasingly detrimental and difficult to predict in more congested operating envi-
ronments (Santi 2005; Goldsmith 2005; Mostofi et al. 2009). Modeling techniques
for these dynamics are provided in Sect. 66.3. Fortunately, large- and medium-scale
UAVs often operate in open skies, which mitigates many of these detrimental effects.
However, actively studied concepts of mini/micro-UAV teams deployed below the
tree line, inside buildings, or in urban settings will be significantly affected by these
dynamics (Warwick 2009). Similarly, noise and interference from other emitting
sources in the environment can be difficult to predict and can change both in time
and by location Wagle and Frew (2011). In some cases, another source may actively
seek to interfere or “jam” communications to degrade the system performance as a
countermeasure (U.S. Air Force Chief Scientist AF/ST 2010). For similar reasons,
measurements of the signal strength, while accurate, may also fluctuate significantly
at time scales of only tens or hundredths of a second. The uncertainties in the model
and the rapid variation in the measurements pose a significant challenge in the
network feedback control loop and lead to limited performance guarantees of the
system (Goldsmith 2005).
Given the ability to control the network topology, the problem of efficiently
routing information between nodes is a challenge in itself (Goldsmith 2005).
Different types of data have different communication requirements, but in general
the typical objective is to route data to its destination with minimal delay, over links
with sufficient data rate capacity and with minimal number of packets dropped due
to error. A wireless network typically operates on one or more communication
channels in a specific range of the electromagnetic spectrum. As such, nodes
transmitting on the same or neighboring channels interfere with each other, which
generates a design trade (Xue and Kumar 2004) best illustrated through a simple
example. Assume that a UAV needs to transmit time sensitive data to a distant base.
One option is to increase its transmitter power enough to send the data directly to
the base, as UAV2 does in Fig. 66.2a. However, this higher transmit power now
interferes more with other UAVs also trying to communicate. Furthermore, the long
link may be strong enough to reach the base under nominal conditions, but may not
be robust to channel fading. Another option then is for the UAV to transmit over a
more robust link and at a lower power to another less distant UAV (as UAV2 now
does in Fig. 66.2b), which can then relay the message, possibly through additional
hops, to the base. This reduces interference from that UAV, but may cause delays
in end-to-end delivery of the data, and cost overhead in the network to establish
the multi-hop route. The challenge of selecting the best route and power settings
is further exacerbated as the number of agents increases, as wireless link qualities
continuously fluctuate under fading, and because the network topology is dynamic
as agents move around to execute the mission (Goldsmith 2005). This is especially
a b
2 2
4 4
1 1
3
3
5 5
Base Base
6
6
Centralized - Hub Network Decentralized - Ad-hoc Network
Fig. 66.2 Network architectures. In (a) all UAVs communicate only with the base, and UAV4 is
out of range and disconnected. In (b) connections can be shorter, and UAVs interconnect and can
route data via multi-hop
relevant for multi-UAV systems, which can involve vehicles traveling rapidly over
large distances.
Multi-agent networks for all types of systems face the communication control
challenges described above. The degree to which these challenges affect the system
changes based on the domain of operation. For instance, unmanned ground vehicles
(UGVs) typically operate in environments with obstacles to line of sight, resulting
in channel fading, but they have the ability to stop when a suitable location
to transmit is encountered (Zhang and Quilling 2011). Autonomous underwater
vehicles (AUVs) have limited underwater communication capabilities and may only
be able to exchange coordinated control information once every few hours upon
surfacing (Leonard et al. 2010). Unattended ground sensors (UGS) forming a fixed
wireless sensor network (WSN) may be less susceptible to dynamic changes in
topology but need to be power conscious when transmitting since their battery
lifetime is limited (Ibrahim et al. 2009). There are a number of challenges especially
relevant for UAVs. First, the size, weight, and power (SWaP) of the UAV radio
device can be limited by design constraints of the vehicle and compete with
requirements imposed by other onboard systems. Next, unless a vertical flight
aircraft is used, the antenna position and orientation, which affects channel quality
(Ahmed et al. 2011), is subject to vehicle dynamics. To mitigate this problem, some
designs have included multiple receivers on the vehicle meeting SWaP constraints
(Kung et al. 2010; Yanmaz et al. 2011). Similarly, a fixed wing vehicle is unable to
hold a fixed position considered to be “optimal” for communications. Also interest-
ingly, some lightweight materials commonly used for UAVs, such as carbon fiber,
have conductive properties which shield radio signals and can cause loss of link
(Krueger 2002). Finally, SWaP constraints in micro-UAVs may limit onboard
computing capability and therefore limit the complexity of the algorithm used for
motion control, relay positioning, or task assignment.
Even with proper system design and implementation, the communication net-
work will still have limitations that can significantly degrade the performance
of the multi-UAV system. Algorithms to control vehicle motion, transmission
power, information routing, and task allocation must rapidly converge to a good
solution and require minimal message exchanges to mitigate delays. Command
delays, however short, may cause a formation of UAVs to perform inefficiently or
churn (effectively go unstable), which can have disastrous effects such as vehicle
collisions (Schwager et al. 2011). Delayed and dropped messages sent to planning
agents, whether centralized or decentralized, can cause inconsistencies in situational
awareness and, as a result, flawed planning (Ponda et al. 2010). Similarly, message
delays in decentralized planning may prevent agents from reaching consensus on
a plan (Johnson et al. 2011). These limitations, therefore, have implications on the
system architecture, choice of algorithms, and performance guarantees for multi-
UAV systems.
66.3 Communications and Networking
This section provides the technical background in wireless communications and

networking that will be utilized by many of the multi-agent network control methods
in this chapter.
66.3.1 System and Network Architectures
The design and control of the communication network depends on the system
architecture and data transfer requirements. For instance, traditional cellular phone
networks form a hub-and-spoke model between a fixed base station and mobile users
in its area. Telephone voice data requires minimal delay to be effective, but does
not need high data rates and has some error tolerance. Conversely, data networks
for Internet downloads require high and bursty (sporadic) data rates and dynamic
bandwidth allocation between a gateway, routers, and end users. These networks
can tolerate some delays but should be relatively error-free (Modiano 2012).
A multi-UAV system network typically needs to support two types of data.
The first type consists of command and control messages including state information
(telemetry), observations of the world (e.g., estimated target location), and control
data (e.g., waypoints or task allocation). These messages have low bandwidth
requirements but must be exchanged with minimal delay and error for effective team
coordination. Second, information gathered from sensors such as video, still images,
atmospheric samples, and other types of remote sensing data may also need to be
transmitted to designated processing, exploitation, and dissemination (PED) centers
(Christmann and Johnson 2007). Again, communication requirements depend on
the data type, but live video, for instance, requires minimal delay and high data
rates and has some fault tolerance (Modiano 2012).
A multi-UAV system architecture built on centralized control, where vehicles
remain within transmission range of the command and control node, can operate
using a hub-and-spoke network model, similar to a cellular system (Fig. 66.2a).
Some implementations can even extend command and control range by using a low
data rate satellite communication system (Marshall et al. 2011). However, many of
the applications envisioned for multi-UAV teams require greater topology flexibility,
with internode information exchanges and relaying (Fig. 66.2b). This second
architecture (Fig. 66.2b) closely follows concepts of ad hoc networks, which have
received significant interest in many disciplines (Santi 2005), and applies to most of
the systems studied in this chapter.
Network architectures are generally abstracted to a multilayer design, where each
layer is individually responsible for a different role in communicating information.
Modern ad hoc networks typically consist of (1) a physical layer responsible for
sending bits across a wireless channel, (2) an access layer to deconflict channel
usage from multiple users, (3) a network layer to route information from node
to node, (4) a transport layer to regulate flow and retransmissions, and (5) the
application layer which uses and creates data (Modiano 2012). While understanding
these layers is important in network design, many of the applications discussed in
this chapter, along with latest research in ad hoc networks, push for cross layer
design optimization (Goldsmith 2005; Xiao et al. 2004; Kaabi and Filali 2010;
Hiertz et al. 2010). As such, further discussion of these layers is limited in this
chapter, and the reader is referred to Bertsekas and Gallager (1992) and Kawadia
and Kumar (2005) for more details.
66.3.2 Wireless Channels
In wireless networks, data is exchanged over communication channels by sending

information encoded and modulated from a transmitter, over the air using a signal,
to a receiver which demodulates and decodes to process the data. The strength of
the received signal power compared to surrounding noise, or signal-to-noise ratio
(SNR), , is a critical parameter in assessing the quality of the link. This value is
commonly modeled with the simplified relationship:
gP
SNR: D (66.1)
No W
˛
do
gDK (66.2)
d
where P is the transmission power, N2o is the power spectral density of the
environment noise, and W is the bandwidth of the signal. The channel gain g is
generally modeled as Eq. (66.2) in its most simple form, in which K is a gain based
on equipment characteristics, d is the Euclidean distance between the transmitter

and receiver, and ˛ is the path loss exponent which equals 2 in free space and up to
6 in environments congested with obstacles (Goldsmith 2005). This relationship
represents the path loss dynamics experienced in wireless channels, in which
decreases with increasing distance.
The next effect to consider is shadowing of the signal due to obstacles in the
line of sight. A practical way to model shadowing in uncertain environments is to
vary according to a lognormal distribution with variance 2 which depends on the
density of obstacles in the environment (Goldsmith 2005). A convenient method to
express this is to convert the channel gain in Eq. (66.2) into decibels (gdB ), where
shadowing now becomes a Gaussian-distributed random variable centered on the
path loss gain:

d
gdB D KdB 10˛ log N .0; dB
2
/ (66.3)
do
The distribution has been shown to be spatially correlated to Xc , the typical size of
obstacles in the area. The spatial covariance A.x/ in the distribution between two
points separated by distance x can be expressed as
A.x/ D dB
2
e x=Xc (66.4)
This expression can be used to predict the amount of time a moving vehicle may
undergo deep shadowing fades, or determine how to reposition the vehicle to exit a
shadowed area (Mostofi et al. 2009; Mostofi 2009).
Multipath fading, which occurs because of constructive and destructive interfer-
ence from scatterings and reflections of the signal by the obstacles, can be modeled
using a nonstationary random process centered on channel gain g subject to path
loss and shadowing in Eq. (66.3). Models using Rayleigh, Rician, or Nakagami
distributions with a tunable fading parameter have been shown to follow empirical
measurements of this dynamic (Mostofi et al. 2009; Goldsmith 2005). Multipath
fading values can be added to the gdB expression in Eq. (66.3) and then used
to predict the overall SNR using Eq. (66.1). This dynamic can vary significantly
over small distances and is often modeled without spatial correlation for simplicity
(Mostofi et al. 2009). The three dynamics can be seen using experimentally
measured wireless channel signal strength measurements in Fig. 66.3 obtained from
(Mostofi et al. 2009).
Many field experiments have studied the dynamics of link performance in greater
detail. Research in Wagle and Frew (2011) used data collected from aerial links to
measure an additional temporal correlation factor in the above dynamics. A study
in Ahmed et al. (2011) observed that link qualities vary depending on the relative
orientations between the transmitter and receiver, even when using omnidirectional
radio modules. Multiple antenna configurations were similarly used in Kung et al.
(2010) and Yanmaz et al. (2011) to improve the overall reception for the vehicle
through diversity.
Fig. 66.3 Measured

