1

An Overview of Recent Progress in the Study of

Distributed Multi-agent Coordination
Yongcan Cao, Member, IEEE, Wenwu Yu, Member, IEEE,
Wei Ren, Member, IEEE, and Guanrong Chen, Fellow, IEEE
arXiv:1207.3231v2 [math.OC] 4 Sep 2012

Abstract—This article reviews some main results and progress central station is available and powerful enough to control
in distributed multi-agent coordination, focusing on papers pub- a whole group of vehicles. Essentially, the centralized ap-
lished in major control systems and robotics journals since proach is a direct extension of the traditional single-vehicle-
2006. Distributed coordination of multiple vehicles, including
unmanned aerial vehicles, unmanned ground vehicles and un- based control philosophy and strategy. On the contrary, the
manned underwater vehicles, has been a very active research distributed approach does not require a central station for
subject studied extensively by the systems and control community. control, at the cost of becoming far more complex in structure
The recent results in this area are categorized into several and organization. Although both approaches are considered
directions, such as consensus, formation control, optimization, practical depending on the situations and conditions of the
task assignment, and estimation. After the review, a short
discussion section is included to summarize the existing research real applications, the distributed approach is believed more
and to propose several promising research directions along with promising due to many inevitable physical constraints such as
some open problems that are deemed important for further limited resources and energy, short wireless communication
investigations. ranges, narrow bandwidths, and large sizes of vehicles to
Index Terms—Distributed coordination, formation control, sen- manage and control. Therefore, the focus of this overview is
sor network, multi-agent system placed on the distributed approach.
In distributed control of a group of autonomous vehicles, the
main objective typically is to have the whole group of vehicles
I. I NTRODUCTION
working in a cooperative fashion throughout a distributed pro-
ONTROL theory and practice may date back to the tocol. Here, cooperative refers to a close relationship among
C beginning of the last century when Wright Brothers
attempted their first test flight in 1903. Since then, control
all vehicles in the group where information sharing plays a
central role. The distributed approach has many advantages in
theory has gradually gained popularity, receiving more and achieving cooperative group performances, especially with low
wider attention especially during the World War II when it operational costs, less system requirements, high robustness,
was developed and applied to fire-control systems, missile nav- strong adaptivity, and flexible scalability, therefore has been
igation and guidance, as well as various electronic automation widely recognized and appreciated.
devices. In the past several decades, modern control theory was The study of distributed control of multiple vehicles was
further advanced due to the booming of aerospace technology perhaps first motivated by the work in distributed comput-
based on large-scale engineering systems. ing [1], management science [2], and statistical physics [3].
During the rapid and sustained development of the modern In the control systems society, some pioneering works are
control theory, technology for controlling a single vehicle, generally referred to [4], [5], where an asynchronous agree-
albeit higher-dimensional and complex, has become relatively ment problem was studied for distributed decision-making
mature and has produced many effective tools such as PID problems. Thereafter, some consensus algorithms were studied
control, adaptive control, nonlinear control, intelligent control, under various information-flow constraints [6]–[10]. There are
and robust control methodologies. In the past two decades in several journal special issues on the related topics published af-
particular, control of multiple vehicles has received increas- ter 2006, including the IEEE Transactions on Control Systems
ing demands spurred by the fact that many benefits can be Technology (vol. 15, no. 4, 2007), Proceedings of the IEEE
obtained when a single complicated vehicle is equivalently (vol. 94, no. 4, 2007), ASME Journal of Dynamic Systems,
replaced by multiple yet simpler vehicles. In this endeavor, Measurement, and Control (vol. 129, no. 5, 2007), SIAM
two approaches are commonly adopted for controlling multiple Journal of Control and Optimization (vol. 48, no.1, 2009), and
vehicles: a centralized approach and a distributed approach. International Journal of Robust and Nonlinear Control (vol.
The centralized approach is based on the assumption that a 21, no. 12, 2011). In addition, there are some recent reviews
and progress reports given in the surveys [11]–[15] and the
Y. Cao is with the Control Science Center of Excellence, Air Force Research books [16]–[23], among others.
Laboratory, Wright-Patterson AFB, OH 45433, USA. W. Yu is with the
Department of Mathematics, Southeast University, Nanjing 210096, China This article reviews some main results and recent progress
and also with the School of Electrical and Computer Engineering, RMIT in distributed multi-agent coordination, published in major
University, Melbourne VIC 3001, Australia. W. Ren is with the Department of control systems and robotics journals since 2006. For results
Electrical Engineering, University of California, Riverside, CA 92521, USA.
G. Chen is with the Department of Electronic Engineering, City University before 2006, the readers are referred to [11]–[14].
of Hong Kong, Hong Kong SAR, China. Specifically, this article reviews the recent research results
 2

in the following directions, which are not independent but directed graphs. The Laplacian L has at least one single zero
actually may have overlapping to some extent: eigenvalue with a corresponding eigenvector 1 consisting of
1. Consensus and the like (synchronization, rendezvous). all numeric 1. Here and throughout, all matrices and vectors
Consensus refers to the group behavior that all the are assumed to have comparable dimensions unless otherwise
agents asymptotically reach a certain common agreement indicated.
through a local distributed protocol, with or without
predefined common speed and orientation.
2. Distributed formation and the like (flocking). Distributed B. Stochastic Matrices
formation refers to the group behavior that all the agents
form a pre-designed geometrical configuration through A nonnegative square matrix is called (row) stochastic
local interactions with or without a common reference. matrix if its every row is summed up to one. The product
3. Distributed optimization. This refers to algorithmic devel- of two stochastic matrices is still a stochastic matrix. A row
opments for the analysis and optimization of large-scale stochastic matrix P ∈ Rn×n is called indecomposable and
distributed systems. aperiodic if limk→∞ P k = 1y T for some y ∈ Rn [24].
4. Distributed task assignment. This refers to the imple-
mentation of a task-assignment algorithm in a distributed
III. C ONSENSUS
fashion based on local information.
5. Distributed estimation and control. This refers to dis- Consider a group of n agents, each with single-integrator
tributed control design based on local estimation about kinematics described by
the needed global information.
The rest of this article is organized as follows. In Section II, ẋi (t) = ui (t), i = 1, · · · , n, (1)
basic notations of graph theory and stochastic matrices are
introduced. Sections III, IV, V, VI, and VII describe the recent where xi (t) and ui (t) are, respectively, the state and the
research results and progress in consensus, formation control, control input of the ith agent. A typical consensus control
optimization, task assignment, and estimation, respectively. Fi- algorithm is designed as
nally, the article is concluded by a short section of discussions
with future perspectives. n
X
ui (t) = aij (t)[xj (t) − xi (t)], (2)
II. P RELIMINARIES j=1

This section introduces basic concepts and notations of where aij (t) is the (i, j)th entry of the corresponding ad-
graph theory and stochastic matrices. jacency matrix at time t. The main idea behind (2) is that
each agent moves towards the weighted average of the states
A. Graph Theory of its neighbors. Given the switching network pattern due
For a system of n connected agents, its network topology to the continuous motions of the dynamic agents, coupling
can be modeled as a directed graph denoted G = (V, W), coefficients aij (t) in (2), hence the graph topologies, are
where V = {v1 , v2 , · · · , vn } and W ⊆ V ×V are, respectively, generally time-varying. It is shown in [9], [10] that consensus
the set of agents and the set of edges which directionally is achieved if the underlying directed graph has a directed
connect the agents together. Specifically, the directed edge spanning tree in some jointly fashion in terms of a union of
denoted by an ordered pair (vi , vj ) means that agent j can its time-varying graph topologies.
access the state information of agent i. Accordingly, agent i The idea behind consensus serves as a fundamental principle
is a neighbor of agent j. A directed path is a sequence of for the design of distributed multi-agent coordination algo-
directed edges in the form of (v1 , v2 ), (v2 , v3 ), · · · , with all rithms. Therefore, investigating consensus has been a main
vi ∈ V. A directed graph has a directed spanning tree if there research direction in the study of distributed multi-agent co-
exists at least one agent that has directed paths to all other ordination. To bridge the gap between the study of consensus
agents. The union of a set of directed graphs with the same algorithms and many physical properties inherited in practical
set of agents, {Gi1 , · · · , Gim }, is a directed graph with the systems, it is necessary and meaningful to study consensus by
same set of agents and its set of edges is given by the union considering many practical factors, such as actuation, control,
of the edge sets of all the directed graphs Gij , j = 1, · · · , m. communication, computation, and vehicle dynamics, which
A complete directed graph is a directed graph in which each characterize some important features of practical systems. This
pair of distinct agents is bidirectionally connected by an edge, is the main motivation to study consensus. In the following
thus there is a directed path from any agent to any other agent part of the section, an overview of the research progress in
in the network. the study of consensus is given, regarding stochastic network
Two matrices are frequently used to represent the network topologies and dynamics, complex dynamical systems, delay
topology: the adjacency matrix A = [aij ] ∈ Rn×n with effects, the sampled-data framework, asynchronous effects,
aij > 0 if (vj , vi ) ∈ W and aij = 0 otherwise,Pand the quantization, convergence speed, and finite-time convergence,
n
Laplacian matrix L = [ℓij ] ∈ Rn×n with ℓii = j=1 aij mainly after 2006. Several milestone results prior to 2006 can
and ℓij = −aij , i 6= j, which is generally asymmetric for be found in [2], [4]–[6], [8]–[10], [25].
 3

A. Stochastic Network Topologies and Dynamics possible dynamic final states for double-integrator dynamics)
caused by the substantial dynamical difference. It is still an
In multi-agent systems, the network topology among all open question as if some general conditions (corresponding
vehicles plays a crucial role in determining consensus. The to some specific algorithms) can be found for consensus with
objective here is to explicitly identify necessary and/or suffi- double-integrator dynamics.
cient conditions on the network topology such that consensus In addition to analyzing the conditions on the network
can be achieved under properly designed algorithms. topology such that consensus can be achieved, a special type of
It is often reasonable to consider the case when the network consensus algorithm, the so-called gossip algorithm [36], [37],
topology is deterministic under ideal communication chan- has been used to achieve consensus in the stochastic setting.
nels. Accordingly, main research on the consensus problem The gossip algorithm can always guarantee consensus almost
was conducted under a deterministic fixed/switching network surely if the available pairwise communication channels satisfy
topology. That is, the adjacency matrix A(t) is deterministic. certain conditions (such as a connected graph or a graph
Some other times, when considering random communication with a directed spanning tree). The way of network topology
failures, random packet drops, and communication channel switching does not play any role in the consideration of
instabilities inherited in physical communication channels, it consensus.
is necessary and important to study consensus problem in the The current study on consensus over stochastic network
stochastic setting where a network topology evolves according topologies has shown some interesting results regarding: (1)
to some random distributions. That is, the adjacency matrix consensus algorithm design for various multi-agent systems,
A(t) is stochastically evolving. (2) conditions of the network topologies on consensus, and
In the deterministic setting, consensus is said to be achieved (3) effects of the stochastic network topologies on the con-
if all agents eventually reach agreement on a common state. vergence rate. Future research on this topic includes, but not
In the stochastic setting, consensus is said to be achieved limited to, the following two directions: (1) when the network
almost surely (respectively, in mean-square or in probability) topology itself is stochastic, how to determine the probability
if all agents reach agreement on a common state almost surely of reaching consensus almost surely? (2) compared with the
(respectively, in mean-square or in probability). Note that the deterministic network topology, what are the advantages and
problem studied in the stochastic setting is slightly different disadvantages of the stochastic network topology, regarding
from that studied in the deterministic setting due to the differ- such as robustness and convergence rate?
ent assumptions in terms of the network topology. Consensus As is well known, disturbances and uncertainties often exist
over a stochastic network topology was perhaps first studied in networked systems, for example, channel noise, commu-
in [26], where some sufficient conditions on the network topol- nication noise, uncertainties in network parameters, etc. In
ogy were given to guarantee consensus with probability one for addition to the stochastic network topologies discussed above,
systems with single-integrator kinematics (1), where the rate the effect of stochastic disturbances [38], [39] and uncertain-
of convergence was also studied. Further results for consensus ties [40] on the consensus problem also needs investigation.
under a stochastic network topology were reported in [27]– Study has been mainly devoted to analyzing the performance
[35], where research effort was conducted for systems with of consensus algorithms subject to disturbances and to present-
single-integrator kinematics [27]–[34] or double-integrator dy- ing conditions on the uncertainties such that consensus can be
namics [35]. Consensus for single-integrator kinematics under achieved. In addition, another interesting direction in dealing
stochastic network topology has been extensively studied in with disturbances and uncertainties is to design distributed
particular, where some general conditions for almost-surely local filtering algorithms so as to save energy and improve
consensus was derived [29], [30], [33]. Loosely speaking, computational efficiency. Distributed local filtering algorithms
almost-surely consensus for single-integrator kinematics can play an important role and are more effective than traditional
be achieved, i.e., xi (t) − xj (t) → 0 almost surely, if and centralized filtering algorithms for multi-agent systems. For
only if the expectation of the network topology, namely, the example, in [41]–[43] some distributed Kalman filters are
network topology associated with expectation E[A(t)], has a designed to implement data fusion. In [44], by analyzing
directed spanning tree. It is worth noting that the conditions consensus and pinning control in synchronization of complex
are analogous to that in [9], [10], but in the stochastic setting. networks, distributed consensus filtering in sensor networks is
In view of the special structure of the closed-loop systems addressed. Recently, Kalman filtering over a packet-dropping
concerning consensus for single-integrator kinematics, basic network is designed through a probabilistic approach [45].
properties of the stochastic matrices play a crucial role in Today, it remains a challenging problem to incorporate both
the convergence analysis of the associated control algorithms. dynamics of consensus and probabilistic (Kalman) filtering
Consensus for double-integrator dynamics was studied in [35], into a unified framework.
where the switching network topology is assumed to be driven
by a Bernoulli process, and it was shown that consensus can be B. Complex Dynamical Systems
achieved if the union of all the graphs has a directed spanning Since consensus is concerned with the behavior of a group
tree. Apparently, the requirement on the network topology for of vehicles, it is natural to consider the system dynamics
double-integrator dynamics is a special case of that for single- for practical vehicles in the study of the consensus problem.
integrator kinematics due to the difference nature of the final Although the study of consensus under various system dynam-
states (constant final states for single-integrator kinematics and ics is due to the existence of complex dynamics in practical
 4

systems, it is also interesting to observe that system dynamics • Nonholonomic mobile robots [54], [77], [83], [90]. The
play an important role in determining the final consensus dynamics are described by
state. For instance, the well-studied consensus of multi-agent
ẋi = ui cos θi , ẏi = ui sin θi , θ̇i = ωi , i = 1, · · · , N,
systems with single-integrator kinematics often converges to a
(6)
constant final value instead. However, consensus for double-
integrator dynamics might admit a dynamic final value (i.e., a where [xi , yi ] denotes the location of the ith agent,
time function). These important issues motivate the study of and ui and ωi denote, respectively, its translational and
consensus under various system dynamics. rotational velocity. Note that there are three states and two
As a direct extension of the study of the consensus prob- control inputs. Therefore, the dynamics for nonholonomic
lem for systems with simple dynamics, for example, with mobile robots are underactuated. This poses substantial
single-integrator kinematics or double-integrator dynamics, difficulties in designing proper consensus algorithms with
consensus with general linear dynamics was also studied corresponding stability analysis.
recently [46]–[51], where research is mainly devoted to finding • Rigid bodies and the like [63]–[66], [78], [79], [85]. One
feedback control laws such that consensus (in terms of the typical (but not unique) description of the dynamics is
output states) can be achieved for general linear systems Mi (qi )q̈i +Ci (qi , q̇i )q̇i +gi (qi ) = τi , i = 1, · · · , N, (7)
ẋi = Axi + Bui , yi = Cxi , (3) where qi ∈ Rp is the vector of generalized coordinates,
Mi (qi ) ∈ Rp×p is the symmetric positive-definite inertia
where A, B, and C are constant matrices with compatible matrix, Ci (qi , q̇i )q̇i ∈ Rp is the vector of Coriolis and
sizes. Apparently, the well-studied single-integrator kinematics centrifugal torques, gi (qi ) is the vector of gravitational
and double-integrator dynamics are special cases of (3) for torques, and τi ∈ Rp is the vector of torques produced
properly choosing A, B, and C. by the actuators associated with the ith agent. In practice,
As a further extension, consensus for complex systems has the dynamics of many mechanical systems are similar
also been extensively studied. Here, the term consensus for to (7). A notable property regarding the dynamics of rigid
complex systems is used for the study of consensus problem bodies is that Ṁi (qi )−2Ci (qi , q̇i ) is skew-symmetric (i.e.,
when the system dynamics are nonlinear [52]–[85] or with z T [Ṁi (qi ) − 2Ci (qi , q̇i )]z = 0 for all z ∈ Rp ), which
nonlinear consensus algorithms [86]–[88]. Examples of the plays a crucial role in finding Lyapunov functions and
nonlinear system dynamics studied in the consensus problem the subsequent stability analysis.
include: Although the aforementioned system dynamics are different
• Nonlinear oscillators [57]. The dynamics are often as- from the well-studied single-integrator kinematics and double-
sumed to be governed by the Kuramoto equation integrator dynamics, the main research problem is same,
namely, to drive all agents to some common states through
N
KX local interactions among agents. Similarly to the consensus
θ̇i = ωi + sin(θj − θi ), i = 1, 2, · · · , N, (4) algorithms proposed for systems with simple dynamics, the
N j=1
consensus algorithms used for these complex models are also
where θi and ωi are, respectively, the phase and natural based on a weighted average of the state differences, with
frequency of the ith oscillator, N is the number of some additional terms if necessary. Main research work has
oscillators, and K is the control gain. Generally, the been conducted to design proper control algorithms and derive
control gain K plays a crucial role in determining the necessary and/or sufficient conditions such that consensus can
synchronizability of the network. be achieved ultimately.
• Complex networks [55], [67]–[71], [80]–[82], [84], [89]. Note that although the objective is same, i.e., to guarantee
The dynamics are typically represented as reaching agreement on some final states, the problem is more
complicated due to the nonlinearity of the closed-loop systems.
N
X In addition, most properties of stochastic matrices cannot
ẋi (t) =f (xi (t)) + c aij (t)Γ(xj (t) − xi (t)), be directly applied to their convergence analysis. The main
j=1,j6=i techniques used in their stability analysis include dissipativity
i = 1, 2, · · · , N, (5) theory [52], nonsmooth analysis [87], [88], [90], and especially
Lyapunov functions [57], [63], [65], [66], [90], [91].
T
where xi = (xi1 , xi2 , · · · , xin ) ∈ Rn is the state vector One particular interesting topic is synchronization in com-
of the ith node, f : Rn 7→ Rn is a nonlinear vector func- plex networks which has been widely investigated in the
tion, c is the overall coupling strength, A(t) = [aij (t)] is past decade [92], [93]. Mathematically, the definitions for
the outer coupling matrix with aij (t) = 1 if node i and synchronization in complex networks and consensus in multi-
node j are connected at time t but otherwise aij (t) = 0, agent systems are very similar, so to differentiate these two
with aii (t) = ki (degree of node i), and Γ is a general definitions and promote research exchanges in these two
inner coupling matrix describing the inner interactions topics, their differences are briefly summarized below.
between different state components of agents. It is easy 1) Different Asymptotic States (Nonlinear Dynamics ver-
to see that model (1) with control input (2) is a special sus Linear Dynamics). In the studies of synchronization in
case of (5) with f = 0. complex networks, researchers focus on synchronization with
 5

self-nonlinear dynamics where each single system is unstable stability. A well-studied consensus algorithm for (1) is given
and thus the final asymptotic synchronization state is typically in (2), where it is now assumed that time delay exists. Two
time-varying [92], [94]. However, in the investigations of types of time delays, communication delay and input delay,
multi-agent systems, the individual self-dynamics on each have been considered in the literature. Communication delay
system is usually linear or zero and therefore the asymptotic accounts for the time for transmitting information from origin
consensus state is usually a constant [6], [8]. to destination. More precisely, if it takes time Tij for agent i
2) Different Focuses (Known Connectivity versus Time- to receive information from agent j, the closed-loop system
varying Distributed Protocol). In synchronization of complex of (1) using (2) under a fixed network topology becomes
networks, the aim is to reveal how the network structure, which n
is known in priori, affects the nonlinear collective dynamics
X
ẋi (t) = aij (t)[xj (t − Tij ) − xi (t)]. (8)
[92], [93], while the aim of consensus in multi-agent systems j=1
is to figure out how the designed distributed local protocol
concerning mobile time-varying network structure affects the An interpretation of (8) is that at time t, agent i receives
consensus behavior [6], [8], [9]. information from agent j and uses data xj (t − Tij ) instead of
3) Different Approaches (Lyapunov Method versus Stochas- xj (t) due to the time delay. Note that agent i can get its own
tic Matrix Theory). Since both complex networks and multi- information instantly, therefore, input delay can be considered
agent systems are networked systems, algebraic graph theory as the summation of computation time and execution time.
[95] is a common approach to use. Because of the nonlinear More precisely, if the input delay for agent i is given by Tip ,
terms in synchronization of complex networks, Lyapunov then the closed-loop system of (1) using (2) becomes
function method is usually used together with matrix theory n
[55], [89], [94]. In order to show consensus in multi-agent aij (t)[xj (t − Tip ) − xi (t − Tip )].
X
ẋi (t) = (9)
systems with time-varying network structures, stochastic ma- j=1
trix theory [4]–[6], [9] and convexity analysis [10] are often
applied. Clearly, (8) refers to the case when only communication
4) Different Inner Matrices Γ (General Inner Matrix versus delay is considered while (9) refers to the case when only
Particular Inner Matrix). In the typical simple consensus input delay is considered. It should be emphasized that both
model, the inner matrices
 Γ are
 usually an identity matrix and communication delay and input delay might be time-varying
0 1 and they might co-exist at the same time.
a rank-one matrix for multi-agent systems with
0 0 In addition to time delay, it is also important to consider
single-integrator kinematics [8] and double-integrator dynam- packet drops in exchanging state information. Fortunately,
ics [96]–[98], respectively. In consensus models with higher- consensus with packet drops can be considered as a special
order dynamics [99], the inner matrix is similar. However, the case of consensus with time delay, because re-sending packets
inner matrix in system (5) is a general one. after they were dropped can be easily done but just having time
In summary, synchronization in complex networks focuses delay in the data transmission channels.
on nonlinear dynamics while consensus in multi-agent systems Thus, the main problem involved in consensus with time
focuses on distributed cooperative control, and thus different delay is to study the effects of time delay on the convergence
approaches are utilized. and performance of consensus, referred to as consensusabil-
The current research on consensus with complex systems ity [100].
focuses on fully-actuated systems although consensus for non- Because time delay might affect the system stability, it is
holonomic mobile robots [90], which are typical underactuated important to study under what conditions consensus can still
systems. Note that many mechanical devices are described be guaranteed even if time delay exists. In other words, can
by systems with underactuation. Therefore, it is important to one find conditions on the time delay such that consensus can
develop appropriate consensus algorithms for underactuated be achieved? For this purpose, the effect of time delay on the
systems. consensusability of (1) using (2) was investigated. When there
exists only (constant) input delay, a sufficient condition on the
C. Delay Effects time delay to guarantee consensus under a fixed undirected
Time delay appears in almost all practical systems due interaction graph is presented in [8]. Specifically, an upper
to several reasons: (1) limited communication speed when bound for the time delay is derived under which consensus
information transmission exists; (2) extra time required by can be achieved. This is a well-expected result because time
the sensor to get the measurement information; (3) compu- delay normally degrades the system performance gradually but
tation time required for generating the control inputs; and will not destroy the system stability unless the time delay is
(4) execution time required for the inputs being acted. In above a certain threshold. Further studies can be found in,
general, time delay reflects an important property inherited e.g., [101]–[111], which demonstrate that for (1) using (2),
in practical systems due to actuation, control, communication, the communication delay does not affect the consensusability
and computation. but the input delay does. In a similar manner, consensus with
Knowing that time delay might degrade the system perfor- time delay was studied for systems with different dynamics,
mance or even destroy the system stability, studies have been where the dynamics (1) are replaced by other more complex
conducted to investigate its effect on system performance and ones, such as double-integrator dynamics [97], [112]–[119],
 6

complex networks [120]–[123], rigid bodies [124], [125], and remain unchanged during each sampling period. Under this cir-
general nonlinear dynamics [126]. cumstance, consensus is studied in a sampled-data framework,
In summary, the existing study of consensus with time delay called sampled-data consensus, which reflects the limitations
mainly focuses on analyzing the stability of consensus algo- inherited in physical measurement and control units. Mean-
rithms with time delay for various types of system dynamics, while, it is also important to point out that the sampled-data
including linear and nonlinear dynamics. Generally speaking, consensus algorithms require much less information exchange
consensus with time delay for systems with nonlinear dy- and computational power than the continuous-time consensus
namics is more challenging. For most consensus algorithms algorithms. Accordingly, consensus under the sampled-data
with time delays, the main research question is to determine framework deserves certain consideration.
an upper bound of the time delay under which time delay Sampled-data consensus was investigated in, e.g., [115],
does not affect the consensusability. For communication delay, [127]–[136]. Consensus for systems with single-integrator
it is possible to achieve consensus under a relatively large kinematics (1) was studied under a sampled-data framework
time delay threshold. A notable phenomenon in this case is with a fixed or a switching network topology, in [128],
that the final consensus state is constant. Considering both [129], where some necessary and/or sufficient conditions were
linear and nonlinear system dynamics in consensus, the main presented to guarantee achieving consensus. Sampled-data
tools for stability analysis of the closed-loop systems include consensus of systems with double-integrator kinematics was
matrix theory [102], [103], Lyapunov functions [120], [121], studied under fixed or switching network topologies in [115],
frequency-domain approach [106], passivity [122], and the [127], [130]–[136]. Due to the fact that an infinitely large
contraction principle [101]. sampling period will cause no information exchange among
Although consensus with time delay has been studied the agents, the main research question is to find conditions
extensively, it is often assumed that time delay is either on the sampling period T , which might be time-varying,
constant or random. However, time delay itself might obey its such that consensus can be achieved. The conditions on the
own dynamics, which possibly depend on the communication network topology for the sampled-data closed-loop systems
distance, total computation load and computation capability, are mostly similar to that for the continuous-time closed-loop
etc. Therefore, it is more suitable to represent the time delay systems. Note that the existing research on consensus in a
as another system variable to be considered in the study of the sampled-data framework mainly focuses on the simple system
consensus problem. In addition, it is also important to consider dynamics and thus the closed-loop system can be represented
time delay and other physical constraints simultaneously in the in terms of a linear matrix equation. The corresponding net-
study of the consensus problem. work stability can be analyzed by investigating the properties
of the system matrices constructed based on the proposed
consensus algorithms and the given network topology. Various
D. Sampled-data Framework approaches, including Lyapunov functions [127], [131], matrix
The previous three subsections describe the main research theory [115], [132], [135], [136], stochastic matrices [133],
work in the study of the consensus problem. The following and linear matrix inequalities [130], [131], have been adopted,
introduces a few other aspects, namely, sampled-data frame- and some necessary and/or sufficient conditions have been
work, quantization, asynchronous effect, convergence speed, derived for guaranteeing sampled-data consensus.
and finite-time convergence, that have been considered in the It is natural to consider the sampled-data effect for consen-
consensus problem as well. Among these topics, sampled- sus with general linear or nonlinear dynamics (see, e.g., [137]).
data framework, quantization, and asynchronous effects are In addition, it is meaningful to consider the case when all
considered due to some physical limitations in practical sys- vehicles do not necessarily share the same sampling period or
tems while convergence speed and finite-time convergence are the sampling period is not necessarily constant. Accordingly,
concerned with the performance for some proposed consensus it is expected that a careful design of the sampling periods
algorithms. (associated with the proposed algorithms) might lead to the
Due to the limitations in the measurement and control units, optimization of the closed-loop systems under the proposed
it is often impossible to acquire information measurements algorithms subject to certain cost functions, such as maximum
at an arbitrarily fast speed and to execute the control in- convergence rate and minimum total information exchange. In
puts instantaneously. Accordingly, the closed-loop systems are another word, it is intriguing to move from analysis to design
modeled in a hybrid fashion. That is, the system plant is when investigating the consensus problem in a sampled-data
described in a continuous-time setting while the measurements framework.
and control inputs are described in a piecewise constant
fashion. For instance, in a sampled-data setting, (2) becomes
E. Asynchronous Effects
n
X In most existing research of the consensus problem, it is
ui (t) = ui (kT ) = aij (kT )[xj (kT ) − xi (kT )] (10)
assumed that all agents update their states synchronously,
j=1
which requires a synchronized clock for the whole group of
for kT ≤ t < (k + 1)T , where T is the sampling period agents. However, such a synchronized clock might not exist
and k is the discrete-time index. Essentially, (10) is a zero- in real applications. This motivates the design of consensus
order-hold version of ui (t) in the sense that the control inputs algorithms in an asynchronous fashion; that is, each agent
 7