dynamics of a wireless
channel. The solid blue curve
represents the signal strength,
which varies at a small scale
due to multipath fading.
Spatially correlated
shadowing can be seen by
averaging locally (light gray
curve). The average of the
large-scale variations then
follows distance-dependent
path loss (dashed line)
(Figure and caption copyright
Mostofi et al. (2009))
Relationships in Eqs. (66.1)–(66.4) model the dynamics of a single wireless

channel. In a multi-user network, multiple agents simultaneously communicating
on the same or neighboring channels interfere with each other. In this case, SNR is
often rewritten as the signal to interference and noise ratio (SINR) represented by
ij for the channel from node i to j (Ribeiro 2011)
gij Pij
SINR: ij D P (66.5)
.No W /j C 1
.m;n/¤.i;j / gmj Pmn
Here, the interference caused by other nodes m transmitting to n adds to the noise at
receiver j in the denominator. While some of these effects can be mitigated through
signal code processing ( 1 ), a common method of deconfliction is to divide channel
usage by time (TDMA) where D 1 (Ribeiro 2011).
The SNR in Eq. (66.1) or SINR in Eq. (66.5) is a fundamental indicator of
wireless link performance and can be used to predict operationally relevant data
transmission metrics. First, the theoretical capacity of the channel can be evaluated
using Shannon’s equation:
cij D Wij log2 .1 C ij / (66.6)
which represents its data rate in bits/sec. While this value is an upper bound,
and actual data transmissions are often set to significantly lower rates due to
fading (Goldsmith 2005), it is a useful indicator in applications where information
throughput must be considered. Another important performance metric is the bit
error rate (BER) or the probability an information bit will be dropped, which
can require packet retransmissions and inefficient data flow. BER decreases with
increasing , and the relationship depends on the specific modulation scheme.
Equation (66.7) represents the relationship for a common modulation method, where
Q is defined in Goldsmith (2005):
p
BERij D Q. 2/ (66.7)
Many of the communication network control methods discussed in this chapter have
an objective to optimize or meet some threshold values for capacity, BERij , or even
information delay which relates to cij (Xiao et al. 2004). The models listed in this
section highlight the challenges associated with predicting and controlling wireless
channel performance due to dynamic uncertainties in fading and coupling with other
agents using the network.
66.3.3 Network Topology and Consensus
Another relevant challenge, especially in ad hoc networks, involves optimizing the

network topology (i.e., determining which nodes need to form interconnections),
to achieve the desired connectivity for information flow. The network topology can
be represented as a graph G D .V; E/ with a set of n nodes V D f1; 2; : : : ; ng
and edges E V V indicating connectivity between nodes. The graph can be
represented using an adjacency matrix, A, with n n elements aij , where aij D 1
if i can communicate with j and aij D 0 otherwise. Each row of the adjacency
matrix can be read directly to identify the set of neighbors for every agent. In some
formulations, aij D w, where w 2 Œ0; 1, depending on the quality of the link
and topology properties. The neighborhood of agent i , Ni D fj 2 V W aij ¤ 0g,
is defined as the set of agents that agent i can communicate with directly (Olfati-
Saber et al. 2007). For instance in the network in Fig. 66.2b, the neighbors of UAV4
are N4 D f2; 3; 5g. Theoretical formulations have established bounds on how many
agents each node should be connected to for the network to become asymptotically
disconnected or connected (Xue and Kumar 2004). For this reason, many multi-
vehicle systems control the network so that each agent remains connected to a
specific number of neighbors (Atay and Bayazit 2008; Zhu et al. 2009; Goddemeier
et al. 2011; Razafindralambo and Simplot-Ryl 2011; Reich et al. 2008; Zavlanos
et al. 2011). Many of these schemes are discussed in Sects. 66.5 and 66.6.
Analyzing the network topology provides useful insights on how information
propagates throughout the network and how the team of agents reaches consensus
on quantities of interest such as plans, parameters, and situational awareness. The
stability and convergence properties of most consensus algorithms can be predicted
by analyzing the spectral properties of the network graph Laplacian, L, for a given
network structure (Olfati-Saber et al. 2007). The Laplacian is another related matrix
representation of the network topology and is formulated as
L D DA (66.8)
P matrix A and the degree matrix, D D diag.d1 ; : : : ; dn /, with

using the adjacency
elements di D j 2Ni aij , representing the outgoing connections of each agent
(sum over its neighbors). The maximum degree of the graph, D maxi di , is useful
in determining spectral properties of the network Laplacian. In particular, all of the
eigenvalues of the Laplacian L in the complex plane are bounded by a closed disk
of radius centered at C 0j . Assuming the graph is undirected, where aij D aj i ,
the Laplacian L is symmetric, and therefore its eigenvalues lie on the real line within
the set Œ0; 2. These can be ordered sequentially as
0 D 1 2 n 2: (66.9)
ThePfirst (smallest) eigenvalue of L is always zero .1 D 0/ since every row sum
is j lij D 0, and is known as the trivial eigenvalue. The second eigenvalue, 2 , is
known as the algebraic connectivity or Fiedler value of the graph and is always
positive for strongly connected graphs where a single- or multi-hop path exists
from any agent to every other agent. The algebraic connectivity 2 determines the
speed of convergence for most consensus algorithms, and as such, many multi-
vehicle applications attempt to maximize this value by controlling the network
topology (Ibrahim et al. 2007, 2009; Yan and Mostofi 2010; Michael et al. 2009;
Zavlanos et al. 2011) (see Sects. 66.5 and 66.6). The last eigenvalue, n , is related to
stability guarantees for reaching an equilibrium in time-delayed networks, providing
associated bounds on the maximum allowable delays (Olfati-Saber et al. 2007).
Further details on consensus algorithms are provided in Sect. 66.4.1.
It must be noted that graph theoretic methods are simplifications of multi-UAV
network representations, since connectivity is often assumed to be binary and
links are considered active if agents are within a threshold distance of each other
(Santi 2005). These methods usually do not consider realistic wireless channel
dynamics as described in the previous subsection or some of the complexities in
information routing discussed next. Nevertheless, they do provide helpful mathe-
matical interpretations of the potential performance of the network and are useful in
guiding the system design to achieve configurations more suitable for information
exchange (Michael et al. 2009).
66.3.4 Information Routing
Since envisioned multi-UAV team applications require similar flexibility to ad hoc

wireless network architectures, network routing becomes a significant challenge
that must be considered in controlling the system. Given the wireless channel
properties and network topology, information must be routed along links with
sufficient capacities, and arrive at its destination with minimum error (BER) and
delay, as dictated by data requirements. Routing is an active area of research in
the wireless networking community (Al-Karaki and Kamal 2004). If the network
involves multiple channels on different frequencies, as is often the case, the problem
extends to finding the optimal route and channel allocation (Kaabi and Filali 2010).
Strategies range from decentralized, highly adaptable methods to more centralized
optimized schemes (Goldsmith 2005). Understanding how a particular network

system routes information is necessary to adequately control communications in
a team of UAVs, as it may affect where agents are positioned to provide a suitable
topology.
Several industry standards (802.11s, 802.15, 802.16) drive the design and
operation of mesh networking modules commonly used in ad hoc networks (Kaabi
and Filali 2010; Hiertz et al. 2010). These standards rely on decentralized routing
mechanisms. A commonly used protocol is Ad hoc On-Demand Distance Vector
(AODV) (Perkins and Belding-Royer 1999). In AODV, a node which needs to
transmit broadcasts a “request to send” message which identifies the destination
node. Neighbors receive the message, compute a cost metric (discussed later), and
rebroadcast the message. The process continues, where cost is computed at every
hop, until the destination is reached. Each node stores its lowest cost originator.
Once a designated amount of time has elapsed, the destination node sends a
unicast response to its lowest cost originator, which similarly relays the response to
backtrack the lowest cost route. Other similar routing techniques include Dynamic
Source Routing (DSR) (Perkins and Belding-Royer 1996), Optimized Link State
Routing (OLSR) (Adjih et al. 2004), and B.A.T.M.A.N (Garroppo et al. 2010).
Different metrics are implemented depending on the objectives, which include
lowest hop count, expected transmission count (ETX in Eq. (66.10), Salonidis et al.
2007; Decouto et al. 2005), and airtime metric (CAij in Eq. (66.11), Ribeiro 2008).
Here, ppk t is the packet loss ratio (based on BER), O is overhead, Bt is the test-
frame length, and r is the data rate (based on capacity):
1
ETXij D (66.10)
1 ppk t

Bt 1
CAij D OC (66.11)
r 1 ppk t
The protocols listed above are implemented in practice, such as on Zigbee modules
common in robotics research (Bhatia and Kaushik 2008), because of their rapid
response to changes in wireless channel performance and topology. A significant
drawback to them, however, is that information from different nodes is routed
mostly independently of other nodes and tends to use the same high-performance
routes. This can lead to congestion in high data rate applications and unfairness in
node usage (Salonidis et al. 2007). Researchers have proposed different network
optimization schemes to address this, which invoke conservation of information
flow at each node and link utilization constraints. For instance, authors in Toumpis
and Gitzenis (2009) optimize routing fairness by considering interference between
nodes and using electrical flow analysis. Similarly, convex optimization methods
have optimized routing through simultaneous link scheduling and transmitter power
allocation (Xiao et al. 2004; Cruz and Santhanam 2003). While these schemes
produce better routing solutions, they mostly consider static networks and would
require frequent recomputation in a dynamic environment to adapt to realistic
changes in the topology (Goldsmith 2005).
66.4 Coordination Under Communication Constraints
For multi-UAV teams to cooperatively execute a set of objectives, the communica-