updates its own states regardless of the update times of other where it was shown that the convergence rate depends on
agents. the accuracy of the quantization but not the coding/decoding
In most studies of the asynchronous consensus problem schemes. In [143], [144], quantized consensus was studied via
for networked systems, due to the intrinsic technical dif- the gossip algorithm, with both lower and upper bounds of the
ficulties, usually only single-integrator kinematics (1) and convergence time derived in terms of the network size. Further
double-integrator dynamics are considered. In [103], such results regarding quantized consensus were reported in [145]–
an asynchronous consensus problem with time delay was [152], where the main research was also on the convergence
investigated by utilizing some basic properties of stochastic time for various proposed quantized consensus algorithms as
matrices. Similarly in [138], the asynchronous consensus prob- well as the quantization effects on the convergence time. It is
lem was studied by using matrix theory and graph theory, and intuitively reasonable that the convergence time depends on
in [139], by employing the paracontracting theory. In [140], both the quantization level and the network topology. It is
the authors studied the asynchronous consensus problem for then natural to ask if and how the quantization methods affect
double-integrator dynamics and presented sufficient conditions the convergence time. This is an important measure of the
to guarantee consensus, where a condition based on linear robustness of a quantized consensus algorithm (with respect
matrix inequalities was given. to the quantization method).
Note that consensus in an asynchronous fashion has been Note that it is interesting but also more challenging to study
considered mainly for single-integrator kinematics and double- consensus for general linear/nonlinear systems with quantiza-
integrator dynamics but not for other system dynamics. For tion. Because the difference between the truncated signal and
certain linear systems, it might be expected that asynchronous the original signal is bounded, consensus with quantization
communication does not affect the consensusability as shown can be considered as a special case of one without quantization
in [103], [138] for single-integrator kinematics. However, a when there exist bounded disturbances. Therefore, if consensus
similar conclusion may not hold for systems with general can be achieved for a group of vehicles in the absence of
dynamics, especially nonlinear dynamics. It is important to quantization, it might be intuitively correct to say that the
quantify the effects of the asynchronous communication on differences among the states of all vehicles will be bounded
the consensus problem. if the quantization precision is small enough. However, it is
still an open question to rigorously describe the quantization
effects on consensus with general linear/nonlinear systems.
F. Quantization
Quantized consensus has been studied recently with motiva- G. Convergence Speed
tion from digital signal processing. Here, quantized consensus In addition to the study on the consensus problem with
refers to consensus when the measurements are digital rather physical constraints mentioned in the previous subsections,
than analog therefore the information received by each agent it is also important to study the control performance of the
is not continuous and might have been truncated due to digital consensus problem. From the control’s perspective, it is natural
finite precision constraints. Roughly speaking, for an analog to propose proper control algorithms and analyze the stabil-
signal s, a typical quantizer with an accuracy parameter δ, ity, and to optimize the proposed control algorithms under
also referred to as quantization step size, is described by certain control performance indexes. In this subsection, the
Q(s) = q(s, δ), where Q(s) is the quantized signal and q(·, ·) convergence speed problem is reviewed, which is an important
is the associated quantization function. For instance [141], a performance measure for consensus algorithms.
quantizer rounding a signal s to its nearest integer can be For (1) using algorithm (2) in a connected undirected graph,
expressed as Q(s) = n, if s ∈ [(n−1/2)δ, (n+1/2)δ], n ∈ Z, the worst-case convergence speed was shown in [8] to be the
T
where Z denotes the integer set. Note that the types of Laplacian spectral gap minX6=0,1T X=0 XkXk LX
2 = λ2 , where
quantizers might be different for different systems, hence Q(s) 0 is an all-zero column vector, X = [x1 , · · · , xn ]T , L is the
may differ for different systems. Due to the truncation of Laplacian matrix with λ2 being smallest nonzero eigenvalue.
the signals received, consensus is now considered achieved if Here, one should recall that the smallest eigenvalue of a
the maximal state difference is not larger than the accuracy Laplacian matrix for a connected undirected graph is zero and
level associated with the whole system. A notable feature all the other eigenvalues are positive.
for consensus with quantization is that the time to reach In order to increase the convergence speed, the above
consensus is usually finite. That is, it often takes a finite spectral gap should be enlarged. For this purpose, an iterative
period of time for all agents’ states to converge to an accuracy algorithm was proposed in [153] to maximize the above
interval. Accordingly, the main research is to investigate the spectral gap, by employing a semidefinite programming solver.
convergence time associated with the proposed consensus Other than the smallest nonzero eigenvalue of the Laplacian
algorithm. matrix, another commonly used measure for the convergence
Quantized consensus was probably first studied in [141], speed is the following ratio, introduced in [154], [155]:
where a quantized gossip algorithm was proposed and its 1/t
kX(t) − X ⋆ k

convergence was analyzed. In particular, the bound of the ρ= lim , (11)
convergence time for a complete graph was shown to be poly- t→∞,X(t)6=X ⋆ kX(0) − X ⋆ k
nomial in the network size. In [142], coding/decoding strate- where X ⋆ represents the final equilibrium given by σ1, where
gies were introduced to the quantized consensus algorithms, σ is a constant.
 8

In [154], this problem of finding the fastest convergence I. Remarks

speed was casted into a semidefinite programming problem.
Furthermore, the convergence speed defined by (11) was In summary, the existing research on the consensus problem
studied in both deterministic and stochastic settings. In the has covered a number of physical properties for practical
deterministic setting, it was studied in [155]–[157] with esti- systems and control performance analysis. However, the study
mation of lower bounds. In the stochastic setting, this problem of the consensus problem covering multiple physical properties
was studied in [26], [29], [158], with a per-step convergence and/or control performance analysis has been largely ignored.
factor introduced and discussed in [158], which itself can be In other words, two or more problems discussed in the above
considered a measure of the convergence speed. subsections might need to be taken into consideration simul-
taneously when studying the consensus problem. In addition,
The existing study mainly focuses on the analysis of the
consensus algorithms normally guarantee the agreement of
convergence speed under various network topologies and op-
a team of agents on some common states without taking
timization of the convergence speed for certain given network
group formation into consideration. To reflect many practical
topologies. Considering the fact that consensus under different
applications where a group of agents are normally required
network topologies may demonstrate different convergence
to form some preferred geometric structure, it is desirable to
speeds, a natural question is how to design an optimal (switch-
consider a task-oriented formation control problem for a group
ing) network topology with proper adjacency matrix such that
of mobile agents, which motivates the study of formation
optimal convergence speed can be achieved.
control presented in the next section.

H. Finite-time Convergence IV. F ORMATION C ONTROL

As an extension of the study of convergence speed for the Compared with the consensus problem where the final
consensus problem, finite-time consensus, reaching consensus states of all agents typically reach a singleton, the final states
in a finite time, has also been studied recently. Compared of all agents can be more diversified under the formation
with most existing research on the consensus problem, finite- control scenario. Indeed, formation control is more desirable
time consensus demonstrates a disturbance rejection property in many practical applications such as formation flying, co-
and robustness against uncertainties. In addition, due to the operative transportation, sensor networks, as well as combat
finite-time convergence, it is often possible to decouple the intelligence, surveillance, and reconnaissance. In addition, the
consensus problem from other control objectives when they performance of a team of agents working cooperatively often
are considered simultaneously. exceeds the simple integration of the performances of all
For a group of n agents, e.g. with dynamics (1), the individual agents. For its broad applications and advantages,
objective is to design ui (t) such that xi (t) = xj (t) for t ≥ T , formation control has been a very active research subject in
where T is a constant. Here, T is called the consensus time. the control systems community, where a certain geometric
Finite-time consensus for networked systems with single- pattern is aimed to form with or without a group reference.
integrator kinematics (1) in the continuous-time setting was More precisely, the main objective of formation control is
solved in [86], [88], [159]–[163]. Finite-time consensus for to coordinate a group of agents such that they can achieve
networked systems with double-integrator dynamics in the some desired formation so that some tasks can be finished by
continuous-time setting was studied in [164]. An important the collaboration of the agents. Generally speaking, formation
common characteristic of the proposed algorithms is the use control can be categorized according to the group reference.
of the signum function. It is well known that linear consensus Formation control without a group reference, called formation
algorithms can normally guarantee asymptotic convergence, producing, refers to the algorithm design for a group of agents
but not finite-time convergence. On the contrary, the finite-time to reach some pre-desired geometric pattern in the absence
consensus algorithms developed in [86], [88], [159]–[161], of a group reference, which can also be considered as the
[163], [164], which utilize the signum function, are able to control objective. Formation control with a group reference,
do so. called formation tracking, refers to the same task but following
Note that the existing research on finite-time consensus the predesignated group reference. Due to the existence of
mainly focuses on systems with simple dynamics, such as the group reference, formation tracking is usually much more
single-integrator kinematics and double-integrator dynamics, challenging than formation producing and control algorithms
in the continuous-time setting. Because many practical sys- for the latter might not be useful for the former. As of today,
tems are better and more proper to be described by general there are still many open questions in solving the formation
linear/nonlinear dynamics, it is natural to study finite-time tracking problem.
consensus for systems with general linear/nonlinear dynamics The following part of the section reviews and discusses
in the future. In addition, it is interesting to study finite-time recent research results and progress in formation control,
consensus in the discrete-time setting. Some recent work on including formation producing, formation tracking, and con-
this topic can be found in [165], [166], where the objective is nectivity maintenance for consensus and formation control,
to compute the final consensus value for all agents in a finite mainly accomplished after 2006. Several milestone results
number of steps. prior to 2006 can be found in [167]–[169].
 9

A. Formation Producing stabilization [178]–[188], and circular formation alike [175],

The existing work in formation producing aims at analyzing [176], [189]–[192]. In the inverse agreement problem [177],
the formation behavior under certain control laws, along with the objective is to force a team of agents to disperse in
stability analysis. space. Roughly speaking, for the single-integrator kinemat-
1) Matrix Theory Approach: Due to the nature of multi- ics (1), thePncorresponding control input takes the form of
agent systems, matrix theory has been frequently used in the ui (t) = j=1 bij (kxi − xj k)[xi (t) − xj (t)], where bij (·)
stability analysis of their distributed coordination. is a nonnegative function. Assuming that each agent can
Note that consensus input to each agent (see e.g., (2)) is communicate with all other agents within a radius R, the
essentially a weighted average of the differences between the agents will disperse in space with the relative distance between
states of the agent’s neighbors and its own. As an extension any two agents being larger than R.
of the consensus algorithms, some coupling matrices were For the case of leaderless flocking, research has been
introduced here to offset the corresponding control inputs by conducted to stabilize a group of agents towards some de-
some angles [170]–[173]. For example, given (1), the control sired geometric formation, where the inter-agent interaction is
input (2) is revised as described directly or indirectly by some nonnegative potential
function Vij (kxi − xj k) regardless of the final group veloc-
n
X ity. Some notable properties of Vij (kxi − xj k) includes: (i)
ui (t) = aij (t)C[xj (t) − xi (t)], (12)
Vij (kxi − xj k) achieves its minimum when kxi − xj k is equal
j=1
to the desired inter-agent distance between agents i and j,
where C is a coupling matrix with compatible size. If xi ∈ R3 , (ii) Vij (kxi − xj k) increases as kxi − xj k decreases from the
then C can be viewed as the 3-D rotational matrix. The main desired distance to zero and Vij (kxi − xj k) could approach
idea behind (12) is that the original control input for reaching infinity as kxi − xj k approaches zero, and (iii) Vij (kxi − xj k)
consensus is now rotated by some angles. The closed-loop increases as kxi − xj k increases from the desired distance to
system can be expressed in a vector form, whose stability can the maximum communication range R. The basic idea behind
be determined by studying the distribution of the eigenvalues the potential function Vij (kxi − xj k) is to drive the inter-
of a certain transfer matrix. Main research work was conducted agent distance to the desired value while avoiding possible
in [170]–[173] to design proper algorithms and analyze the inter-agent collision. The corresponding control law for each
collective motions for systems with single-integrator kinemat- agent is usually chosen to be the same as the corresponding
ics and double-integrator dynamics. Analogously, the collec- consensus algorithm except that the xi − xj term is replaced
tive motions for a team of nonlinear self-propelling agents by ∇xi V (kxi − xj k) here. A fundamental limitation is that
were shown to be affected by the coupling strength among the all agents will normally converge to some (constant) inter-
agents, the time delay, the noise, and the initial states [174]. agent configuration locally in the sense that some nonnegative
Note that the collective motions for nonholonomic mobile potential function achieves its local minimum. Accordingly,
robots were also studied recently in, e.g., [175], [176]. Al- the inter-agent distance might not converge to the desired value
though the study in [170]–[173] is different from that in [175], globally. It is an interesting future research topic to study how
[176], similarities exist in the sense that all agents will not to ensure the desired inter-agent distance be achieved globally
move to the weighted average of the states of neighboring under a properly designed control algorithm. In addition, the
agents, but to some offsetted state. Noticeably, the offsetted network topology associated with a team of agents is usually
state in [170]–[173] is properly designed, yet the one in [175], assumed to be undirected, which is not applicable to many
[176] is induced by some special nonlinear system dynamics. practical systems which are described by directed networks.
In the study of formation producing with linear closed-loop For the case of circular formation and the like, the main
systems, it is observed that the associated system matrix has research in [175], [176], [189], [191], [193] was devoted to
two interesting properties: (1) the existence of at least one the collective motion for nonhonolomic mobile robots√with
zero eigenvalue, and (2) the existence of at least one pair of dynamics given in (6). Denote ri = xi + ιyi , where ι = −1.
eigenvalues on the imaginary axis. The two properties play Then, (6) becomes ṙi = ui eιθi , θ̇i = ωi , i = 1, · · · , N.
an important role in the formation producing problem under Due to the nature of the nonlinear dynamics, a consensus-
a fixed network topology. However, the two properties might like algorithm often renders a circular-like ultimate formation
not be able to solve the formation producing problem under where the trajectories of all agents are circular and the relative
a switching network topology, which remains a challenging phase difference (namely, θi − θj ) is constant. The current
problem due to the complexity in the analysis of switching work mainly focuses on the case when all agents share
systems. a common unit speed. Similar circular-like formation was
2) Lyapunov Function Approach: Although matrix theory analyzed in [190], [192], where the system dynamics are
is a relatively simple approach for stability analysis of the different from (6) but share a similar nonlinearity. Due to the
formation producing problem, it is not applicable in many for- nonlinearity of the system dynamics, it is a challenging task
mation producing scenarios, especially with nonlinear systems. to incorporate time delay, disturbances, quantization, etc, into
It is then natural to consider the Lyapunov function approach. the existing research.
By using the Lyapunov function approach, several typical 3) Graph Rigidity: For a network with a given number
formation producing scenarios have been studied, including of agents, the edges are closely related to the shape of
the inverse agreement problem [177], leaderless flocking and formation. Roughly speaking, if enough information regarding
 10

edge distances for a team of agents is available, the geometric worth mentioning that the group reference can be arbitrarily
structure of all agents is determined. Then the graph for the chosen as long as its derivative is bounded. In [205]–[207],
agents is rigid. According to [194], a graph of n agents is the synchronization of a group of linear systems to the output
rigid if at least 2n − 3 edge distances are available. of another linear exosystem was investigated with or without
Motivated by the graph rigidity, research has been conducted parameter uncertainties. In [203], [204], a general group
in [195]–[199] to drive a group of agents to the desired inter- reference was discussed while in [205]–[207] a general system
agent configuration by ensuring that a certain number of edge model was considered. How to solve formation tracking for
distances are identical to the desired ones. The graph rigidity general linear systems with a general group reference is still
recovery after loss of an agent has also been investigated. an open problem.
Compared with other formation producing algorithms which The formation tracking problem can be converted to a tradi-
require edge vector information (i.e., xi −xj ), less information tional stability problem by redefining the variables as the errors
is required in edge distance information (i.e., kxi − xj k). As between each agent’s state and the group reference. Then,
a tradeoff, some unstable equilibria, such as a collinear initial the formation tracking problem is solved if the corresponding
configuration (the initial states of all agents are linearly depen- errors can be driven to zero. However, the formation producing
dent) and a common initial state (the initial states of all agents problem, in general, cannot be solved in this way. Therefore,
are identical), rather than the desired inter-agent configuration, under a switching network topology, the formation tracking
might exist. In practical applications, it is important to design problem is generally easier to manager than the formation
proper control algorithms such that a team of agents can avoid producing problem.
converging to the unstable equilibria. 2) Lyapunov Function Approach: Due to the broad applica-
4) Receding Horizon Approach: Receding horizon con- tions of the Lyapunov function approach in stability analysis,
trol (RHC), known also as model predictive control, has it has become an important tool in the study of the formation
been introduced into the formation stabilization problem. By tracking problem as well.
nature, RHC is a finite-horizon optimization problem. The Flocking with a dynamic group reference has been stud-
employment of RHC in the formation stabilization problem ied recently [208]–[212], where the objective is to design
is motivated by the fact that RHC is more capable of dealing distributed control algorithms such that a team of agents
with constraints. move cohesively along the group reference. Compared with
The main research on this topic [200]–[202] has been leaderless flocking where no specific final group velocity is
devoted to deriving proper distributed control algorithms for required, the study of flocking with a dynamic group reference
a team of agents such that they can reach some desired is much more challenging both theoretically and technically
formation by optimizing some finite-horizon cost functions due to the existence of the dynamic group reference and the
in scenarios with or without time delay. Because RHC is requirement on the cohesive movement of the agents along
essentially an optimization-based control strategy, the dis- the dynamic group reference. In other words, the agents not
tributed control algorithms are typically given by solving only have to maintain some desired geometric formation but
optimization problems. Therefore, more computational time also need to follow the group reference as a whole. The
is required by RHC than other control approaches. Therefore, combination of the two objectives makes the problem much
the potential computation-induced time delay needs to be taken more difficult than the leaderless flocking problem where only
into consideration in practical applications. the first objective is involved. If enough information of the
group reference is known, such as the acceleration and/or
velocity information of the group reference, flocking with
B. Formation Tracking a dynamic group reference can be solved by employing a
Although formation control without a group reference is gradient-based control law [208]–[210]. Another approach was
interesting in theory, it is more realistic to study formation proposed in [211], where a variable structure-based control
control in the presence of a group reference because it may law was used to solve the problem with less information
represent a control objective or a common interest of the whole required. Similarly to the study of the leaderless flocking
group. This scenario is now reviewed in this subsection. problem, the existing approaches on flocking with a dy-
1) Matrix Theory Approach: Similarly to the case of for- namic group reference can only reach a local minimization
mation producing, matrix theory is often used in the study of of certain potential functions because the potential function
the formation tracking problem. is generally unspecified but satisfies the conditions stated in
An interesting problem in formation tracking is to design a Subsection IV-A. Accordingly, the inter-agent distance is not
distributed control algorithm to drive a team of agents to track identical to the desired one. However, the potential-based
some desired state. For example, given the single-integrator control can be easily designed to guarantee collision avoidance
kinematics, control algorithms were designed in [203], [204], and maintain the initial inter-agent communication patterns.
where the algorithms are similar to those consensus algorithms Nevertheless, it is still an open problem to accomplish the
except for that an extra term is introduced here due to the task with global inter-agent distance stabilization, collision
existence of the group reference. If properly designed, all avoidance, and initial communication pattern maintenance.
agents can track the group reference accurately as reported Formation control with a group reference was studied in
in [203], with bounded tracking errors analyzed in [204], both linear systems [213]–[215] and nonlinear systems [216]–
where a discretized version in [203] was considered. It is [224] when the potential function V (kxi − xj k) is re-
 11

placed by some known functions, generally in the form of D. Remarks

2
kxi − xj − dij k , where dij denotes the desired distance be- Current research on formation control mainly focuses on
tween agents i and j. Briefly, the nonlinear systems studied in a fixed formation where the inter-agent distance is fixed.
this case include nonholonomic mobile robots (see (6)) [218]– Considering practical applications, however, it might require
[222], rigid bodies (see (7)) [223], [224], and linear systems the formation be adaptive with respect to the events performed
with other nonlinear terms [216], [217]. Compared with the by the team of agents. In addition, it is important to consider
flocking problem, the problem studied here is relatively easier constraints, such as input saturation, quantization, and power
due to the known V (kxi − xj k). In general, the inter-agent limitation, in the formation control problem. Meanwhile, the
distance can be driven to the desired one. As a tradeoff, robustness is another important factor that deserves consider-
the collision avoidance and initial communication pattern able attention in real applications where noise and disturbances
maintenance need to be considered separately. exist.
In terms of connectivity maintenance for consensus and for-
C. Connectivity Maintenance for Consensus and Formation
mation control, research has been devoted to continuous-time
Control
systems but practical systems are more suitable to be modeled
In both consensus and formation control problems, it is in a discrete-time setting, which makes the study of connec-
often assumed that the network topology satisfies certain tivity maintenance for discrete-time systems more meaningful.
fundamental conditions, for example, is connected or has a In general, the connectivity maintenance for discrete-time
directed spanning tree. However, a practical communication systems is more challenging due to the fundamental limitation
model is typically distance-based, i.e., two agents can commu- of the corresponding control input, which is usually piecewise
nicate with each other if and only if their distance is smaller constant rather than continuous.
than a certain threshold, called communication range. This is
particularly true for sensor networks. In order to guarantee
V. O PTIMIZATION
consensus or formation control be achieved asymptotically, a
connectivity maintenance mechanism is essential, which has Optimization is an important subject in the studies of control
been studied recently. systems. The main objective of optimization is to find an
The main approach to maintaining the connectivity of a optimal strategy subject to some given constraints such as
team of agents is to define some artificial potentials (between cost functions. Recently optimization in distributed multi-
any pair of agents) in a proper way such that if two agents are agent coordination has been studied concerning convergence
neighbors initially then they will always communicate with speed and some specific cost functions, which are respectively
each other thereafter [211], [225]–[234]. In general, the artifi- reviewed below.
cial potential between a pair of agents grows to be sufficiently
large (could be unbounded) when the distance between them A. Convergence Speed
increases to be equal to the communication range. For properly
designed control algorithms, which are usually composed of As discussed above, one important problem in consensus is
the gradients of the artificial potentials, the total artificial the convergence speed, which characterizes how fast consensus
potential is nonincreasing. This then indicates that the initial can be achieved therefore is desirable to optimize. In this
communication patterns can be preserved because otherwise regard, the convergence speed is the cost function to be
the total potential will become larger than the initial total maximized.
artificial potential, as soon as some communication pattern Consider a group of n agents with dynamics described by
is broken. Although this approach provides a systematic way the single-integrator kinematics (1). Equipped with (2), the
to guarantee the connectivity, the corresponding control al- dynamical system (1) can be written in a matrix form, as
gorithms might require infinite large control inputs, which is Ẋ(t) = −LX(t), (13)
not practical. Meanwhile, it is not even necessary to always
maintain the initial communication patterns in order to guar- where X(t) = [x1 (t), · · · , xn (t)]T and L is the Laplacian
antee the connectivity. Therefore, how to find a more effective matrix. For a network with a fixed topology, L is a constant
way to guarantee connectivity deserves further investigation. matrix.
In contrast to the studies in [211], [225]–[234], the authors Motivated by the observation that the smallest nonzero
in [235] investigated an interesting problem where the number eigenvalue of the Laplacian matrix, λ2 (L), determines the
of initially existing communication patterns plays a role in worst-case convergence speed [8], research has been con-
the connectivity maintenance for the consensus problem with ducted to maximize the convergence speed [153], [236] by
single-integrator kinematics (1) and control input (2). Roughly choosing optimal weights associated with edges. In contrast
speaking, if the initial graph is “sufficiently” connected in to [153], [236], where the systems are assumed to have single-
the sense that each agent has at least a certain number integrator kinematics, optimization of the convergence speed
of neighbors, consensus can be guaranteed to be achieved. for double-integrator dynamics was considered in [237], where
Note that the result can only be applied to systems with the convergence speed is defined in a similar way to the λ2 (L)
single-integrator kinematics therefore further investigation is for the case with single-integrator kinematics. It is worth
expected for systems with high-order linear dynamics or mentioning that optimal convergence for general linear and
nonlinear dynamics. nonlinear systems is still an open problem.
 12

Other than λ2 (L), another commonly used measure for algorithm (15) can only find sub-optimal solutions, determined
the convergence speed is given by (11). The corresponding by the constant step size α. Further results on this topic can be
optimization problem is found in [241]–[243], where a similar distributed multi-agent
optimization problem was studied within various scenarios,
max ρ, (14) under constraints [241], over random networks [242], and
ui ∈U
with broadcast-based communications in an asynchronous
where ρ is defined in (11) and U is a set of admissible
setting [243]. In the existing literature, time delay and dis-
controllers. Existing research in [154], [238] focuses on the
turbances have not been taken into consideration. Therefore,
case when all agentsP converge to the average of the initial
it is important to consider time delay and disturbances in the
states, i.e., X ⋆ = [ n1 ni=1 xi (0)]1. For an arbitrary fixed or
distributed multi-agent optimization problem due to their wide
switching network topology, the optimization problem (14)
existence in wireless sensor networks.
is challenging if X ⋆ is unknown. But if X ⋆ is chosen as
1
Pn In addition to the distributed multi-agent optimization prob-
[ n i=1 xi (t)]1, then the problem becomes much easier.
lem, where the cost function is a sum of a series of convex
functions, distributed optimization has also been considered
B. Specific Cost Functions for both
R ∞infinite-horizon cost functions [244]–[248] given by
In addition to the fastest convergence speed requirement, Ji = 0 [X T (t)QX(t) + U T (t)RU (t)]dt and finite-horizon
Rt
various cost functions are also subject to minimization. cost functions [249]–[252] given by Jf = 0 f [X T (t)QX(t)+
One interesting problem studied in this setting is distributed U T (t)RU (t)]dt, where X ∈ Rn is the state, U ∈ Rn is the
multi-agent optimization, which is motivated by solving one control input, and tf is a positive constant. It is worth mention-
challenge in wireless sensor networks, namely, to estimate ing that the RHC approach discussed in Section IV-A typically
the environment parameters and/or some functions of inter- has finite-horizon cost functions. Differing from the research
est, such as temperature and source location [239]. As a reported in [240]–[243], which is to find the optimal estimated
simple strategy, each sensor node can send its data to an state, the objective here is to find the optimal control laws
existing central location which can then process the data subject to the minimization of certain cost functions. Due to
if it is sufficiently powered. However, due to the limited requirements of optimizing the cost functions when designing
power resources and communication capabilities, this strategy the control laws, the computational complexity becomes an
is often not applicable. An alternative approach to achieving important problem to study. Meanwhile, the network topology
a similar objective is to estimate the environment parameters plays a significant role in the optimization problem with
and/or some functions of interest locally, which requires much certain cost functions, therefore it is also important to optimize
less communication bandwidth and power. In wireless sensor the network topology subject to certain cost functions.
networks [239], the estimation problem is usually modeled
as a distributed multi-agent optimization problem. Roughly VI. D ISTRIBUTED TASK A SSIGNMENT
speaking, the objective of distributed multi-agent optimization
Pn Distributed task assignment refers to the study of task
is to cooperatively minimize the cost function i=1 fi (x), assignment of a group of dynamical agents in a distributed
where the function fi : Rn 7→ R represents the cost of agent manner, which can be roughly categorized into coverage
i, known by this agent only, and x ∈ Rm is a decision vector. control, scheduling, and surveillance. Compared with the pre-
In [239], an incremental subgradient approach was used to vious studies discussed in Sections III, IV, and V, distributed
solve the optimization problem for a ring type of network. It task assignment focuses on the three task-oriented research
should be noted that [239] does not provide much discussion problems, where each topic has its unique features.
on the optimization problem under other types of network
topologies.
A. Coverage Control
Ref. [240] was probably the first paper studying the dis-
tributed multi-agent optimization problem under a consensus- Coverage control is an active research direction in mobile
based framework. The problem considered therein is formu- sensor networks, where the objective is to properly assign the
lated so as to mobile sensors’ motion in order to maximize the detection
Pn probability.
minimize i=1 fi (x), Let Q be a convex space with φ represent the distribution
s.t. x ∈ Rn , density function which indicates the probability that some
where each fi : Rn → R is assumed to be a convex function. event takes place over Q [253]. Consider a group of n mobile
Inspired by the average consensus algorithm and the standard sensors whose locations are specified by P = [p1 , · · · , pn ].
subgradient method, a consensus-like algorithm was proposed The sensor performance at a point q degrades with respect
as to the distances kq − pi k, which are all described by a
n
X nondecreasing differentiable function, f . The coverage con-
xi (k + 1) = aij (k)xj (k) − αgi (xi (k)), (15)
trol problem is to find a local controller for each mobile
j=1
sensorPsuch Rthat the following cost function is minimized:
n
where α is the step size and gPi (xi (k)) is the subgradient of J = i=1 f (kq − pi k)φ(q)dq . This coverage control by
n
fi (x) at x = xi (k). In [239], j=1 aij (k)xj (k) in (15) was nature is an optimization problem. Main research on this topic
replaced by xi−1 (k) with x0 (k) = xn (k − 1). Note that the was reported in, e.g., [254]–[261], where the coverage control
 13