tion network must exchange command and control messages and, when necessary,
remotely sensed data. The inability to communicate sensed data, such as live video,
to the designated processing centers may render the multi-UAV system ineffective
for its mission. Worse however, failure to properly exchange command and control
messages can lead to dangerous system failures or unintended consequences. This
is particularly true in decentralized systems, where these messages enable vehicle
control and team decision making. Inadequate team control can lead to formation
instability where one or multiple agents increasingly diverge from the desired
behavior and risk colliding or separating from the team. Poor decision making
may prevent the team from reaching consensus on a plan and remain idle without
accomplishing anything or, sometimes worse, take action on an undesirable plan
leading to unwanted consequences. Despite taking appropriate measures to control
the communication network, transmission delays and dropouts are virtually un-
avoidable and may occur unexpectedly. This naturally limits the rate of information
agents have available to make control decisions or formulate a team plan. Graph
theoretic properties introduced in Sect. 66.3, such as the graph Laplacian L and its
set of eigenvalues i , can provide insight into network bounds on formation stability
or team consensus (Olfati-Saber et al. 2007). This section focuses on identifying
these limitations, and the following two sections describe network control strategies
to improve team performance and network connectivity.
66.4.1 Information Consensus
A critical component of team coordination strategies involves performing consen-

sus, which is defined as reaching an agreement on quantities of interest that depend
upon the states of all agents (Olfati-Saber et al. 2007). Most distributed planning
and control approaches employ consensus algorithms to reach agreement on plans,
situational awareness, or other desired parameters. A consensus algorithm is a set
of rules, or protocol, that determines how the information is exchanged between
agents to ensure that the team will convergence to an equilibrium on the parameters
of interest. For example, a simple algorithm to perform consensus on a continuous
parameter z can be expressed as an nth order linear system where each agent
executes the following protocol:
X
xP i .t/ D .xj .t/ xi .t// 8i (66.12)
j 2Ni
xi .0/ D zi 2 R
Collectively the team dynamics can be written as the linear system
xP .t/ D Lx.t/ (66.13)

where L is the graph Laplacian as described in Sect. 66.3.3. Since Eq. (66.13) is a
linear system, the spectral properties of L determine the stability and convergence
rate of the consensus algorithm. In particular, as mentioned in Sect. 66.3.3, for
any strongly connected graph, all of the nontrivial eigenvalues of L are positive,
and thus the consensus algorithm is globally asymptotically stable. Furthermore,
the algorithm has an exponential convergence rate determined by the algebraic P
connectivity 2 and is guaranteed to achieve a unique equilibrium zN D 1=n i zi ,
which is the average of all the agents’ initial values (Almost identical guarantees
can be made for discrete representations of Eq. (66.13) as well.) (Olfati-Saber
et al. 2007). Note that, in general, these convergence and stability properties are
dependent upon L and 2 , which are defined by the network structure, and not on
the particular consensus algorithm used.
In realistic environments, there are several effects that complicate the transmis-
sion of messages between agents and thus impact the performance of consensus
algorithms. Some examples include time delays for transmitted messages and
varying network topologies where links between agents are subject to change over
time. Recent research has focused on analyzing consensus convergence and stability
properties in the presence of time delays and dynamic networks. It turns out that
the global exponential stability guarantees can be extended to dynamic networks.
As long as the graph G.t/ is strongly connected for all time t, the agents will
asymptotically reach an average consensus with convergence rate greater than or
equal to ?2 D mint 2 .G.t//, where 2 .G.t// represents the algebraic connectivity
value associated with the Laplacian for graph G.t/. Likewise, in time-delayed
networks, where messages are received after a delay instead of instantaneously,
the dynamics become
xP .t/ D Lx.t / (66.14)
and global exponential stability to the average-consensus equilibrium can be

guaranteed for delays within the range <
=2n . Note that here the last eigenvalue
of the Laplacian, n , provides bounds on the maximum allowable delay , and thus
networks with lower n are more robust to time delays. Recall from Sect. 66.3.3
that n < 2, favoring network topologies with lower maximum degree , such
as random graphs and ad hoc networks (Fig. 66.2b), instead of networks with large
hubs (Fig. 66.2a) (Olfati-Saber et al. 2007).
The remainder of this section describes how control and decision-making
algorithms employ consensus to achieve common goals for teams of UAVs and
the associated challenges and limitations. In particular, the stability and con-
vergence properties described by the network Laplacian affect the performance
of these planning and control algorithms as well, in very similar ways to the
effects on consensus discussed here. As such, many multi-vehicle approaches
explore strategies to directly control the network topology in order to increase
mission performance (Ibrahim et al. 2007, 2009; Yan and Mostofi 2010; Michael
et al. 2009; Zavlanos et al. 2011; Ponda et al. 2011). Sections 66.5 and 66.6
describe some of these methods and associated challenges in further
detail.
66.4.2 Effects on Cooperative Control
The feedback control process uses sensed feedback information with a model of
the system dynamics to select a control input. If feedback information is delayed or
corrupted due to communication degradations, control decisions may be flawed. For
instance, in military MQ-1 Predator procedures, a pilot remotely controls the UAV
manually using live forward-looking video on the nose of the aircraft as reference.
Delays in video feedback and control commands from link latencies have resulted in
pilot-induced oscillations (PIO), an instability from poor feedback control which has
resulted in numerous aircraft mishaps (Defense Research and Development Canada
2010). The same phenomenon can occur in autonomously controlled cooperative
teams of UAVs.
In general, network controlled systems (NCS) require a proper balance of control
and feedback update rates: too slow can prevent the system from being controlled
sufficiently well, but too fast can overload the communication network (Branicky
et al. 2000) or excite system dynamics resulting in instabilities (Schwager et al.
2011). The stability of the system also depends on control implementation and
feedback filtering techniques, as described in detail in Fax and Murray (2004).
For networked multi-agent systems, it is helpful to define the control graph
G ctrl , similar but separate from the communication graph G comm introduced in
Sect. 66.3 as G (Schwager et al. 2011). G ctrl is a directed graph that defines
control dependencies (Fig. 66.4). In “leader-follower” formations, each vehicle
selects control inputs based on the dynamics of the vehicle(s) directly preceding
it (Fig. 66.4a). Tracking errors can propagate down the formation and under certain
conditions exceed control authority, leading to instability, even with no delay in
information sharing (Al-Karaki and Kamal 2004). In “string” formations studied
in the context of an Automated Highway System, each vehicle communicates with
the vehicle directly ahead and must follow it by a prescribed safe distance (Liu
et al. 2001). Agents are slotted to communicate at separate times, which propagates
a b
1
1 2 3 2 3
2 3 4
c
5 6 7 1 2 3
Control Graph Optional Ctrl Cycle Comm. Graph
Fig. 66.4 Control and communication graphs. (a) represents a “leader-follower” formation free
of network delays as discussed in Al-Karaki and Kamal (2004). (b) is the configuration of the
Automated Highway System in Liu et al. (2001). (c) illustrates the graph in the three UAV
formation flight test in Michael et al. (2010) (Figure (c) copyright Michael et al. (2010))
an information delay down the chain. Each vehicle bases its control inputs on
the dynamics of the vehicle ahead and of it and on the lead vehicle dynamics
(Fig. 66.4b). In this structure, even infinitely small communication delays cause
string instabilities if agents compensate for error immediately. The formation can,
however, be stabilized given even greater communication delays if the vehicles
delay their control response to all apply it in near unison.
In Schwager et al. (2011) and Michael et al. (2010), the effects of network
delays on formation stability of a team of decentralized UAVs are explored in indoor
flight testing. Vehicles can communicate only with their neighbors and are similarly
slotted to broadcast knowledge of vehicle states at different times. Agents store
outdated team member states xj and update them as more recent information is
relayed to them. Control inputs ui .t/ in Eq. (66.15) for each UAV are based on this
information:
X
ui .t/ D Kij .yOij .t/ yi .t/ ıij /
j 2Nictrl
yOi i .t/ D Ci xi .t/ (66.15)

yOij .t/ D Cj xj .t ij T /
where G ctrl specifies the set of neighbors Nictrl , T is the overall time cycle of com-
munications (for all agents to transmit once), ij is the number of communication
hops from j to i , ıij is the desired interagent distance, and Kij is a control gain.
To derive stability bounds on the system, the authors rewrite the system to be in
continuous time at T second intervals and couple the entire system of agents as
(fully defined in Schwager et al. (2011)):
X.t C T / D AX.t/ C (66.16)
The structure of matrix A depends on G ctrl and G comm , and its eigenvalues i are
a function of Kij and T . The convergence rate of the formation is then directly
related to ˛ D jmax .A/j1=T , where ˛ 1 indicates Lyapunov stability and
˛ < 1 asymptotic stability. The indoor flight experiment consists of three UAVs
with control and communication graphs in Fig. 66.4c and second-order control as
in Eq. (66.17). Results show the following: (1) the formation is stable with small
gains and converges faster with faster network cycles T , (2) at higher control
gains, decreasing network cycle time may actually result in slower converge due
to unstable dynamics being excited, and (3) introducing feedback cycles in G ctrl can
also introduce instabilities.
Communication delays can also be problematic in multi-UAV path planning for
collision avoidance. In an outdoor flight test in Kopeikin et al. (2012),
decentralized vehicles continuously replan the next steps of their receding horizon
paths (not in formation). UAVs deconflict by not planning crossing paths to go
around each other. Short network delays and rapidly changing dynamics cause the
vehicles to sometimes plan using outdated information. On occasion, two UAVs
approaching each other head on may alternate in replanning deconflicting routes
left or right multiple times. This phenomenon, known as “churning,” is similar
to the common awkward situation where two people mutually block each other
in a hallway or sidewalk, and causes delays, hazards, and system performance
degradation.
66.4.3 Effects on Cooperative Decision Making
In order for multiple UAVs to cooperate in a mission, the team as a whole must
plan and make decisions collectively. The objective in cooperative decision making
is therefore for the networked team to reach an agreement on the tasks, plans, and
actions required to execute the mission. Different strategies exist to achieve conflict-
free plans, which require information exchange through the network. The dynamics
and uncertainties of the communication network have several implications on
decision-making strategies: (1) bandwidth limitations can restrict the content and
frequency of planning messages, (2) even small network delays can cause agents
to plan asynchronously making consensus difficult to achieve, and (3) network
dropouts may prevent agents from participating in team decision making which can
impact plan execution.
Because network bandwidth is a limited resource, cooperative decision-making
methods need to limit the content of planning messages, efficiently select which
pieces of information need to be shared with which members, and overall reach
consensus with as few messages as possible to mitigate delays and conflicts.
If the system operates under centralized control, every UAV needs to communicate
every required bit of situational awareness information to a central leader (e.g.,
the base station). This strategy minimizes the number of messages required to
reach consensus, but may require large amounts of low-level planning data to
be communicated to the base, and can cause the system to be slow to react
in a rapidly changing environment due to network delays. To overcome this,
decentralized planning strategies provide individual agents with greater decision-
making capability, where information can be shared more selectively and agents
can locally decide on courses of action based on their own situational awareness,
thus improving reaction time. Now, however, consensus on plans must be explic-
itly reached by exchanging messages over the network. A common distributed
strategy, referred to as implicit coordination, consists of agents independently
solving the entire centralized planning problem for all agents and carrying out
actions assigned to them. This requires agents to reach consensus on individ-
ual situational awareness and planning objectives prior to planning (Ren and
Beard 2005; Olfati-Saber and Murray 2004; Alighanbari et al. 2006), which
may require large exchanges of data across the network, especially if incon-
sistencies exist (Alighanbari and How 2005). Another popular planning method
involves using distributed auction algorithms, where agents individually plan

for themselves and consensus is reached through an auction mechanism (Sariel
and Balch 2005; Ahmed et al. 2005; Atkinson 2004). A particular distributed
auction approach that performs consensus on agent plans rather than on infor-
mation is the Consensus-Based Bundle Algorithm (CBBA) (Choi et al. 2009).
This algorithm employs a consensus protocol that explicitly specifies rules for
reaching agreements on tasks, thus guaranteeing a conflict-free solution despite
possible inconsistencies in situational awareness. In addition to being robust
to inconsistent information, performing consensus in the task space is usually
significantly faster, requiring fewer messages between the agents (Choi et al. 2009;
Johnson et al. 2011).
Another challenge that affects distributed planning algorithms is that they usually
rely on synchronous message passing in order to guarantee convergence. This is
difficult to enforce in dynamic environments where the number of tasks or agents in
the network may be changing and the network is subjected to the uncertainties and
dynamics described in the previous section. When messages are asynchronous, time
delays are not usually constant and messages may be received out of order, which
impacts typical consensus algorithms since old information may be processed as
new, thus hindering consensus. Being robust to these effects requires developing
new consensus protocols or plan deconfliction rules that can recognize and process
out of order messages. For instance, an asynchronous extension to CBBA addresses
these issues in the decentralized task planning domain (Johnson et al. 2011), and
several algorithms have been developed to perform linear asynchronous consensus
in the information domain (parameter consensus) (Ren et al. 2005; Olfati-Saber et al.
2007; Blondel et al. 2005; Bemporad et al. 2010). This however remains a general
problem for most multi-UAV systems of interest.
Finally, rapid link degradation or specific mission requirements (e.g., sending
a reconnaissance UAV on a long duration mission at a location far from the
base) may cause the network to disconnect. Most current distributed tasking
algorithms, however, require continued connectivity in order to converge, and
literature on limited communication environments has focused on maintaining
this network connectivity through relay teams (Nguyen et al. 2003; Ibrahim
et al. 2007; Ponda et al. 2011) by optimally placing vehicles (Dixon and Frew
2009) or by restricting allowable tasks. In realistic mission scenarios, where
communication links are broken and created in real time and teams of agents
often reconfigure themselves dynamically into different subnetwork structures, it
is desirable to have distributed tasking algorithms that can ensure conflict-free
solutions even when the network is disconnected. A recent study has considered
distributed methods to locally adapt the task planning space within each subnetwork
of agents, thus ensuring conflict-free assignments in the presence of network
disconnects (Ponda et al. 2010). The approach in Ponda et al. (2010) results in
assignments that would be equivalent to a mission operations center subdividing
task sets among the different subnetworks, but without requiring the centralized
overhead.
66.5 Motion Planning for Communications
Envisioned multi-UAV system applications involve controlling the agents to achieve

a defined primary mission objective. This commonly includes maximizing area
coverage, tracking a target, rendezvous in space, or formation and flocking con-
trol (Razafindralambo and Simplot-Ryl 2011; Goddemeier et al. 2011; Ghaffarkhah
and Mostofi 2009; Zavlanos et al. 2011). Oftentimes, the primary objective does
not specifically define connectivity objectives, but implicitly requires network
communications control for the system to operate effectively. As described in the
previous section, network performance depends on the quality of wireless links,
the number of interconnected neighbors, and the ability to route information based
on the topology. A common way to control network communications in a team of
multiple UAVs is to plan the motion of the vehicles in such a way that the resulting
topology supports the exchange of data required by the system. The general problem
then becomes controlling a set of N D f1; : : : ; ng team members to achieve the
mission objective and maintain the required state of connectivity. For mathematical
simplification, vehicles are often considered to be point nodes with first-order
dynamics, xPi .t/ D u.t/, where u.t/ is the control input at time t and xi .t/ is the
position of agent i . Some formulations go further and extend control to second-order
dynamics as in Eq. (66.17) (Zavlanos et al. 2011) where v is velocity,
xPi .t/ D vi .t/ and vPi .t/ D u.t/: (66.17)
whereas others include greater detail in modeling actual vehicle kinematics (Dixon
and Frew 2007). Network communication requirements vary from maintaining
connectivity with a specific number of neighbors to establishing routes with
sufficient throughput to send data to a base station. One of the earliest studies
where motion control was employed to maintain connectivity consisted of a team
of distributed robots with the task of meeting at an undefined point (rendezvous
problem). The strategy involved each robot independently moving a step amount
toward the geometric center of the area outlined by its perceived neighbors (Ando
et al. 1999). This process was repeated iteratively until robots converged on a central
location. Since, methods have expanded to include graph theoretic techniques,
continuous control through potential fields, reactive control strategies, and adaptive
techniques based on learning from the environment. Many of the efforts described
here were focused on ground robotic systems but are relevant to aerial applications
as well.
66.5.1 Graph Theoretic Methods and Potential Fields
Section 66.3 introduced the network graph G and the algebraic connectivity metric
2 , the second eigenvalue of the graph Laplacian L. Several studies have formulated
L by assigning weights w 2 Œ0; 1 to each interagent link based on relative distance,
with bounds wij D 1 if dij do where the link is strong and wij D 0 if
dij dmax where the signal is considered too weak to connect Zavlanos et al.
(2011). Connectivity objectives in multi-agent control problems often consist of
meeting some threshold 2 or even maximizing its value. Initially, discrete methods
were proposed to maximize 2 using iterative control inputs for each agent using
greedy algorithms in both centralized (Kim and Mesbahi 2005) and decentralized
frameworks (Gennaro and Jadbabaie 2006). These ensure that the team drives to
connected configurations throughout the mission. Continuous control methods to
maximize 2 using properties of L were then formulated in Zavlanos and Pappas
(2007) using potential fields in a centralized framework. Potential fields, which are
commonly used in multi-vehicle control applications, are virtual energy fields that
lead to attractive or repelling control inputs formulated as
X
ui .t/ D kri i ri ij (66.18)
j ¤i
1 1
ij D C 2 (66.19)
jjxij jj 2 dmax jjxij jj2
One benefit of this formulation is that it allows multiple, sometimes competing,

control objectives to be considered (Fig. 66.5). Here ij is designed according to
the desired connectivity dynamics between i and j , as illustrated by the simple
function in Eq. (66.19) which keeps the nodes within communication range and
prevents them from colliding (Michael et al. 2009). i can be any other steering
function, used, for example, for obstacle avoidance or target tracking (Hsieh et al.
2008) as implemented in Fig. 66.5 (obtained from Goddemeier et al. (2011)). As the
system approaches an unwanted state, such as a vehicle becoming disconnected
or dangerously approaching an obstacle, that effect dominates the effective force
generated by the potential field to correct the situation.
A decentralized version of this framework was proposed in Yang et al. (2010)
using a power iteration algorithm for each agent to compute its local 2 using
information from its neighbors and then estimate the global 2 to generate potential
fields. Potential fields have also been used to control multi-vehicle systems for
other connectivity requirements, using both graph theory and considering realistic
channel dynamics and measurements. For instance, Ajorlou et al. (2010) and
Dimarogonas and Johansson (2010) used bounded forces to minimize distance
with connected neighbors for a team to rendezvous or assemble in a formation.
One study considered Rician fading in wireless links to control multiple robots to
track a target and maintain connectivity with the base (Ghaffarkhah and Mostofi
2009). A primal-dual optimization algorithm was used to control agents through a
deployment objective while subjected to stochastic fading effects (Ny et al. 2011).
Finally, in Goddemeier et al. (2011) potential fields controlled a team of eight UAVs
to form a moving aerial communication network. Forces were created based on
the deployment objective of maximizing area coverage over a moving point, and
connectivity requirements which included maintaining a prescribed SNR over each
link subject to realistic fading, and staying connected with a specified number of
other UAVs.
Fig. 66.5 Potential fields for motion. (a) shows a set of potential fields which dominate as vehicles
approach the objects defined, and (b) represents potential fields dominating when further away to
generally guide the vehicles (Figure copyright Goddemeier et al. (2011))
66.5.2 Reactive and Adaptive Techniques
Instead of controlling motion for connectivity at every time, another strategy is

to allow agents to navigate freely according to mission objectives and take action
only when connectivity is threatened. One method involves each robot iteratively
planning how far it can travel at every step before losing connectivity with a required
minimum number of neighbors (Razafindralambo and Simplot-Ryl 2011). Other
studies were motivated by the difficulty of predicting wireless link performance
with enough accuracy to make control decisions in a realistic environment. In
these cases, each robot travels according to deployment objectives (using potential
fields or random motion) until the measured signal strength with a set number k of
neighbors drops below a designated threshold. At this point, the robot is commanded
to halt (Reich et al. 2008) or even backtrack toward its nearest neighbor (Hsieh et al.
2008) and can only resume deployment objectives if communication requirements
are again met. In these strategies, the required number of neighbors k is part
of the design trade. A higher value increases network robustness to unpredicted
disconnects but constrains the configuration of agents to be tighter and therefore
cover less surface area (Reich et al. 2008). When warranted, the team may benefit
from two agents disconnecting to add flexibility to the formation (so long as the
graph does not disconnect). For such cases, the study in Zavlanos and Pappas (2008),
which is explored in greater detail in the next subsection, employs a market-based al-
gorithm for agents to come to consensus on which link to remove to meet formation
objectives. Finally, certain scenarios involve agents taking on tasks beyond the reach
of communications where the only option is to temporarily disconnect the network.
Authors in Hollinger and Singh (2010) therefore define an implicit coordination
approach which is used to find optimal agent paths to execute the mission but reach
a point of full connectivity at some specific time interval.
Many recent robotic efforts have turned to adaptive and learning techniques to ad-
just their behavior and overcome unpredicted changes in the environment (Leonard
et al. 2010). Because it is so difficult to predict actual wireless channel performance
in dynamic environments, several approaches have used measurements in the field
to adapt the motion planning strategy. One basic approach is to take measurements
of the environment a priori and construct a “radio map” used to plan constraints on
relative node positions to stay connected (Hsieh et al. 2008). However, it may not
be practical to perform this type of pre-mission survey as access to the location may
be denied, and the environment may change over time due to different obstacles
or other emitting devices. For this reason, several studies propose probabilistic
methods based on the spatial correlation of channel fading to select locations to
transmit. For instance, in Lindhe and Johansson (2010) a single mobile agent
tracks a moving target and must send collected data to a fixed base. Current
measures of the SNR are used with link dynamics previously described to adaptively
schedule stops where the channel is predicted to be strong to transmit information
rather than waste energy transmitting during fades. Bounds on the dynamics are
derived to ensure system stability and prevent the agent from falling behind the
target. In a similar scenario, a single agent must repeatedly visit multiple areas
and send information to the base. Here the planner sequentially solves for the
optimal visit path using a Mixed-Integer Linear Program (MILP) and budgets the
power needed to transmit given the environment throughout and the path using
a nonlinear program (Ghaffarkhah et al. 2011) for energy conservation. Finally
in Mostofi (2009), a team of multiple UAVs learn link dynamics online to plan
motion cooperatively and track a target. At each step, the vehicles seek to gain
information Ii Œk: X
Ii Œk D Iis Œk C Ijc Œk; (66.20)
j ¤i
about the target either through direct sensing Iis Œk (which degrades with distance)
or by exchanging information with other agents Ijc Œk (which degrades with poor
communications). The measured SNR is used to learn the communication envi-
ronment and properly balance sensing and communication objectives. Furthermore,
channel fades correlated over multiple steps are used to identify areas of shadowing,
and adapt the motion strategy to improve communications.
66.5.3 Implementation Examples
Motion planning to control network communications is a relatively new research