problem was considered in two directions, namely, analysis to the power constraints, each agent might be subjected to
of coverage control under various practical constraints, such constraints such as bounded control input, limited distance to
as limited sensing/communication capacities [254], load bal- travel, and finite accuracy level, etc., therefore it is interesting
ancing [260], and nonholonomic mobile robots [261], and and important to consider these limitations in the distributed
algorithms for coverage control [257]. Noting that time delay surveillance problem in the future.
and uncertainties have not been considered in the coverage
control problem, it is interesting to consider the effect of VII. E STIMATION
time delay and uncertainties in the coverage control problem.
Due to the absence of global information, used for achieving
Moreover, the density function φ might be time-dependent in
group coordination in many cases, a distributed estimation
real systems, which is another interesting research topic for
scheme is often needed for estimation.
further study.
The first problem is to design local distributed estimators
such that some global information can be estimated asymp-
B. Scheduling totically or in finite time. The second problem is to design
Another interesting topic in distributed task assignment is local controllers based on the local estimators such that the
distributed scheduling, which refers to the scheduling of a closed-loop system is stable. The estimation-based distributed
group of dynamical agents in a distributed manner. Distributed control is essentially a combination of both centralized control
scheduling has many potential applications in military and and distributed control in such a way that distributed control
civilian sectors, and can be roughly categorized into two is used in the estimation of some global information and the
typical problems, namely sequence optimization [262] and task centralized control idea is used for local controllers design.
allocation [263]–[267]. The objective of sequence optimization The estimation-based distributed control strategy often inherits
is to schedule a team of agents such that some metrics can the merits of both centralized control and distributed control.
be optimized. For instance, in [262], an optimal scheduling However, it is worth emphasizing that a closed-loop system
sequence was designed to fuel a group of UAVs via dynamic with distributed estimators is much more complicated to
programming, where the metric is the total spending time. The design than one without distributed estimators.
objective of task allocation is to distribute certain number of Main research on this topic has been reported in [161],
tasks to a team of agents such that they can balance the total [276]–[281], where the joint estimation and control problem
tasks. If the number of tasks for each agent is considered a was considered subject to disturbances [276], [278], [279]
variable in the consensus problem, the task allocation might or without disturbances [161], [277], [280], [281]. In [161],
be viewed as a consensus problem except that a limitation [276]–[281], a joint estimation and control problem is solved
on the total number of tasks for all agents exists. A notable in the sense that the distributed estimator is used in the design
feature of the distributed task assignment problem is that of proper control algorithms such that certain global objective
various constraints may exist due to the physical properties can be achieved. Without the aid of distributed estimators,
associated with the agents. In view of the difference objec- the control design is very hard and even impossible. As
tives for sequence optimization and task allocation, it is an can be noticed from [282]–[288], the distributed estimation
interesting topic to consider combining both objectives of the problem has been considered without much discussion on
two problems simultaneously. specific control problems. In general, the joint estimation and
control idea has provided an important approach in the study
C. Surveillance of distributed multi-agent coordination where only neighbor-
based information is not sufficient for the controllers design.
Distributed surveillance means to monitor a certain area by
On the other hand, in real applications properly designed
using a group of mobile agents in distributed coordination.
distributed estimators might be used to replace some expensive
Distributed surveillance has a number of potential applications,
sensors.
such as board security guarding, forest fire monitoring, and oil
In general, it remains a challenging problem to study
spill patrolling.
task-oriented coordination control systems where the use of
The main motivation of distributed surveillance is that a
distributed estimation is either necessary or an appropriate
team of agents can monitor a given (large) area more effec-
replacement of certain expensive measurement devices, at
tively than a single agent when the team of agents works in a
the costs of difficult control system design and complex
cooperative fashion. Accordingly, an important research prob-
system stability analysis. Moreover, physical limitations such
lem in distributed surveillance is to design environment-based
as bounded control input, asynchronous communication, and
cooperative control laws for all coordinated agents such that
information quantization, could potentially reduce the appli-
the given area can be monitored efficiently. Recent research
cability of the joint estimation and control scheme in various
on distributed surveillance has been reported in [268]–[275],
distributed multi-agent coordination systems.
where a number of physical limitations were identified and
considered such as time delay and uncertainties [271], [274],
collision avoidance between agents [272], and heterogeneously VIII. D ISCUSSIONS
distributed agents [274]. The current research is conducted This article has reviewed some recent research and devel-
under the assumption that each agent has enough power such opment in distributed multi-agent coordination. In addition to
that any designed control law can be applied. However, due the theoretical results reviewed above, many experiments have
 14

also been conducted to validate the theoretical designs and • Centralization and decentralization. Although decentral-
analysis, as can be found in [289]–[294], to name just a couple ization shows obvious advantages over centralization,
of representative reports. Although the existing theoretical such as scalability and robustness, decentralization also
research and experiments have solved a number of technical has its own drawbacks. One shortcoming is that, under
problems in distributed multi-agent coordination, there are decentralized protocols, some agents cannot predict the
still many interesting, important and yet challenging research group behavior based only on the available local infor-
problems deserving further investigation. Some of them are mation. Consequently, some group behavior cannot be
briefly summarized as follows: controlled. As a sensible example, current economic cri-
sis actually illustrates some disadvantages of behavioral
decentralization. One interesting question, therefore, is
• Quantization effects in distributed coordination algo- how to balance decentralization and centralization so as
rithms. The current research efforts focus on studying to further improve the overall systems performance.
distributed coordination problems with control inputs
and measurements being analog signals with continuous R EFERENCES
values. However, digital signal processing techniques [1] N. A. Lynch, Distributed Algorithms. San Francisco, California:
require digital inputs and sampled-data measurements. Morgan Kaufmann Publishers, Inc., 1996.
Although quantization effects have been studied in several [2] M. H. DeGroot, “Reaching a consensus,” Journal of American Statis-
tical Association, vol. 69, no. 345, pp. 118–121, 1974.
coordination problems, the quantization effect in some [3] T. Vicsek, A. Czirók, E. B. Jacob, I. Cohen, and O. Schochet, “Novel
other distributed coordination problems remain unsolved type of phase transitions in a system of self-driven particles,” Physical
and even untouched. Review Letters, vol. 75, no. 6, pp. 1226–1229, 1995.
[4] J. N. Tsitsiklis, “Problems in decentralized decision making and
• Optimization with integrated individual and global cost computation,” Ph.D. dissertation, MIT, 1984.
functions. Optimization problem in distributed coordi- [5] J. N. Tsitsiklis, D. P. Bertsekas, and M. Athans, “Distributed asyn-
nation has been studied with various cost functions. In chronous deterministic and stochastic gradient optimization algo-
rithms,” IEEE Transactions on Automatic Control, vol. 31, no. 9, pp.
real systems, each individual agent has both local and 803–812, 1986.
global objectives, contributing to an integration of both [6] A. Jadbabaie, J. Lin, and A. S. Morse, “Coordination of groups
individual and global cost functions. Therefore, optimiz- of mobile autonomous agents using nearest neighbor rules,” IEEE
Transactions on Automatic Control, vol. 48, no. 6, pp. 988–1001, June
ing a combined objective is more realistic but also more 2003.
challenging. Another interesting problem is to investigate [7] J. A. Fax and R. M. Murray, “Information flow and cooperative control
how to balance the individual cost functions and the of vehicle formations,” IEEE Transactions on Automatic Control,
vol. 49, no. 9, pp. 1465–1476, September 2004.
global cost function toward a common objective. [8] R. Olfati-Saber and R. M. Murray, “Consensus problems in networks
• Intelligent coordination. Intelligent coordination refers to of agents with switching topology and time-delays,” IEEE Transactions
the distributed coordination of a team of agents in the on Automatic Control, vol. 49, no. 9, pp. 1520–1533, September 2004.
[9] W. Ren and R. W. Beard, “Consensus seeking in multiagent systems
presence of artificial intelligence, namely, each agent is under dynamically changing interaction topologies,” IEEE Transactions
intelligent in the sense that they can choose the best on Automatic Control, vol. 50, no. 5, pp. 655–661, May 2005.
possible responses based on its own objective. Intelligent [10] L. Moreau, “Stability of multi-agent systems with time-dependent com-
munication links,” IEEE Transactions on Automatic Control, vol. 50,
coordination has potential applications not only in engi- no. 2, pp. 169–182, February 2005.
neering and technology but also in economics and social [11] N. E. Leonard, D. A. Paley, F. Lekien, R. Sepulchre, D. M. Fratantoni,
studies. Although research problems, such as pursuer- and R. E. Davis, “Collective motion, sensor networks, and ocean
sampling,” Proceedings of the IEEE, vol. 95, no. 1, pp. 48–74, January
invader problem [295]–[298] and the game theory in 2007.
distributed coordination [299]–[304], have been studied [12] W. Ren, R. W. Beard, and E. M. Atkins, “Information consensus in
recently, there are still many open questions especially multivehicle cooperative control: Collective group behavior through
local interaction,” IEEE Control Systems Magazine, vol. 27, no. 2, pp.
the understanding of group behaviors in the presence 71–82.
of agent intelligence. One interesting problem is how [13] R. M. Murray, “Recent research in cooperative control of multivehicle
to interpret the underlying complex networks as well as systems,” ASME Journal of Dynamic Systems, Measurement, and
Control, vol. 129, no. 5, pp. 571–583, September 2007.
to stabilize and optimize the network in the presence of [14] R. Olfati-Saber, J. A. Fax, and R. M. Murray, “Consensus and coop-
agent intelligence. eration in networked multi-agent systems,” Proceedings of the IEEE,
• Competition and cooperation. Today, most research is vol. 95, no. 1, pp. 215–233, January 2007.
[15] P. Y. Chebotarev and R. P. Agaev, “Coordination in multiagent systems
conducted based on local cooperation but not competi- and Laplacian spectra of digraphs,” Automation and Remote Control,
tion. This posts an obvious limitation because competition vol. 70, no. 3, pp. 128–142, 2009.
not only exists but also plays an positive role in group [16] C. W. Wu, Synchronization in Complex Networks of Nonlinear Dynam-
ical Systems. World Scientific, 2007.
coordination. For example, due to the lack of competition, [17] J. Shamma, Ed., Cooperative Control of Distributed Multi-Agent Sys-
the final states of the traditional consensus algorithms are tems. Hoboken, NJ: Wiley-Interscience, 2008.
always limited to be within some region in the state space [18] W. Ren and R. W. Beard, Distributed Consensus in Multi-vehicle
Cooperative Control, ser. Communications and Control Engineering.
determined by the initial agent states. One interesting London: Springer-Verlag, 2008.
question is how to introduce competition into distributed [19] F. Bullo, J. Cortés, and S. Martı́nez, Distributed Control of Robotic
coordination so as to arrive at different desired regions Networks, ser. Applied Mathematics Series. Princeton University
Press, 2009.
and to improve the system performance by rewarding [20] Z. Qu, Cooperative Control of Dynamical Systems: Applications to
different agents with different benefits. Autonomous Vehicles. Springer-Verlag, 2009.
 15