topic in robotics, and as such, there have only been a few implementations of these
principles using actual vehicles. Most of the studies described previously in this
section were conducted in simulation. The following discusses two experiments that
are particularly relevant because they incorporate several of the motion planning
methods discussed previously.
In Reich et al. (2008), four UGVs are each tasked to a designated location,
and the mission objective is to transmit imagery back to the base, using multiple
hops if needed. A radio map is constructed by measuring the signal strength
between many pairs of locations in the area, and it is employed in assigning
tasks to the robots so that a connected multi-hop path exists to the base. During
the deployment, a centralized scheme uses potential fields to navigate the robots
to their targets. A reactive controller based on signal strength and throughput
measurements halts individual robots if connectivity thresholds are violated. The
experiment successfully shows the ability to control robots using the scheme. It also
highlights that signal strengths and throughputs vary significantly during the mission
with previous static measurements due to fading and multi-user interference. Results
support the choice to implement a reactive controller.
In Michael et al. (2009), potential fields are implemented in a decentralized
framework on a team of seven UGVs (Fig. 66.6). The objective is for each robot
to maintain distance-based connectivity with k neighbors while a leader robot is
given navigation commands. When flexibility is needed in the formation, links are
switched off from the potential field using a market-based consensus algorithm. In
this process, robots place bids on their link most at risk with bid value based on
their number of neighbors. The bids are propagated through the network, and the
highest bid link is allowed to be deleted. The framework only allows one link to
be deleted at a time to ensure that the graph stays connected, and the approach
introduces methods to handle asynchronous planning issues (see Sect. 66.4). In this
effort, robots compute their own local estimate of the network and exchange position
and bid information using Transmission Control Protocol (TCP) messages only with
their neighbors.
Fig. 66.6 Experiment with

seven UGVs maintaining
connectivity through motion
control. Each colored line in
the visualization depicts the
current connectivity graph
estimate for a particular robot
(Figure and caption copyright
Michael et al. (2009))
66.6 Communication Relays for Network Support
Because of their rapid and flexible deployment capabilities, persistence, ability

to fly above obstacles, and relative low cost, UAVs have generated significant
interest to serve as communication relays. Relays can be used to extend the range
of a network, add capacity for increased throughput, and increase the robustness
of the system to failure. Large-scale military UAVs are already delivering wide-
area communication connectivity to ground troops in mountainous areas (Jonson
et al. 2008), and similar smaller tactical concepts have been proposed for disaster
area recovery (Federal Communications Commission 2011), and wild land fire
fighting (Zajkowski et al. 2006) and to connect scattered wireless sensors and
mobile ad hoc networks (Palat et al. 2005). When deploying UAVs to serve as
communication relays, their primary mission objective is to maximize connectivity.
Many of the communication control methods described in the previous section
apply to this problem, but here the underlying network to be supported cannot
usually be controlled and may be dynamic. Control objectives typically involve
guiding relay agents to optimized locations to support data exchange requirements
at the lowest resource cost possible. Two classes of problems are provided in this
section. First, methods are investigated to optimize the control of a chain of relays
linking two mobile end nodes. Second, the problem of optimal relay deployment to
support a larger, typically fixed network is considered. In all cases, controlling the
communication network involves considering the dynamics of wireless links, the
topology, and information routing options.
66.6.1 Controlling Relay Chains
Several applications have called for a chain of relay nodes to extend communica-
tions through multiple hops from one end of the chain to the other. In this case,
once the chain forms, the relationship between pairs of nodes remains constant
(same two neighbors), and information routing is known since it simply follows
the nodes in the chain by default. A common application under which this is
studied involves maintaining communications between a moving agent and its base
station. This problem has been examined in complex obstructed environments where
the objective is to optimally navigate and position N relays to establish multi-
hop line of sight with the base. Optimal deployment paths can be generated in a
centralized framework using a Mixed-Integer Linear Program (Schouwenaars et al.
2006) or a Gauss pseudospectral solver initialized using a rapidly-exploring random
tree (RRT) solution (Kulling 2009). A suboptimal decentralized receding horizon
planner can also be used to provide faster real-time control to repeatedly “redeploy”
relays as the mission progresses (Kulling 2009). In a similar problem, a chain of
UGV relays must connect a mobile node to a base and optimize their formation
to minimize the end-to-end bit error rate (BER) (see Sect. 66.3 and Eq. (66.7)).
Here, wireless links are subjected to path loss, shadowing, and multipath fading
(Yan and Mostofi 2010; Bezzo and Fierro 2011). The resulting optimization can be
written as
XN
max J .x/ D ln.1 BERi 1;i / (66.21)
i D2
which is reformulated as a set of decentralized potential fields used for vehicle con-
trol as in Eq. (66.19). In Yan and Mostofi (2010), the formulation does not assume
deterministic link dynamics, since actual channel qualities can vary significantly
with predicted models. It instead employs a probabilistic framework with least
squares regression to estimate the value of the path loss exponent ˛ in Eq. (66.2) and
shadowing variance and correlation in Eq. (66.4) from several SNR measurements.
These estimates are constantly updated to plan motion to optimize the chain BER.
Since multipath fading is assumed to be spatially uncorrelated and varies quickly
over small distances, once the relay reaches its destination, it “jitters” around its
position to further reduce the quantity BERi 1;i . With a slight modification to the
problem formulation, the objective is changed to provide a chain configuration
which meets a threshold end-to-end BER but minimizes relay energy costs. This
allows motion cost and transmission costs (through variable power transmission) to
be traded off in the optimization, with tunable weights associated with each to favor
one over the other (Yan and Mostofi 2010).
Again, due to the recency of these studies, most of the above efforts were
conducted in simulation or controlled hardware experiments with simplifying
assumptions. However, authors in Dixon and Frew (2007) and Dixon (2010) tackle
the problem of forming a chain of relays using actual fixed wing UAVs flying
over a mountainous terrain. In this study, the objective is to maximize the end-
to-end capacity c of the chain found from Eq. (66.6) or in other words maximize
the minimum link capacity in the chain (Fig. 66.7). Two complicating but realistic
factors in this work include: (1) fixed wing aircraft have velocity limits 0 <
vmin < v < vmax and must loiter in orbits also constrained by max bank angle,
and (2) the authors chose to control the chain based on SINR signal strength
measurements only and avoid using channel models deemed inaccurate in dynamic
environments. Because of the first challenge, vehicle dynamics are modeled as being
nonholonomic with bicycle kinematics:
xPi D vi cos i; yPi D vi sin i; and Pi D vi si (66.22)
where si D g tan
v2i

is the path curvature and i is the heading angle. Consequently
control inputs take on the form ui D vi si . Since each relay is controlled based on
its SINR measurements only, an extremal seeking algorithm is implemented where
each UAV hones in on the heading which maximizes with its neighbors using

1
.t/ D argmax : (66.23)
minfi;i 1 ; i 1;i ; i;i C1 ; i C1;i g
3500
3000
UA-1 UA-2
2500
Y-Position [m]
Source Dest
2000
1500
Noise
1000
500
0
0 1000 2000 3000 4000 5000
X-Position [m]
Fig. 66.7 Flight testing a chain of UAV relays. The evolution of a simulation with two UAV
relays (left) and the type of fixed wing aircraft used in the experiment (right) (Figure copyright
Dixon (2010))
This framework does not assume equal interference levels for neighboring UAVs,
and as such, the SINR measurements are asymmetric on each side of the link. The
optimization therefore maximizes the minimum link capacity on either side of each
link. Then, a Lyapunov Guidance Vector Field (LGVF) (Dixon and Frew 2007)
drives the UAVs to an orbital pattern about a center point i , and a consensus
algorithm is used to synchronize the UAVs on a common orbiting phase angle.
Flight test results demonstrate the ability to control an aircraft to climb the SINR
gradient and improve capacity using signal strength measurements only. However,
RF fluctuations encountered are detrimental to the chaining algorithm and at times
cause the measured to actually increase while flying away from a neighbor.
This exemplifies the complexity in controlling the communication network with
operational dynamics using feedback measurements alone.
66.6.2 Deploying Relays to Support a Network
So far this section has explored problems involving the control of a chain of relays
between two designated points. In a number of other applications, however, UAVs
(or other mobile agents) serving as relays can be deployed to provide general
connectivity support to an existing network consisting of many nodes. The problem
here is to determine how best to deploy each relay and which nodes they should
form links with to satisfy the data requirements of the network. Objectives here
include maximizing the performance of the supported network or minimizing the
number of relays deployed to meet communication thresholds. These problems are
often studied in the context of UAVs relaying information from poorly connected
ground nodes or for wireless sensor network (WSN) nodes with connectivity and
energy limitations sending data to a base station.
Network graph theoretic processes described in Sect. 66.3 have also been
explored in these problems. One example consists of maximizing the lifetime of a
fixed WSN collecting data which depletes its battery over time as the energy is used
to transmit information (Ibrahim et al. 2007, 2009). Here K relays are available
to be deployed and redeployed to support the network. The graph Laplacian L
in Eq. (66.8) is composed of weights wij in its adjacency matrix A based on
the transmitter power needed to maintain a threshold BER across the link. This
formulation assumes all starting nodes have equal battery life, and therefore the
algebraic connectivity of the graph 2 can be used as a metric of remaining
network life. A centralized semi-definite programming (SDP) algorithm optimizes
the placement of the K relays to maximize 2 over the network. It discretizes the
area into cells, finds the best combination of placements for the K relays at the
center of these cells, and then rediscretizes selected cells to refine relay placement
over several iterations. From there, a minimum cost path routing algorithm (see
Sect.66.3) is implemented where relays are encoded with reduced cost to incentivize
their use. The deployment process is repeated as needed as other nodes deplete.
The formulation is also modified to solve for the minimum number of additional
relays required to support the network at a threshold level (Ibrahim et al. 2009).
In a similar problem, WSN nodes compute their k-redundancy in a distributed
framework using neighbor discovery messages (Atay and Bayazit 2008). The kired
is the number of links that need to be removed to disconnect any two neighbors
of i and is a measure of the importance of that node in global connectivity.
Here if any kired falls below some threshold because of a change in topology, the
network requests a relay which is positioned using several different optimization
schemes to meet connectivity requirements. Finally, the study in Zhu et al. (2009)
discusses how the deployment strategy changes for one UAV relay to support an ad
hoc network according to different connectivity objectives. Optimization methods
are presented to maximize (1) global and (2) worst-case connectivity based on
properties of the minimum spanning tree (MST) graph (Bertsimas and Tsitsiklis
1997) or (3) minimize network bisection probability and (4) maximize the Fiedler
value using graph Laplacian L. Each objective results in a different UAV relay
deployment plan.
In the previous three examples, the deployed relays essentially became nodes in
the supported ad hoc network, thus changing the topology and information routing
options. However, another strategy is to assume that relays are similar to gateways
and have a separate strong communication network to connect to other relays and
the rest of the world. These can then be used as a communication backbone to
support a disconnected ad hoc network (as illustrated with the “UAV Comm. Relay”
in Fig. 66.1). The network then takes on a form more consistent with a cellular
network architecture, where relays serve as mobile base stations and connect to
nodes in their area of responsibility. This is how current military large-scale UAV
communication relays operate (Jonson et al. 2008). Mobile backbone network
(MBN) optimization seeks to simultaneously position and assign K MBN relays
over a network in order to (1) maximize the minimum throughput in the network
(Srinivas and Modiano 2008), (2) maximize the aggregate network throughput
(Srinivas and Modiano 2008), (3) maximize the number of nodes meeting a
threshold throughput (Craparo and How 2010), and (4) optimize relay deployment
and node positioning (if those nodes can be controlled) for objective (3) (Craparo
and How 2010). In objectives (3) and (4), the optimization is solved using a Mixed-
Integer Linear Program (MILP) for a network optimization formulation (Bertsimas
and Tsitsiklis 1997). Similar to other previous studies, the optimal solver scales
poorly with increasing numbers of nodes and relays, and a suboptimal greedy
solution based on solving the Max-Flow problem (Bertsimas and Tsitsiklis 1997)
is provided in parallel for fast and provably good solutions.
66.6.3 Controlling Communications Through Task Allocation
A similar, yet distinct method to control network communications in a team of

UAVs is through task allocation. Here, the system consists of a multi-agent team
with a set of objectives, and similar to motion planning problems discussed in
Sect. 66.5, connectivity is an implicitly derived requirement. The framework can
use task allocation to ensure agents only take on tasks which result in a suitable
network topology, or better yet, allocate additional tasks to certain agents to serve as
communication relays as needed (Ponda et al. 2011). The challenges now include:
(1) determining which tasks should be assigned to which agents, (2) identifying
when the network topology will not support the assignment, (3) designating which
agents to take away from executing mission tasks to assign as relays and where
to position them, and (4) adapting to dynamic scenarios and uncertainty. These
frameworks follow the same general principles of multi-UAV mission planning
methods addressed in chapter Cooperative Mission Planning for Multi-UAV
Teams.
The authors in Tardioli et al. (2010) discuss a task allocation-based network
control mechanism. A set of N UGVs have a set of J possible tasks to accomplish
and must maintain a threshold receive signal strength (RSS), through multiple hops
if needed, to a base. A centralized task allocation planner initializes by dividing
the team of robots into subset clusters. Each cluster is assigned a common task
to accomplish. Then, motion planning potential fields similar to Eq. (66.19) are
used to attract the cluster to the goal, to the base, and to maintain connectivity
within the cluster while accomplishing the task. If the cluster cannot achieve the
goal and stay connected, it is merged with a neighboring cluster to expand the
multi-hop capability. Clusters can monotonically grow in size until the entire team
is included in one large cluster. Unachievable tasks are then discarded. At every
planning cycle, clusters are reset to their original size to again allow more flexibility
in task assignment. Different optimization strategies can be implemented to select
which task gets assigned to each cluster at every step (Tardioli et al. 2010). The
framework is implemented in an outdoor hardware system experiment with three
small UGVs. Results, while successful, again highlight the difficulty in predicting
link quality due to rapid variations and uncertainty in the dynamics.
Fig. 66.8 Flight testing a

task allocation framework
which maintains connectivity.
On the left, UAV2 relays
UAV3 as it executes a task,
and both plan to implement a
similar strategy for UAV3 to
accomplish the task in red at
the top of the image
A different task allocation strategy is proposed in Ponda et al. (2011) and

Kopeikin et al. (2011). A decentralized task allocation planner is used to coordinate
a team of N UAVs in maximizing reward in the mission by taking on high-value
tasks while maintaining distance-based connectivity requirements. The scenario
is dynamic, similar to realistic operations, and tasks are created at random times
and locations throughout the mission and have a limited time window of validity.
Because the assignment problem is combinatorial and difficult to solve optimally,
an auction-based multi-agent multi-task assignment algorithm (Consensus-Based
Bundle Algorithm) allows agents to place bids on each task (Choi et al. 2009).
Bids are based on task values, agent capabilities, and availability for execution.
Agents then share bids, assign winners, and iterate through the algorithm to reach
consensus on a conflict-free plan. The framework then predicts the network topology
for the execution of each task assigned, identifies disconnected configurations, and
creates new relay tasks to repair the network. Underutilized UAVs can then bid on
these relay tasks similar to any other. Two challenges in this planning framework
discussed in Ponda et al. (2011) involve (1) properly incentivizing agents to take on
the relay tasks which by themselves add no reward to the mission and (2) ensuring
the algorithm converges quickly to mitigate planning delays. This strategy does
not plan for connectivity through motion as UAVs travel to and from tasks, and
instead ensures the network is connected during task execution only. The system
was implemented in an outdoor flight test with a team of three UAVs to demonstrate
initial functionality (Fig. 66.8) and is being modified to account for more complex
wireless connectivity models (Kopeikin et al. 2011).
66.7 Concluding Remarks and Remaining Challenges
Teams of multiple UAVs show tremendous potential in a growing number of