[21] M. Mesbahi and M. Egerstedt, Graph Theoretic Methods for Multiagent [46] Z. Qu, J. Wang, and R. A. Hull, “Cooperative control of dynamical
Networks. Princeton University Press, 2010. systems with application to autonomous vehicles,” IEEE Transactions
[22] W. Ren and Y. Cao, Distributed Coordination of Multi-agent Networks: on Automatic Control, vol. 53, no. 4, pp. 894–911, 2008.
Emergent Problems, Models, and Issues, ser. Communications and [47] S. E. Tuna, “Conditions for synchronizability in arrays of coupled linear
Control Engineering. London: Springer-Verlag, 2011. systems,” IEEE Transactions on Automatic Control, vol. 54, no. 10, pp.
[23] H. Bai, M. Arcak, and J. Wen, Cooperative Control Design: A Sys- 2416–2420, 2009.
tematic, Passivity-Based Approach, ser. Communications and Control [48] L. Scardovi and R. Sepulchre, “Synchronization in networks of identi-
Engineering. New York: Springer-Verlag, 2011. cal linear systems,” Automatica, vol. 45, no. 11, pp. 2557–2562, 2009.
[24] J. Wolfowitz, “Products of indecomposable, aperiodic, stochastic ma- [49] J. H. Seo, H. Shim, and J. Back, “Consensus of high-order linear sys-
trices,” Proceedings of the American Mathematical Society, vol. 15, tems using dynamic output feedback compensator: Low gain approach,”
pp. 733–736, 1963. Automatica, vol. 45, no. 11, pp. 2659–2664, 2009.
[25] Z. Lin, B. Francis, and M. Maggiore, “Necessary and sufficient graph- [50] Z. Li, Z. Duan, G. Chen, and L. Huang, “Consensus of multiagent sys-
ical conditions for formation control of unicycles,” IEEE Transactions tems and synchronization of complex networks: A unified viewpoint,”
on Automatic Control, vol. 50, no. 1, pp. 121–127, 2005. IEEE Transactions on Circuits And Systems–I: Regular Papers, vol. 57,
[26] Y. Hatano and M. Mesbahi, “Agreement over random networks,” IEEE no. 1, pp. 213–224, 2010.
Transactions on Automatic Control, vol. 50, no. 11, pp. 1867–1872, [51] T. Yang, S. Roy, Y. Wan, and A. Saberi, “Constructing consensus con-
November 2005. trollers for networks with identical general linear agents,” International
[27] C. W. Wu, “Synchronization and convergence of linear dynamics in Journal of Robust and Nonlinear Control, vol. 21, no. 11, pp. 1237–
random directed networks,” IEEE Transactions on Automatic Control, 1256, 2011.
vol. 51, no. 7, pp. 1207–1210, 2006. [52] G.-B. Stan and R. Sepulchre, “Analysis of interconnected oscillators by
[28] M. Porfiri and D. J. Stilwell, “Consensus seeking over random weighted dissipativity theory,” IEEE Transactions on Automatic Control, vol. 52,
directed graphs,” IEEE Transactions on Automatic Control, vol. 52, no. 2, pp. 256–270, 2007.
no. 9, pp. 1767–1773, 2007. [53] Z. Lin, B. Francis, and M. Maggiore, “State agreement for continuous-
[29] M. Akar and R. Shorten, “Distributed probabilistic synchronization time coupled nonlinear systems,” SIAM Journal on Control and Opti-
algorithms for communication networks,” IEEE Transactions on Au- mization, vol. 46, no. 1, pp. 288–307, 2007.
tomatic Control, vol. 53, no. 1, pp. 389–393, 2008. [54] D. V. Dimarogonas and K. J. Kyriakopoulos, “A connection between
[30] A. Tahbaz-Salehi and A. Jadbabaie, “A necessary and sufficient con- formation infeasibility and velocity alignment in kinematic multi-agent
dition for consensus over random networks,” IEEE Transactions on systems,” Automatica, vol. 44, no. 10, pp. 2648–2654, 2008.
Automatic Control, vol. 53, no. 3, pp. 791–795, 2008. [55] J. Zhou, J.-A. Lu, and J. Lü, “Adaptive synchronization of an uncer-
tain complex dynamical network,” IEEE Transactions on Automatic
[31] M. Huang, S. Dey, G. N. Nair, and J. H. Manton, “Stochastic con-
Control, vol. 51, no. 4, pp. 652–656, 2006.
sensus over noisy networks with markovian and arbitrary switches,”
Automatica, vol. 46, no. 10, pp. 1571–1583, 2010. [56] ——, “Pinning adaptive synchronization of a general complex dynam-
ical network,” Automatica, vol. 44, no. 4, pp. 996–1003, 2008.
[32] M. Huang and J. H. Manton, “Stochastic consensus seeking with noisy
[57] N. Chopra and M. W. Spong, “On exponential synchronization of ku-
and directed inter-agent communication: Fixed and randomly varying
ramoto oscillators,” IEEE Transactions on Automatic Control, vol. 54,
topologies,” IEEE Transactions on Automatic Control, vol. 55, no. 1,
no. 2, pp. 353–357, 2009.
pp. 235–241, 2010.
[58] W. Yu, J. Cao, G. Chen, J. Lü, J. Han, and W. Wei, “Local synchro-
[33] A. Tahbaz-Salehi and A. Jadbabaie, “Consensus over ergodic stationary
nization of a complex network model,” IEEE Transactions on Systems,
graph processes,” IEEE Transactions on Automatic Control, vol. 55,
Man, and Cybernetics, Part B: Cybernetics, vol. 39, no. 1, pp. 230–241,
no. 1, pp. 225–230, 2010.
February 2009.
[34] G. Yin, Y. Sun, and L. Y. Wang, “Asymptotic properties of consensus-
[59] F. Chen, Z. Chen, L. Xiang, Z. Liu, and Z. Yuan, “Reaching a consensus
type algorithms for networked systems with regime-switching topolo-
via pinning control,” Automatica, vol. 45, no. 5, pp. 1215–1220, 2009.
gies,” Automatica, vol. 47, no. 7, pp. 1366–1378, 2011.
[60] Z.-G. Hou, L. Cheng, and M. Tan, “Decentralized robust adaptive
[35] Y. Zhang and Y.-P. Tian, “Consentability and protocol design of multi- control for the multiagent system consensus problem using neural
agent systems with stochastic switching topology,” Automatica, vol. 45, networks,” IEEE Transactions on Systems, Man, and Cybernetics, Part
no. 5, pp. 1195–1201, 2009. B: Cybernetics, vol. 39, no. 3, pp. 636–647, June 2009.
[36] S. Boyd, A. Ghosh, B. Prabhakar, and D. Shah, “Randomized gossip [61] P. DeLellis, M. diBernardo, and F. Garofalo, “Novel decentralized
algorithms,” IEEE Transactions on Information Theory, vol. 52, no. 6, adaptive strategies for the synchronization of complex networks,”
pp. 2508–2530, 2006. Automatica, vol. 45, no. 5, pp. 1312–1318, 2009.
[37] A. G. Dimakis, S. Kar, J. M. F. Moura, M. G. Rabbat, and A. Scaglione, [62] G. Shi and Y. Hong, “Global target aggregation and state agreement of
“Gossip algorithms for distributed signal processing,” Proceedings of nonlinear multi-agent systems with switching topologies,” Automatica,
the IEEE, vol. 98, no. 11, pp. 1847–1864, 2010. vol. 45, no. 5, pp. 1165–1175, 2009.
[38] T. Li and J.-F. Zhang, “Mean square average-consensus under measure- [63] A. Sarlette, R. Sepulchre, and N. E. Leonard, “Autonomous rigid body
ment noises and fixed topologies: necessary and sufficient conditions,” attitude synchronization,” Automatica, vol. 45, no. 2, pp. 572–577,
Automatica, vol. 45, no. 8, pp. 1929–1936, 2009. 2009.
[39] Z. Li, Z. Duan, and G. Chen, “On H∞ and H2 performance regions of [64] H. Bai, M. Arcak, and J. T. Wen, “Rigid body attitude coordination
multi-agent systems,” Automatica, vol. 47, no. 4, pp. 797–803, 2011. without inertial frame information,” Automatica, vol. 44, no. 12, pp.
[40] H. Kim, H. Shim, and J. H. Seo, “Output consensus of heterogeneous 3170–3175, 2008.
uncertain linear multi-agent systems,” IEEE Transactions on Automatic [65] S.-J. Chung and J.-J. Slotine, “Cooperative robot control and concur-
Control, vol. 56, no. 1, pp. 200–206, 2011. rent synchronization of lagrangian systems,” IEEE Transactions on
[41] R. Olfati-Saber, “Distributed kalman filter with embedded consensus Robotics, vol. 25, no. 3, pp. 686–700, 2009.
filters,” in Proceedings of the IEEE Conference on Decision and [66] S. Nair and N. E. Leonard, “Stable synchronization of mechanical
Control, Seville, Spain, December 2005, pp. 8179–8184. system networks,” SIAM Journal on Control and Optimization, vol. 47,
[42] R. Olfati-Saber and J. S. Shamma, “Consensus filters for sensor no. 2, pp. 661–683, 2008.
networks and distributed sensor fusion,” in Proceedings of the IEEE [67] J. Yao, Z.-H. Guan, and D. J. Hill, “Passivity-based control and
Conference on Decision and Control, Seville, Spain, December 2005, synchronization of general complex dynamical networks,” Automatica,
pp. 6698–6703. vol. 45, no. 9, pp. 2107–2113, 2009.
[43] R. Olfati-Saber, “Distributed kalman filtering for sensor networks,” in [68] J. Zhao, D. J. Hill, and T. Liu, “Synchronization of complex dynamical
Proceedings of the IEEE Conference on Decision and Control, New networks with switching topology: A switched system point of view,”
Orleans, LA, USA, December 2007, pp. 5492–5498. Automatica, vol. 45, no. 11, pp. 2502–2511, 2009.
[44] W. Yu, G. Chen, Z. Wang, and W. Yang, “Distributed consensus [69] P. P. Menon and C. Edwards, “Decentralised static output feedback
filtering in sensor networks,” IEEE Transactions on Systems, Man, and stabilisation and synchronisation of networks,” Automatica, vol. 45,
Cybernetics, Part B: Cybernetics, vol. 39, no. 6, pp. 1568–1577, 2009. no. 12, pp. 2910–2916, 2009.
[45] L. Shi, M. Epstein, and R. M. Murray, “Kalman filtering over a packet- [70] Y.-W. Wang, H. O. Wang, J.-W. Xiao, and Z.-H. Guan, “Synchro-
dropping network: a probabilistic perspective,” IEEE Transactions on nization of complex dynamical networks under recoverable attacks,”
Automatic Control, vol. 55, no. 3, pp. 594–604, 2010. Automatica, vol. 46, no. 1, pp. 197–203, 2010.
 16

[71] J. Lu, D. W. Ho, and J. Cao, “A unified synchronization criterion for [95] C. Godsil and G. Royle, Algebraic Graph Theory. New York:
impulsive dynamical networks,” Automatica, vol. 46, no. 7, pp. 1215– Springer-Verlag, 2001.
1221, 2010. [96] W. Ren and E. Atkins, “Distributed multi-vehicle coordinated control
[72] E. Estrada, S. Gago, and G. Caporossi, “Design of highly synchroniz- via local information exchange,” International Journal of Robust and
able and robust networks,” Automatica, vol. 46, no. 11, pp. 1835–1842, Nonlinear Control, vol. 17, no. 11, pp. 1002–1033, July 2007.
2010. [97] W. Yu, G. Chen, and M. Cao, “Some necessary and sufficient condi-
[73] A. Das and F. L. Lewis, “Distributed adaptive control for synchroniza- tions for second-order consensus in multi-agent dynamical systems,”
tion of unknown nonlinear networked systems,” Automatica, vol. 46, Automatica, vol. 46, no. 6, pp. 1089–1095, 2010.
no. 12, pp. 2014–2021, 2010. [98] W. Yu, G. Chen, M. Cao, and J. Kurths, “Second-order consensus for
[74] H. Su, G. Chen, X. Wang, and Z. Lin, “Adaptive second-order consen- multiagent systems with directed topologies and nonlinear dynamics,”
sus of networked mobile agents with nonlinear dynamics,” Automatica, IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cyber-
vol. 47, no. 2, pp. 368–375, 2011. netics, vol. 40, no. 3, pp. 881–891, 2010.
[75] W. Yu, G. Chen, and M. Cao, “Consensus in directed networks of [99] W. Yu, G. Chen, W. Ren, J. Kurths, and W. X. Zheng, “Distributed
agents with nonlinear dynamics,” IEEE Transactions on Automatic higher order consensus protocols in multiagent dynamical systems,”
Control, vol. 56, no. 6, pp. 1436–1441, 2011. IEEE Transactions on Circuits And Systems–I: Regular Papers, vol. 58,
[76] L. Scardovi, M. Arcak, and E. D. Sontag, “Synchronization of intercon- no. 8, pp. 1924–1932, 2011.
nected systems with applications to biochemical networks: An input- [100] C.-Q. Ma and J.-F. Zhang, “Necessary and sufficient conditions for
output approach,” IEEE Transactions on Automatic Control, vol. 55, consensusability of linear multi-agent systems,” IEEE Transactions on
no. 6, pp. 1367–1379, 2010. Automatic Control, vol. 55, no. 5, pp. 1263–1268, 2010.
[77] Q. Li and Z.-P. Jiang, “Global analysis of multi-agent systems based [101] W. Wang and J.-J. Slotine, “Contraction analysis of time-delayed com-
on vicsek’s model,” IEEE Transactions on Automatic Control, vol. 54, munications and group cooperation,” IEEE Transactions on Automatic
no. 12, pp. 2876–2881, 2009. Control, vol. 51, no. 4, pp. 712–717, 2006.
[78] A. Abdessameud and A. Tayebi, “Attitude synchronization of a group [102] F. Xiao and L. Wang, “Consensus protocols for discrete-time multi-
of spacecraft without velocity measurements,” IEEE Transactions on agent systems with time-varying delays,” Automatica, vol. 44, no. 10,
Automatic Control, vol. 54, no. 11, pp. 2642–2648, 2009. pp. 2577–2582, 2008.
[79] G. Chen and F. L. Lewis, “Distributed adaptive tracking control for [103] ——, “Asynchronous consensus in continuous-time multi-agent sys-
synchronization of unknown networked lagrangian systems,” IEEE tems with switching topology and time-varying delays,” IEEE Trans-
Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, actions on Automatic Control, vol. 53, no. 8, pp. 1804–1816, 2008.
vol. 41, no. 3, pp. 805–816, June 2011. [104] Y. G. Sun, L. Wang, and G. Xie, “Average consensus in networks of
[80] H. Zhang, T. Ma, G.-B. Huang, and Z. Wang, “Robust global expo- dynamic agents with switching topologies and multiple time-varying
nential synchronization of uncertain chaotic delayed neural networks delays,” Systems and Control Letters, vol. 57, no. 2, pp. 175–183, 2008.
via dual-stage impulsive control,” IEEE Transactions on Systems, Man, [105] P. Lin, Y. Jia, and L. Li, “Distributed robust H∞ consensus control
and Cybernetics, Part B: Cybernetics, vol. 40, no. 3, pp. 831–844, June in directed networks of agents with time-delay,” Systems and Control
2010. Letters, vol. 57, no. 8, pp. 643–653, 2008.
[81] J. Lu and D. W. C. Ho, “Globally exponential synchronization and [106] Y.-P. Tian and C.-L. Liu, “Consensus of multi-agent systems with
synchronizability for general dynamical networks,” IEEE Transactions diverse input and communication delays,” IEEE Transactions on Au-
on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 40, no. 2, tomatic Control, vol. 53, no. 9, pp. 2122–2128, 2008.
pp. 350–361, April 2010. [107] Y. G. Sun and L. Wang, “Consensus of multi-agent systems in directed
networks with nonuniform time-varying delays,” IEEE Transactions on
[82] L. Wang and X. Wang, “New conditions for synchronization in dynam-
Automatic Control, vol. 54, no. 7, pp. 1607–1613, 2009.
ical communication networks,” Systems and Control Letters, vol. 60,
[108] U. Münz, A. Papachristodoulou, and F. Allgower, “Delay robustness
no. 4, pp. 219–225, 2011.
in consensus problems,” Automatica, vol. 46, no. 8, pp. 1252–1265,
[83] N. D. Powel and K. A. Morgansen, “Communication-based perfor-
2010.
mance bounds in nonlinear coordinated control,” International Journal
[109] X. Liu, W. Lu, and T. Chen, “Consensus of multi-agent systems
of Robust and Nonlinear Control, vol. 21, no. 12, pp. 1410–1420, 2011.
with unbounded time-varying delays,” IEEE Transactions on Automatic
[84] Y.-W. Wang, J.-W. Xiao, and H. O. Wang, “Global synchronization Control, vol. 55, no. 10, pp. 2396–2401, 2010.
of complex dynamical networks with network failures,” International [110] A. Papachristodoulou, A. Jadbabaie, and U. Münz, “Effects of delay
Journal of Robust and Nonlinear Control, vol. 20, no. 15, pp. 1667– in multi-agent consensus and oscillator synchronization,” IEEE Trans-
1677, 2010. actions on Automatic Control, vol. 55, no. 6, pp. 1471–1477, 2010.
[85] W. Ren, “Distributed leaderless consensus algorithms for networked [111] P. Lin and Y. Jia, “Multi-agent consensus with diverse time-delays and
EulerLagrange systems,” International Journal of Control, vol. 82, jointly-connected topologies,” Automatica, vol. 47, no. 4, pp. 848–856,
no. 11, pp. 2137–2149, 2009. 2011.
[86] Q. Hui, W. M. Haddad, and S. P. Bhat, “Finite-time semistability and [112] Y.-P. Tian and C.-L. Liu, “Robust consensus of multi-agent systems
consensus for nonlinear dynamical networks,” IEEE Transactions on with diverse input delays and asymmetric interconnection perturba-
Automatic Control, vol. 53, no. 8, pp. 1887–1890, 2008. tions,” Automatica, vol. 45, no. 5, pp. 1347–1353, 2009.
[87] Q. Hui and W. M. Haddad, “Distributed nonlinear control algorithms [113] P. Lin and Y. Jia, “Consensus of second-order discrete-time multi-
for network consensus,” Automatica, vol. 44, no. 9, pp. 2375–2381, agent systems with nonuniform time-delays and dynamically changing
2008. topologies,” Automatica, vol. 45, no. 9, pp. 2154–2158, 2009.
[88] J. Cortés, “Finite-time convergent gradient flows with applications to [114] B. Yang and H. Fang, “Forced consensus in networks of double
network consensus,” Automatica, vol. 42, no. 11, pp. 1993–2000, 2006. integrator systems with delayed input,” Automatica, vol. 46, no. 3, pp.
[89] W. Yu, G. Chen, and J. Lü, “On pinning synchronization of complex 629–632, 2010.
dynamical networks,” Automatica, vol. 45, no. 2, pp. 429–435, 2009. [115] Y. Zhang and Y.-P. Tian, “Consensus of data-sampled multi-agent
[90] D. V. Dimarogonas and K. J. Kyriakopoulos, “On the rendezvous systems with random communication delay and packet loss,” IEEE
problem for multiple nonholonomic agents,” IEEE Transactions on Transactions on Automatic Control, vol. 55, no. 4, pp. 939–943, 2010.
Automatic Control, vol. 52, no. 5, pp. 916–922, 2007. [116] P. Lin and Y. Jia, “Consensus of a class of second-order multi-
[91] H. Bai, M. Arcak, and J. T. Wen, “Adaptive motion coordination: Using agent systems with time-delay and jointly-connected topologies,” IEEE
relative velocity feedback to track a reference velocity,” Automatica, Transactions on Automatic Control, vol. 55, no. 3, pp. 778–784, 2010.
vol. 45, no. 4, pp. 1020–1025, 2009. [117] F. Xiao, L. Wang, and J. Chen, “Partial state consensus for networks
[92] X. Wang and G. Chen, “Synchronization in scale-free dynamical of second-order dynamic agents,” Systems and Control Letters, vol. 59,
networks: robustness and fragility,” IEEE Transactions on Circuits And no. 12, pp. 775–781, 2010.
Systems–I: Regular Papers, vol. 49, no. 1, pp. 54–62, 2002. [118] U. Münz, A. Papachristodoulou, and F. Allgower, “Robust consensus
[93] ——, “Synchronization in small-world dynamical networks,” Interna- controller design for nonlinear relative degree two multi-agent systems
tional Journal of Bifurcation and Chaos, vol. 12, no. 1, pp. 187–192, with communication constraints,” IEEE Transactions on Automatic
2002. Control, vol. 56, no. 1, pp. 145–151, 2011.
[94] W. Yu, J. Cao, and J. Lü, “Global synchronization of linearly hybrid [119] J. Qin, H. Gao, and W. X. Zheng, “Second-order consensus for multi-
coupled networks with time-varying delay,” SIAM Journal on Applied agent systems with switching topology and communication delay,”
Dynamical Systems, vol. 7, no. 1, pp. 108–133, 2008. Systems and Control Letters, vol. 60, no. 6, pp. 390–397, 2011.
 17

[120] J. Cao, G. Chen, and P. Li, “Global synchronization in an array of [142] R. Carli and F. Bullo, “Quantized coordination algorithms for ren-
delayed neural networks with hybrid coupling,” IEEE Transactions on dezvous and deployment,” SIAM Journal on Control and Optimization,
Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 38, no. 2, vol. 48, no. 3, pp. 1251–1274, 2009.
pp. 488–498, April 2008. [143] J. Lavaei and R. M. Murray, “On quantized consensus by means of
[121] J. Liang, Z. Wang, Y. Liu, and X. Liu, “Global synchronization control gossip algorithm Part I: Convergence proof,” in Proceedings of the
of general delayed discrete-time networks with stochastic coupling and American Control Conference, 2009, pp. 394–401.
disturbances,” IEEE Transactions on Systems, Man, and Cybernetics, [144] ——, “On quantized consensus by means of gossip algorithm–Part II:
Part B: Cybernetics, vol. 38, no. 4, pp. 1073–1083, August 2008. Convergence time,” in Proceedings of the American Control Confer-
[122] J. Yao, H. O. Wang, Z.-H. Guan, and W. Xu, “Passive stability and ence, 2009, pp. 2958–2965.
synchronization of complex spatio-temporal switching networks with [145] D. Yuan, S. Xu, H. Zhao, and Y. Chu, “Distributed average consensus
time delays,” Automatica, vol. 45, no. 7, pp. 1721–1728, 2009. via gossip algorithm with real-valued and quantized data for 0 < q <
[123] Y. Wang, H. Zhang, X. Wang, and D. Yang, “Networked synchroniza- 1,” Systems and Control Letters, vol. 59, no. 9, pp. 536–542, 2010.
tion control of coupled dynamic networks with time-varying delay,” [146] A. Nedic, A. Olshevsky, A. Ozdaglar, and J. N. Tsitsiklis, “On
IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cyber- distributed averaging algorithms and quantization effects,” IEEE Trans-
netics, vol. 40, no. 6, pp. 1468–1479, December 2010. actions on Automatic Control, vol. 54, no. 11, pp. 2506–2517, 2009.
[124] N. Chopra, M. W. Spong, and R. Lozano, “Synchronization of bilateral [147] M. Franceschelli, A. Giua, and C. Seatzu, “A gossip-based algorithm
teleoperators with time delay,” Automatica, vol. 44, no. 8, pp. 2142– for discrete consensus over heterogeneous networks,” IEEE Transac-
2148, 2008. tions on Automatic Control, vol. 55, no. 5, pp. 1244–1249, 2010.
[148] T. Li, M. Fu, L. Xie, and J.-F. Zhang, “Distributed consensus with
[125] E. Nuno, R. Ortega, L. Basanez, and D. Hill, “Synchronization of net-
limited communication data rate,” IEEE Transactions on Automatic
works of nonidentical euler-lagrange systems with uncertain parameters
Control, vol. 56, no. 2, pp. 279–292, 2011.
and communication delays,” IEEE Transactions on Automatic Control,
[149] M. Zhu and S. Martı́nez, “On the convergence time of asynchronous
vol. 56, no. 4, pp. 935–941, 2011.
distributed quantized averaging algorithms,” IEEE Transactions on
[126] V. S. Bokharaie, O. Mason, and M. Verwoerd, “D-stability and delay- Automatic Control, vol. 56, no. 2, pp. 386–390, 2011.
independent stability of homogeneous cooperative systems,” IEEE [150] R. Carli, G. Como, P. Frasca, and F. Garin, “Distributed averaging
Transactions on Automatic Control, vol. 55, no. 12, pp. 2882–2885, on digital erasure networks,” Automatica, vol. 47, no. 1, pp. 115–121,
2010. 2011.
[127] T. Hayakawa, T. Matsuzawa, and S. Hara, “Formation control of multi- [151] D. V. Dimarogonas and K. H. Johansson, “Stability analysis for multi-
agent systems with sampled information,” in Proceedings of the IEEE agent systems using the incidence matrix: Quantized communication
Conference on Decision and Control, San Diego, CA, December 2006, and formation control,” Automatica, vol. 46, no. 4, pp. 695–700, 2010.
pp. 4333–4338. [152] R. Carli, F. Fagnani, P. Frasca, and S. Zampieri, “Gossip consensus
[128] G. Xie, H. Liu, L. Wang, and Y. Jia, “Consensus in networked algorithms via quantized communication,” Automatica, vol. 46, no. 1,
multi-agent systems via sampled control: Fixed topology case,” in pp. 70–80, 2010.
Proceedings of the American Control Conference, 2009, pp. 3902– [153] Y. Kim and M. Mesbahi, “On maximizing the second smallest eigen-
3907. value of a state-dependent graph laplacian,” IEEE Transactions on
[129] ——, “Consensus in networked multi-agent systems via sampled Automatic Control, vol. 51, no. 1, pp. 116–120, 2006.
control: Switching topology case,” in Proceedings of the American [154] L. Xiao and S. Boyd, “Fast linear iterations for distributed averaging,”
Control Conference, 2009, pp. 4525–4530. Systems and Control Letters, vol. 53, no. 1, pp. 65–78, 2004.
[130] Y. Gao, L. Wang, G. Xie, and B. Wu, “Consensus of multi-agent [155] A. Olshevsky and J. N. Tsitsiklis, “Convergence speed in distributed
systems based on sampled-data control,” International Journal of consensus and averaging,” SIAM Journal on Control and Optimization,
Control, vol. 82, no. 12, pp. 2193–2205, 2009. vol. 48, no. 1, pp. 33–55, 2009.
[131] Y. Gao and L. Wang, “Consensus of multiple double-integrator agents [156] D. Angeli and P.-A. Bliman, “Convergence speed of unsteady dis-
with intermittent measurement,” International Journal of Robust and tributed consensus: Decay estimate along the settling spanning-trees,”
Nonlinear Control, vol. 20, no. 10, pp. 1140–1155, 2010. SIAM Journal on Control and Optimization, vol. 48, no. 1, pp. 1–32,
[132] Y. Cao and W. Ren, “Multi-vehicle coordination for double-integrator 2009.
dynamics under fixed undirected/directed interaction in a sampled- [157] ——, “Tight estimates for convergence of some non-stationary con-
data setting,” International Journal of Robust and Nonlinear Control, sensus algorithms,” Systems and Control Letters, vol. 57, no. 12, pp.
vol. 20, no. 9, pp. 987–1000, 2010. 996–1004, 2008.
[133] ——, “Sampled-data discrete-time consensus algorithms for double- [158] J. Zhou and Q. Wang, “Convergence speed in distributed consensus
integrator dynamics under dynamic directed interaction,” International over dynamically switching random networks,” Automatica, vol. 45,
Journal of Control, vol. 83, no. 3, pp. 506–515, 2010. no. 6, pp. 1455–1461, 2009.
[134] W. Yu, W. X. Zheng, G. Chen, W. Ren, and J. Cao, “Second-order [159] F. Xiao, L. Wang, J. Chen, and Y. Gao, “Finite-time formation control
consensus in multi-agent dynamical systems with sampled position for multi-agent systems,” Automatica, vol. 45, no. 11, pp. 2605–2611,
data,” Automatica, vol. 47, no. 7, pp. 1496–1503, 2011. 2009.
[135] Y. Gao and L. Wang, “Sampled-data based consensus of continuous- [160] F. Jiang and L. Wang, “Finite-time information consensus for multi-
time multi-agent systems with time-varying topology,” IEEE Transac- agent systems with fixed and switching topologies,” Physica D, vol.
tions on Automatic Control, vol. 56, no. 5, pp. 1226–1231, 2011. 238, no. 16, pp. 1550–1560, 2009.
[161] Y. Cao, W. Ren, and Z. Meng, “Decentralized finite-time sliding mode
[136] H. Liu, G. Xie, and L. Wang, “Necessary and sufficient conditions for
estimators and their applications in decentralized finite-time formation
solving consensus problems of double-integrator dynamics via sampled
tracking,” Systems and Control Letters, vol. 59, no. 9, pp. 522–529,
control,” International Journal of Robust and Nonlinear Control,
2010.
vol. 20, no. 15, pp. 1706–1722, 2010.
[162] L. Wang and F. Xiao, “Finite-time consensus problems for networks
[137] G. Wen, Z. Duan, W. Yu, and G. Chen, “Consensus of multi-agent of dynamic agents,” IEEE Transactions on Automatic Control, vol. 55,
systems with intrinsic nonlinear dynamics and sampled-data informa- no. 4, pp. 950–955, 2010.
tion,” in Asian Control Conference, Kaohsiung, Taiwan, May 2011, pp. [163] Q. Hui, “Finite-time rendezvous algorithms for mobile autonomous
107–112. agents,” IEEE Transactions on Automatic Control, vol. 56, no. 1, pp.
[138] M. Cao, A. S. Morse, and B. D. O. Anderson, “Agreeing asyn- 207–211, 2011.
chronously,” IEEE Transactions on Automatic Control, vol. 53, no. 8, [164] X. Wang and Y. Hong, “Finite-time consensus for multi-agent networks
pp. 1826–1838, 2008. with second-order agent dynamics,” in IFAC World Congress, 2008, pp.
[139] L. Fang and P. J. Antsaklis, “Asynchronous consensus protocols using 15 185–15 190.
nonlinear paracontractions theory,” IEEE Transactions on Automatic [165] S. Sundaram and C. N. Hadjicostis, “Finite-time distributed consensus
Control, vol. 53, no. 10, pp. 2351–2355, 2008. in graphs with time-invariant topologies,” in Proceedings of the Amer-
[140] Y. Gao and L. Wang, “Asynchronous consensus of continuous-time ican Control Conference, New York City, NY, July 2007, pp. 711–716.
multi-agent systems with intermittent measurements,” International [166] Y. Yuan, G.-B. Stan, M. Barahona, L. Shi, and J. Gonçalves, “De-
Journal of Control, vol. 83, no. 3, pp. 552–562, 2010. centralised minimal-time consensus,” in Proceedings of the IEEE
[141] A. Kashyap, T. Başar, and R. Srikant, “Quantized consensus,” Auto- Conference on Decision and Control, Orlando, FL, December 2011,
matica, vol. 43, no. 7, pp. 1192–1203, 2007. pp. 4282–4289.
 18