military, first response, exploration, and other applications. As technology advances,
these cooperative systems will make their way to the operational domain. To realize
this, the research community is involved in a wide range of studies including aircraft
design, control, cooperative mission planning, sensing, and exploitation, among
many others. Effective and robust control of the communication network will be
a near universal requirement for these systems.
This chapter presented several topics on controlling the network for a team
of UAVs to ensure command and control messages are exchanged properly, and
remotely sensed data is transmitted back to desired locations. Control methods dis-
cussed include motion planning to create an effective network topology, transmitter
power regulation, deployment and control of mobile communication relay nodes,
and task allocation algorithms to ensure connectivity. While progress is ongoing in
all of these areas, there are nevertheless several high-level challenges which need to
be considered:
1. Robustness to uncertainty: Because of the recent surge in wireless devices for
every day applications, the electromagnetic spectrum is becoming more crowded
and communication systems are competing for bandwidth. Furthermore, as
described earlier, it is often difficult to accurately predict wireless channel
performance in dynamic environments. The need to control and implement
multi-vehicle systems to be robust to uncertainties and interference, and degrade
gracefully when the network underperforms, is critical (Goldsmith 2005).
2. Limiting bandwidth usage: Each bit transmitted costs valuable resources in both
bandwidth and energy. In order to maximize resources, all sensing and control
functions must minimize the amount of information to transmit. Consensus in
cooperative control and decision making must converge accurately across team
members with as few messages as possible. Methods to process and sort sensed
data onboard the UAV to minimize transfer requirements for analysis is also
paramount.
3. Ad hoc networking protocols: Researchers in the cooperative control and ad hoc
networking communities need to partner to jointly optimize control and routing
strategies to maximize resources. Network protocols need to quickly adapt to
the dynamics of multi-UAV networks, be flexible enough for the wide scope of
envisioned applications, and ensure that the network is secure from unwanted
users.
The communication network is one of many critical components of a multi-UAV
system. Many of the methods presented in this chapter require a more in-depth
understanding of other focused disciplines. Interested readers are encouraged to
read other chapters in this handbook describing Multi-UAV Cooperative Mission
Planning, Aircraft Flight Control, Trajectory Planning, and Advanced UAV
Communication Systems. In addition, the references listed in this chapter contain
detailed information on all of the topics covered. The following references are
recommended for readers interested in specific topics: wireless communications
and modeling (Goldsmith 2005; Mostofi et al. 2009), ad hoc topology control
(Santi 2005), mesh networking standards (Hiertz et al. 2010), consensus in network
graphs (Olfati-Saber et al. 2007), and graph theoretic vehicle control (Zavlanos et al.
2011).
References
C. Adjih, A. Laouiti, P. Minet, P. Muhlethaler, A. Qayyum, L. Viennot, The Optimised Routing
Protocol for Mobile Ad-Hoc Networks: Protocol Specification – Report N5145, Tech. rep.,
Institut National de Recherche en Informatique et en Automatique, France, 2004
A. Ahmed, A. Patel, T. Brown, M. Ham, M. Jang, G. Agha, Task assignment for a physical
agent team via a dynamic forward/reverse auction mechanism, in International Conference
on Integration of Knowledge Intensive Multi-Agent Systems, 2005
N. Ahmed, S. Kanhere, S. Jha, Link characterization for aerial wireless sensor networks, in IEEE
Globecom, Dec 2011
A. Ajorlou, A. Momeni, A.G. Aghdam, A class of bounded distributed control strategies
for connectivity preservation in multi-agent systems. IEEE Trans. Autom. Control 55(12),
2828–2833 (2010)
M. Alighanbari, J. How, Decentralized task assignment for unmanned aerial vehicles, in IEEE
Conference on Decision and Control (CDC) (2005), pp. 5668–5673
M. Alighanbari, L. Bertuccelli, J. How, A robust approach to the UAV task asignment problem, in
IEEE Conference on Decision and Control (CDC) (2006), pp. 5935–5940
J.N. Al-Karaki, A.E. Kamal, Routing techniques in wireless sensor networks: a survey. IEEE Wirel.
Commun. 12, 6–28 (2004)
J.N. Al-Karaki, A.E. Kamal, Leader-to-formation stability. IEEE Trans. Robot. Autom. 20(3),
443–455 (2004)
S. Anderson, R. Simmons, D. Goldberg, Maintaining line of sight communications networks
between planetary rovers, in International Conference on Intelligent Robots and Systems, Oct
2003
H. Ando, Y. Oasa, I. Suzuki, M. Yamashita, Distributed memoryless point convergence algorithm
for mobile robots with limited visibility. IEEE Trans. Robot. Autom. 15(5), 2418–2428 (1999)
N. Atay, B. Bayazit, Mobile wireless sensor network connectivity repair with K-redundancy.
Workshop on the Algorithmic Foundations of Robotics, Dec 2008
M.L. Atkinson, Results analysis of using free market auctions to distribute control of UAVs, in
AIAA 3rd Unmanned Unlimited Technical Conference, Workshop and Exhibit, 2004
A. Bhatia, P. Kaushik, A cluster based minimum battery cost AODV routing using multipath route
for zigbee, in IEEE Conference on Networks, St. Louis, MO, Dec 2008
A. Bemporad, M. Heemels, M. Johansson, Networked Control Systems (Springer,
Berlin/Heidelberg, 2010)
D. Bertsekas, R. Gallager, Data Networks (Prentice-Hall, Englewood Cliffs, 1992)
D. Bertsimas, J.N. Tsitsiklis, Introduction to Linear Optimization (Athena Scientific, Belmont,
1997)
N. Bezzo, R. Fierro, Decentralized connectivity and user localization via wireless robotic networks,
in IEEE Globecom, Dec 2011
V.D. Blondel, J.M. Hendrickx, A. Olshevsky, J.N. Tsitsiklis, Convergence in multiagent coordina-
tion, consensus, and flocking, in Proceedings of the IEEE Conference on Decision and Control,
2005
M.S. Branicky, S.M. Phillips, W. Zhang, Stability of networked control systems: explicit analysis
of delay, in American Control Conference (ACC), 2000
P. Chandler, M. Pachter, Hierarchical control for autonomous teams, in Proceedings of AIAA
Guidance, Navigation, and Control Conference and Exhibit, 2001
P.R. Chandler, M. Pachter, S. Rasmussen, UAV cooperative control, in American Control Confer-
ence (ACC), June 2001
H. Choi, L. Brunet, J.P. How, consensus-based decentralized auctions for robust task allocation.
IEEE Trans. Robot. 25, 443–455 (2009)
H.C. Christmann, E.N. Johnson, Design and implementation of a self-configuring ad-hoc network
for unmanned aerial systems, in AIAA Infotech@Aerospace Conference, May 2007
E. Craparo, How J.P., E.M., Throughput optimization in mobile backbone networks, in IEEE
Transactions on Mobile Computing, March 2010
R.L. Cruz, A. Santhanam, Optimal routing, link scheduling and power control in multihop wireless
networks, in IEEE INFOCOM, April 2003
D. Decouto, D. Aguayo, J. Bicket, R. Morris, A high-throughput path metric for multi-hop wireless
routing. Wirel. Netw. 11, 419–434 (2005). Springer
Defense Research and Development Canada, Human Factors Issues for Controlling Uninhabited
Aerial Vehicles, Tech. rep., 2010
D.V. Dimarogonas, K.H. Johansson, Bounded control of network connectivity in multi-agent
systems. IET Control Theory Appl. 4(8), 1330–1338 (2010)
C. Dixon, Controlled Mobility of Unmanned Aircraft Chains to Optimize Network Capacity
in Realistic Communication Environments, Ph.D. thesis, University of Colorado at Boulder,
Boulder, 2010
C. Dixon, E. Frew, Decentralized Extremum-Seeking Control of Nonholonomic Vehicles to Form a
Communication Chain, Lecture Notes in Control and Information Sciences, vol. 369 (Springer,
Berlin/New York, 2007), pp. 311
C. Dixon, E. Frew, Maintaining optimal communication chains in robotic sensor networks using
mobility control. Mobile Netw. Appl. 14(3), 281–291 (2009)
J.A. Fax, R.M. Murray, Information flow and cooperative control of vehicle formations, in
Automatic Control, IEEE Transactions on, 49(9), 1465–1476, (2004)
Federal Communications Commission, The Role of Deployable Aerial Communications Architec-
ture in Emergency Communications and Recommended Next Steps, Tech. rep., Public Safety
and Homeland Security Bureau, Washington, DC, 2011
R.G. Garroppo, S. Giordano, L. Tavanti, Experimental evaluation of two open source solutions
for wireless mesh routing at layer two, in International Symposium on Wireless Pervasive
Computing, 2010
M.C.D. Gennaro, A. Jadbabaie, Decentralized control of connectivity for multi-agent systems, in
IEEE Conference on Decision and Control, Dec 2006
A. Ghaffarkhah, Y. Mostofi, Communications-aware navigation functions for cooperative target
tracking, in American Control Conference (ACC), June 2009
A. Ghaffarkhah, Y. Yan, Y. Mostofi, Dynamic coverage of time varying environments using a
mobile robot – a communication aware perspective, in IEEE Globecom, Dec 2011
N. Goddemeier, S. Rohde, J. Pojda, C. Wietfeld, Evaluation of potential fields mobility strategies
for aerial network provisioning, in IEEE Globecom, Dec 2011
A. Goldsmith, Wireless Communications (Cambridge University Press, New York, 2005)
G.R. Hiertz, D. Denteneer, S. Max, R. Taori, J. Cardona, L. Berlemann, B. Walke, IEEE 802.11s:
the wlan mesh standard. IEEE Wirel. Commun. 2, 104–111 (2010)
G. Hollinger, S. Singh, Multi-robot coordination with periodic connectivity, IEEE International
Conference on Robotics and Automation (ICRA), May 2010
M.A. Hsieh, A. Cowley, R.V. Kumar, C.J. Taylor, Maintaining network connectivity and perfor-
mance in robot teams. IEEE J. 25(1–2), 111–131 (2008)
A. Ibrahim, K. Seddik, K. Liu, Improving Connectivity via Relays Deployment in Wireless Sensor
Networks (2007), pp. 1159–1163 http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4
411133&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs all.jsp%3Farnumber%3D
4411133
A.S. Ibrahim, K. Seddik, K.J.R. Liu, Connectivity-aware network maintenance via relays deploy-
ment. IEEE Trans. Wirel. Commun. 8(1), 356–366 (2009)
L.B. Johnson, S. Ponda, H.-L. Choi, J.P. How, Asynchronous decentralized task allocation for
dynamic environments, in Proceedings of the AIAA Infotech@Aerospace Conference, St. Louis,
MO, Mar 2011
T. Jonson, J. Pezeshki, V. Chao, K. Smith, J. Fazio, Applications of delay tolerant networking
(DTN) in airborne networks, in IEEE MILCOM, 2008.
F. Kaabi, F. Filali, Channel allocation and routing in wireless mesh networks: a survey and
qualitative comparison between schemes. Int. J. Wirel. Mobile Netw. 2(1), 104–111 (2010)
V. Kawadia, P.R. Kumar, Simultaneous routing and resource allocation via dual decomposition.
IEEE J. Sel. Areas Commun. 1, 76–88 (2005)
Y. Kim, M. Mesbahi, On maximizing the second smallest eigenvalue of a state-dependent graph
laplacian, in American Control Conference (ACC), July 2005
A.N. Kopeikin, S.S. Ponda, H.-L. Choi, J.P. How, Real-time dynamic planning to maintain
network connectivity in a team of heterogeneous unmanned systems, in Wi-UAV 2011, 2nd
International Workshop on Wireless Networking for Unmanned Autonomous Vehicles at the
IEEE GlobeComm Conference, Dec 2011
A.N. Kopeikin, A. Clare, O. Toupet, J.P. How, M.L. Cummings, Flight testing a heterogeneous
multi-UAV system with human supervision, in AIAA Guidance, Navigation, and Control
Conference (GNC), August 2012
Q.J. Krueger, Electromagnetic Interference and Radio Frequency Interference Shielding of
Carbon-Filled Conductive Resins, Master’s thesis, Michigan Technological University, Hous-
ton, MI, Feb 2002
K. Kulling, Optimal and Receding-Horizon Path Planning Algorithms for Communications
Relay Vehicles in Complex Environments, Master’s thesis, Department of Aeronautics and
Astronautics, Massachusetts Institute of Technology, Cambridge, MA, June 2009
H.T. Kung, C.-K. Lin, T.-H. Lin, S.J. Tarsa, D. Vlah, Measuring diversity on a low-altitude UAV
in ground-to-air wireless 802.11 mesh network, in IEEE Globecom, Dec 2010
B.M. Leiner, R.J. Ruth, A.R. Sastry, goals and challenges of the DARPA GloMo program. IEEE
Pers. Commun. 3(6), 34–43 (1996)
N.E. Leonard, D.A. Paley, R.E. Davis, D.M. Fratantoni, F. Lekien, F. Zhang, Coordinated control
of an underwater glider fleet in an adaptive ocean sampling field experiment in monterey bay.
J. Field Robot. 26(6), 718–740 (2010)
M. Lindhe, K.H. Johansson, Adaptive exploitation of multipath fading for mobile sensors, IEEE
International Conference on Robotics and Automation (ICRA), May 2010
X. Liu, S.S. Mahal, A. Goldsmith, J.K. Hedrick, Effects of communication delay on string stability
in vehicle platoons, in IEEE International Conference on Intelligent Transportation Systems,
Aug 2001
C. Marshall, M. Mears, S. Rasmussen, 2010 ICE-T cooperative control flight testing, in AIAA
Infotech@Aerospace Conference, April 2011
N. Michael, M.M. Zavlanos, V. Kumar, G.J. Pappas, Maintaining Connectivity in Mobile Robot
Networks, Springer Tracts on Advanced Robotics, vol. 54 (2009), pp. 117–126
N. Michael, M. Schwager, V. Kumar, D. Rus, An experimental study of time scales and stability
in networked multi-robot systems, in Springer Tracts on Advanced Robotics – International
Symposium on Experimental Robotics, 2010
E. Modiano, 16.36: Communication Systems and Networks, Lecture 1 (Massachusetts Institute of
Technology, Cambridge, 2012)
Y. Mostofi, Decentralized communication-aware motion planning in mobile networks: an
information-gain approach. J. Intell. Robot Syst. 56(2), 718–740 (2009)
Y. Mostofi, A.G. Ruiz, A. Ghaffarkhah, D. Li, Characterization and modeling of wireless channels
for networked robotic and control systems – a comprehensive overview, in IEEE International
Conference on Intelligent Robots and Systems, 2009
T. Mulkerin, M. Zernic, General aviation aircraft data communications analysis using a web-based
tool, in Digital Avionics Systems Conference, Dec 2003
K. Negus, A. Stephens, J. Lansford, Goals and challenges of the DARPA GloMo program. IEEE
Pers. Commun. 7(1), 20–27 (2000)
H. Nguyen, N. Pezeshkian, M. Raymond, A. Gupta, J. Spector, Autonomous communication
relays for tactical robots, in Proceedings of the International Conference on Advanced Robotics
(ICAR), 2003
J.L. Ny, A. Ribeiro, G.J. Pappas, Adaptive Communication-Constrained Deployment of Unmanned
Vehicle Systems – (Internet Draft), Tech. rep., University of Pennsylvania, Philadelphia, PA.,
2011, http://www.seas.upenn.edu/aribeiro/preprints/2012lenyetal.pdf
Office of the Secretary of Defense, FY2009-2034 Unmanned Systems Integrated Roadmap, 2009
R. Olfati-Saber, R.M. Murray, Consensus problems in networks of agents with switching topology
and time-delays. IEEE Trans. Autom. Control 49(9), 1520–1533 (2004)
R. Olfati-Saber, A. Fax, R.M. Murray, Consensus and cooperation in networked multi-agent
systems. IEEE Trans. Autom. Control 95(1), 215–233 (2007)
R.C. Palat, A. Annamalai, J.H. Reed, Cooperative relaying for ad-hoc ground networks using
swarm UAVs, in IEEE Military Communications Conference, 2005
C. Perkins, E. Belding-Royer, Dynamic Source Routing in Ad-Hoc Wireless Networks (Kluwer
Academic, Dordrecht, 1996)
C. Perkins, E. Belding-Royer, Ad-hoc on-demand distance vector (AODV) routing, in IEEE
Workshop on Mobile Computing Systems, Feb 1999
S. Ponda, J. Redding, H.-L. Choi, J.P. How, M. Vavrina, J. Vian, Decentralized planning for
complex missions with dynamic communication constraints, in American Control Conference
(ACC), June 2010
S. Ponda, L.B. Johnson, H.-L. Choi, J.P. How, Ensuring network connectivity for decentralized
planning dynamic environments, in AIAA Infotech@Aerospace Conference, April 2011
T. Razafindralambo, D. Simplot-Ryl, Connectivity preservation and coverage schemes for wireless
sensor networks. IEEE Trans. Autom. Control 56(10), 2418–2428 (2011)
J. Reich, V. Misra, D. Rubenstein, G. Zussman, Spreadable Connected Autonomic Networks
(SCAN) – Technical Report CUCS-016-08, Tech. rep., Columbia University, New York, 2008
W. Ren, R. Beard, Consensus seeking in multiagent systems under dynamically changing
interaction topologies. IEEE Trans. Autom. Control 50(5), 655–661 (2005)
W. Ren, R.W. Beard, D.B. Kingston, Multi-agent Kalman consensus with relative uncertainty, in
American Control Conference (ACC), vol. 3 (2005), pp. 1865–1870
A. Ribeiro, Notes on Implementing a IEEE 802.11s Mesh Point, in Euro NGI Workshop, Barcelona,
Spain, Jan 2008
A. Ribeiro, Stochastic learning algorithms for optimal design of wireless fading networks, in IEEE
Workshop on Signal Processing and Advances in Wireless Communications, June 2011
M.W. Ritter, R.J. Friday, M. Cunningham, T.G. Logan, The architecture of Metricom’s microcel-
lular data network and details of its implementation as the second and third generation ricochet
wide-area mobile data service, in IEEE Emerging Technology Symposium on Broadband
Communications, Sept 2001, pp. 143–152
T. Salonidis, M. Garetto, A. Saha, E. Knightly, Identifying high throughput paths in 802.11 mesh
networks: a model-based approach, in IEEE International Conference on Network Protocols,
Oct 2007
P. Santi, Topology Control in Wireless Ad Hoc and Sensor Networks (Wiley, Chichester, 2005)
S. Sariel, T. Balch, Real time auction based allocation of tasks for multi-robot exploration problem
in dynamic environments, in Proceedings of the AIAA Workshop on Integrating Planning Into
Scheduling, 2005
T. Schouwenaars, E. Feron, J. How, Multi-vehicle path planning for non-line of sight communi-
cation, in Proceedings American Control Conference, Jun 2006, Minneapolis, Minnesota, pp.
5757–5762.
M. Schwager, N. Michael, V. Kumar, D. Rus, Time scales and stability in networked multi-robot
systems, in IEEE International Conference on Robotics and Automation (ICRA), May 2011
A. Srinivas, E. Modiano, Joint node placement and assignment for throughput optimization in
mobile backbone networks. IEEE INFOCOM, Dec 2008
D.M. Stipanović, G. Inalhan, R. Teo, C.J. Tomlin, Decentralized overlapping control of a formation
of unmanned aerial vehicles. Automatica 40, 1285–1296 (2004)
D. Tardioli, A. Mosteo, L. Riazuelo, J. Villarroel, L. Montano, Enforcing Network Connectivity in
Robot Team Missions. Int. J. Robot. Res. 29(4), 460–480 (2010)
S. Toumpis, S. Gitzenis, Load balancing in wireless sensor networks using Kirchhoff’s voltage law,
in IEEE INFOCOM, April 2009
U.S. Air Force Chief Scientist (AF/ST), Technology Horizons: A Vision for Air Force Science &
Technology During 2010–2030, Tech. rep., United States Air Force, 2010
N. Wagle, E. Frew, Spatio-temporal characterization of airborne radio frequency environments,

IEEE Globecom, Dec 2011
G. Warwick, AFRL Aims for MAV Lead (2009), Available: www.aviationweek.com, Accessed
July 2011
L. Xiao, M. Johansson, S.P. Boyd, Simultaneous routing and resource allocation via dual
decomposition. IEEE Trans. Commun. 52(7), 1136–1144 (2004)
F. Xue, P. Kumar, The number of neighbors needed for connectivity of wireless networks. Wirel.
Netw. 10, 169–181 (2004)
Y. Yan, Y. Mostofi, Robotic router formation in realistic communication environments- a bit error
rate approach, in IEEE Military Communications Conference, 2010
P. Yang, R.A. Freeman, G.J. Gordon, K.M. Lynch, S.S. Srinivasa, R. Sukthankar, Decentralized
estimation and control of graph connectivity for mobile sensor networks. Automatica 46,
390–396 (2010)
E. Yanmaz, R. Kuschnig, C. Bettstetter, Channel measurements over 802.11a-based UAV-to-
ground links, in IEEE Globecom, Dec 2011
T. Zajkowski, S. Dunagan, J. Eilers, Small UAS communications mission, in 11th Biennial USDA
Forest Service Remote Sensing Application Conference, April 2006
M.M. Zavlanos, G.J. Pappas, Potential fields for maintaining connectivity of mobile networks.
IEEE Trans. Robot. 23(4), 812–816 (2007)
M.M. Zavlanos, G.J. Pappas, Distributed connectivity control of mobile networks, IEEE Trans.
Robot. 24(6), 1416–1428 (2008)
M.M. Zavlanos, M.B. Egerstedt, G.J. Pappas, Graph theoretic connectivity control of mobile robot
networks. IEEE J. 99(9), 1525–1540 (2011)
Y. Zhang, M. Quilling, Optimal backbone generation for robotic relay networks, in IEEE
International Conference on Computer Communications and Networks, 2011
H. Zhu, A.L. Swindlehurst, K. Liu, Optimization of MANET connectivity via smart deploy-
ment/movement of unmanned air vehicles. IEEE J. Veh. Technol. 58(7), 3533–3546 (2009)

Control of Communication Networks For Teams of UAVs Handbook of Unmanned Aerial Vehicles

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Control of Communication Networks For Teams of UAVs Handbook of Unmanned Aerial Vehicles

Uploaded by

Copyright:

Available Formats

Control of Communication Networks

for Teams of UAVs 66

A. Kopeikin () • S.S. Ponda • J.P. How

UAV Comm. Relay Manned

properly consider the benefits, costs, and implications in computation, communica-

66.2 Design Challenges

wireless channels is a significant challenge in this process. In general, links degrade

66.3 Communications and Networking

This section provides the technical background in wireless communications and

66.3.1 System and Network Architectures

66.3.2 Wireless Channels

In wireless networks, data is exchanged over communication channels by sending

on equipment characteristics, d is the Euclidean distance between the transmitter

Fig. 66.3 Measured

Relationships in Eqs. (66.1)–(66.4) model the dynamics of a single wireless

cij D Wij log2 .1 C ij / (66.6)

66.3.3 Network Topology and Consensus

Another relevant challenge, especially in ad hoc networks, involves optimizing the

P matrix A and the degree matrix, D D diag.d1 ; : : : ; dn /, with

66.3.4 Information Routing

Since envisioned multi-UAV team applications require similar flexibility to ad hoc

optimized schemes (Goldsmith 2005). Understanding how a particular network

66.4 Coordination Under Communication Constraints

For multi-UAV teams to cooperatively execute a set of objectives, the communica-

66.4.1 Information Consensus

A critical component of team coordination strategies involves performing consen-

Collectively the team dynamics can be written as the linear system

xP .t/ D Lx.t/ (66.13)

xP .t/ D Lx.t / (66.14)

and global exponential stability to the average-consensus equilibrium can be

66.4.2 Effects on Cooperative Control

Control Graph Optional Ctrl Cycle Comm. Graph

yOi i .t/ D Ci xi .t/ (66.15)

X.t C T / D AX.t/ C  (66.16)

66.4.3 Effects on Cooperative Decision Making

involves using distributed auction algorithms, where agents individually plan

66.5 Motion Planning for Communications

Envisioned multi-UAV system applications involve controlling the agents to achieve

xPi .t/ D vi .t/ and vPi .t/ D u.t/: (66.17)

66.5.1 Graph Theoretic Methods and Potential Fields

One benefit of this formulation is that it allows multiple, sometimes competing,

66.5.2 Reactive and Adaptive Techniques

Instead of controlling motion for connectivity at every time, another strategy is

66.5.3 Implementation Examples

Motion planning to control network communications is a relatively new research

Fig. 66.6 Experiment with

66.6 Communication Relays for Network Support

Because of their rapid and flexible deployment capabilities, persistence, ability

66.6.1 Controlling Relay Chains

xPi D vi cos i; yPi D vi sin i; and Pi D vi si (66.22)

66.6.2 Deploying Relays to Support a Network

66.6.3 Controlling Communications Through Task Allocation

A similar, yet distinct method to control network communications in a team of

Fig. 66.8 Flight testing a

A different task allocation strategy is proposed in Ponda et al. (2011) and

66.7 Concluding Remarks and Remaining Challenges

Teams of multiple UAVs show tremendous potential in a growing number of

N. Wagle, E. Frew, Spatio-temporal characterization of airborne radio frequency environments,

You might also like

X.t C T / D AX.t/ C (66.16)