[167] C. W. Reynolds, “Flocks, herds, and schools: A distributed behavioral [192] S. Hernandez and D. A. Paley, “Three-dimensional motion coordina-
model,” Computer Graphics, vol. 21, no. 4, pp. 25–34, July 1987. tion in a spatiotemporal flowfield,” IEEE Transactions on Automatic
[168] N. E. Leonard and E. Fiorelli, “Virtual leaders, artificial potentials and Control, vol. 55, no. 12, pp. 2805–2810, 2010.
coordinated control of groups,” in Proceedings of the IEEE Conference [193] D. A. Paley, N. E. Leonard, and R. Sepulchre, “Stabilization of
on Decision and Control, Orlando, Florida, December 2001, pp. 2968– symmetric formations to motion around convex loops,” Systems and
2973. Control Letters, vol. 57, no. 3, pp. 209–215, 2008.
[169] P. Ogren, E. Fiorelli, and N. E. Leonard, “Cooperative control of [194] G. Laman, “On graphs and rigidity of plane skeletal structures,” Journal
mobile sensor networks: Adaptive gradient climbing in a distributed of Engineering Mathematics, vol. 4, no. 4, pp. 331–340, 1970.
environment,” IEEE Transactions on Automatic Control, vol. 49, no. 8, [195] R. Olfati-Saber and R. M. Murray, “Graph rigidity and distributed
pp. 1292–1302, August 2004. formation stabilization of multi-vehicle systems,” in Proceedings of
[170] M. Pavone and E. Frazzoli, “Decentralized policies for geometric the IEEE Conference on Decision and Control, 2002, pp. 2965–2971.
pattern formation and path coverage,” ASME Journal of Dynamic [196] J. M. Hendrickx, B. D. O. Anderson, J.-C. Delvenne, and V. D.
Systems, Measurement, and Control, vol. 129, no. 5, pp. 633–643, 2007. Blondel, “Directed graphs for the analysis of rigidity and persistence
[171] W. Ren, “Collective motion from consensus with cartesian coordinate in autonomous agent systems,” International Journal of Robust and
coupling,” IEEE Transactions on Automatic Control, vol. 54, no. 6, pp. Nonlinear Control, vol. 17, no. 11, pp. 960–981, July 2007.
1330–1336, 2009. [197] C. Yu, B. D. O. Anderson, S. Dasgupta, and B. Fidan, “Control of
[172] P. Lin and Y. Jia, “Distributed rotating formation control of multi-agent minimally persistent formations in the plane,” SIAM Journal on Control
systems,” Systems and Control Letters, vol. 59, no. 10, pp. 587–595, and Optimization, vol. 48, no. 1, pp. 206–233, 2009.
2010. [198] B. Fidan, J. M. Hendrickx, and B. D. O. Anderson, “Closing ranks
[173] P. Lin, K. Qin, Z. Li, and W. Ren, “Collective rotating motions of in rigid multi-agent formations using edge contraction,” International
second-order multi-agent systems in three-dimensional space,” Systems Journal of Robust and Nonlinear Control, vol. 20, no. 18, pp. 2077–
and Control Letters, vol. 60, no. 6, pp. 365–372, 2011. 2092, 2010.
[174] L. M. y Teran-Romero, E. Forgoston, and I. B. Schwartz, “Coherent [199] M. Cao, C. Yu, and B. D. Anderson, “Formation control using range-
pattern prediction in swarms of delay-coupled agents,” IEEE Transac- only measurements,” Automatica, vol. 47, no. 4, pp. 776–781, 2011.
tions on Robotics, 2012, in press. [200] W. B. Dunbar and R. M. Murray, “Distributed receding horizon control
[175] R. Sepulchre, D. A. Paley, and N. E. Leonard, “Stabilization of planar for multi-vehicle formation stabilization,” Automatica, vol. 42, no. 4,
collective motion: All-to-all communication,” IEEE Transactions on pp. 549–558, 2006.
Automatic Control, vol. 52, no. 5, pp. 811–824, 2007. [201] W. B. Dunbar, “Distributed receding horizon control of dynamically
[176] ——, “Stabilization of planar collective motion with limited commu- coupled nonlinear systems,” IEEE Transactions on Automatic Control,
nication,” IEEE Transactions on Automatic Control, vol. 53, no. 3, pp. vol. 52, no. 7, pp. 1249–1263, 2007.
706–719, 2008. [202] E. Franco, L. Magni, T. Parisini, M. M. Polycarpou, and D. M.
[177] D. V. Dimarogonas and K. J. Kyriakopoulos, “Inverse agreement Raimondo, “Cooperative constrained control of distributed agents with
protocols with application to distributed multi-agent dispersion,” IEEE nonlinear dynamics and delayed information exchange: A stabilizing
Transactions on Automatic Control, vol. 54, no. 3, pp. 657–663, 2009. receding-horizon approach,” IEEE Transactions on Automatic Control,
vol. 53, no. 1, pp. 324–338, 2008.
[178] R. Olfati-Saber, “Flocking for multi-agent dynamic systems: Algo-
[203] W. Ren, “Multi-vehicle consensus with a time-varying reference state,”
rithms and theory,” IEEE Transactions on Automatic Control, vol. 51,
Systems and Control Letters, vol. 56, no. 7, pp. 474–483, 2007.
no. 3, pp. 401–420, 2006.
[204] Y. Cao, W. Ren, and Y. Li, “Distributed discrete-time coordinated track-
[179] K. D. Do, “Bounded controllers for formation stabilization of mobile
ing with a time-varying reference state and limited communication,”
agents with limited sensing ranges,” IEEE Transactions on Automatic
Automatica, vol. 45, no. 5, pp. 1299–1305, 2009.
Control, vol. 52, no. 3, pp. 569–576, 2007.
[205] J. Xiang, W. Wei, and Y. Li, “Synchronized output regulation of linear
[180] F. Cucker and S. Smale, “Emergent behavior in flocks,” IEEE Trans-
networked systems,” IEEE Transactions on Automatic Control, vol. 54,
actions on Automatic Control, vol. 52, no. 5, pp. 852–862, 2007.
no. 6, pp. 1336–1341, 2009.
[181] H. G. Tanner, A. Jadbabaie, and G. J. Pappas, “Flocking in fixed and [206] X. Wang, Y. Hong, J. Huang, and Z.-P. Jiang, “A distributed control
switching networks,” IEEE Transactions on Automatic Control, vol. 52, approach to a robust output regulation problem for multi-agent linear
no. 5, pp. 863–868, 2007. systems,” IEEE Transactions on Automatic Control, vol. 55, no. 12,
[182] L. E. Barnes, M. A. Fields, and K. P. Valavanis, “Swarm formation pp. 2891–2895, 2010.
control utilizing elliptical surfaces and limiting functions,” IEEE Trans- [207] M. Porfiri, D. G. Roberson, and D. J. Stilwell, “Tracking and for-
actions on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 39, mation control of multiple autonomous agents:a two-level consensus
no. 6, pp. 1434–1445, December 2009. approach,” Automatica, vol. 43, no. 8, pp. 1318–1328, 2007.
[183] M. Kumar, D. P. Garg, and V. Kumar, “Segregation of heterogeneous [208] H. Su, X. Wang, and Z. Lin, “Flocking of multi-agents with a virtual
units in a swarm of robotic agents,” IEEE Transactions on Automatic leader,” IEEE Transactions on Automatic Control, vol. 54, no. 2, pp.
Control, vol. 55, no. 3, pp. 743–748, 2010. 293–307, 2009.
[184] J. Park, H. J. Kim, and S.-Y. Ha, “Cucker-smale flocking with inter- [209] H. Shi, L. Wang, and T. Chu, “Flocking of multi-agent systems with
particle bonding forces,” IEEE Transactions on Automatic Control, a dynamic virtual leader,” International Journal of Control, vol. 82,
vol. 55, no. 11, pp. 2617–2623, 2010. no. 1, pp. 43–58, January 2009.
[185] S.-Y. Ha, T. Ha, and J.-H. Kim, “Emergent behavior of a cucker- [210] W. Yu, G. Chen, and M. Cao, “Distributed leader-follower flocking
smale type particle model with nonlinear velocity couplings,” IEEE control for multi-agent dynamical systems with time-varying veloci-
Transactions on Automatic Control, vol. 55, no. 7, pp. 1679–1683, ties,” Systems and Control Letters, vol. 59, no. 9, pp. 543–552, 2010.
2010. [211] Y. Cao and W. Ren, “Distributed coordinated tracking with reduced
[186] F. Cucker and J.-G. Dong, “Avoiding collisions in flocks,” IEEE interaction via a variable structure approach,” IEEE Transactions on
Transactions on Automatic Control, vol. 55, no. 5, pp. 1238–1243, Automatic Control, vol. 57, no. 1, pp. 33–48, 2012.
2010. [212] G. Wen, Z. Duan, Z. Li, and G. Chen, “Flocking of multi-agent
[187] ——, “A general collision-avoiding flocking framework,” IEEE Trans- dynamical systems with intermittent nonlinear velocity measurements,”
actions on Automatic Control, vol. 56, no. 5, pp. 1124–1129, 2011. International Journal of Robust and Nonlinear Control, 2011.
[188] H.-T. Zhang, C. Zhai, and Z. Chen, “A general alignment repulsion [213] W. Ni and D. Cheng, “Leader-following consensus of multi-agent
algorithm for flocking of multi-agent systems,” IEEE Transactions on systems under fixed and switching topologies,” Systems and Control
Automatic Control, vol. 56, no. 2, pp. 430–435, 2011. Letters, vol. 59, no. 3-4, pp. 209–217, 2010.
[189] F. Zhang and N. E. Leonard, “Coordinated patterns of unit speed [214] W. Ren, “Consensus tracking under directed interaction topologies:
particles on a closed curve,” Systems and Control Letters, vol. 56, Algorithms and experiments,” IEEE Transactions on Control Systems
no. 6, pp. 397–407, 2007. Technology, vol. 18, no. 1, pp. 230–237, January 2010.
[190] D. A. Paley, “Stabilization of collective motion on a sphere,” Automat- [215] J. Hu and G. Feng, “Distributed tracking control of leader-follower
ica, vol. 45, no. 1, pp. 212–216, 2009. multi–agent systems under noisy measurement,” Automatica, vol. 46,
[191] R. Mellish, S. Napora, and D. A. Paley, “Backstepping control design no. 8, pp. 1382–1387, 2010.
for motion coordination of self-propelled vehicles in a flowfield,” [216] W. Wang and J.-J. Slotine, “A theoretical study of different leader roles
International Journal of Robust and Nonlinear Control, vol. 21, no. 12, in networks,” IEEE Transactions on Automatic Control, vol. 51, no. 7,
pp. 1452–1466, 2011. pp. 1156–1161, 2006.
 19

[217] Q. Song, J. Cao, and W. Yu, “Second-order leader-following consensus [241] A. Nedic, A. Ozdaglar, and P. A. Parrilo, “Constrained consensus
of nonlinear multi-agent systems via pinning control,” Systems and and optimization in multi-agent networks,” IEEE Transactions on
Control Letters, vol. 59, no. 9, pp. 553–562, 2010. Automatic Control, vol. 55, no. 4, pp. 922–938, 2010.
[218] K. D. Do, “Formation tracking control of unicycle-type mobile robots [242] I. Lobel and A. Ozdaglar, “Distributed subgradient methods for convex
with limited sensing ranges,” IEEE Transactions on Control Systems optimization over random networks,” IEEE Transactions on Automatic
Technology, vol. 16, no. 3, pp. 527–538, May 2008. Control, vol. 56, no. 6, pp. 1291–1306, 2011.
[219] T. Gustavi and X. Hu, “Observer-based leader-following formation [243] A. Nedic, “Asynchronous broadcast-based convex optimization over a
control using onboard sensor information,” IEEE Transactions on network,” IEEE Transactions on Automatic Control, vol. 56, no. 6, pp.
Robotics, vol. 24, no. 6, pp. 1457–1462, 2008. 1337–1351, 2011.
[220] W. Dong and J. A. Farrell, “Cooperative control of multiple non- [244] Y. Cao and W. Ren, “Optimal linear consensus algorithms: An LQR
holonomic mobile agents,” IEEE Transactions on Automatic Control, perspective,” IEEE Transactions on Systems, Man, and Cybernetics,
vol. 53, no. 6, pp. 1434–1448, 2008. Part B: Cybernetics, vol. 40, no. 3, pp. 819–830, 2010.
[221] ——, “Decentralized cooperative control of multiple nonholonomic [245] F. Borrelli and T. Keviczky, “Distributed LQR design for identical
dynamic systems with uncertainty,” Automatica, vol. 45, no. 3, pp. dynamically decoupled systems,” IEEE Transactions on Automatic
706–710, 2009. Control, vol. 53, no. 8, pp. 1901–1912, 2008.
[222] W. Dong, “Flocking of multiple mobile robots based on backstepping,” [246] E. Semsar-Kazerooni and K. Khorasani, “Optimal consensus seeking
IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cyber- in a network of multiagent systems: An LMI approach,” IEEE Transac-
netics, vol. 41, no. 2, pp. 414–514, April 2011. tions on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 40,
[223] E. J. Rodrı́guez-Seda, J. J. Troy, C. A. Erignac, P. Murray, D. M. no. 2, pp. 540–547, April 2010.
Stipanović, and M. W. Spong, “Bilateral teleoperation of multiple [247] J. Wang and M. Xin, “Multi-agent consensus algorithm with obstacle
mobile agents: Coordinated motion and collision avoidance,” IEEE avoidance via optimal control approach,” International Journal of
Transactions on Control Systems Technology, vol. 18, no. 4, pp. 984– Control, vol. 83, no. 12, pp. 2606–2621, 2010.
992, July 2010. [248] W. Dong, “Distributed optimal control of multiple systems,” Interna-
[224] W. Ren, “Distributed cooperative attitude synchronization and tracking tional Journal of Control, vol. 83, no. 10, pp. 2067–2079, 2010.
for multiple rigid bodies,” IEEE Transactions on Control Systems [249] J. Hu, M. Prandini, and C. Tomlin, “Conjugate points in formation
Technology, vol. 18, no. 2, pp. 383–392, March 2010. constrained optimal multi-agent coordination: A case study,” SIAM
[225] H. Su, X. Wang, and G. Chen, “Rendezvous of multiple mobile agents Journal on Control and Optimization, vol. 45, no. 6, pp. 2119–2137,
with preserved network connectivity,” Systems and Control Letters, 2006.
vol. 59, no. 5, pp. 313–322, 2010. [250] E. Franco, T. Parisini, and M. M. Polycarpou, “Design and stability
[226] M. M. Zavlanos, H. G. Tanner, A. Jadbabaie, and G. J. Pappas, “Hybrid analysis of cooperative receding-horizon control of linear discrete-
control for connectivity preserving flocking,” IEEE Transactions on time agents,” International Journal of Robust and Nonlinear Control,
Automatic Control, vol. 54, no. 12, pp. 2869–2875, 2009. vol. 17, no. 11, pp. 982–1001, July 2007.
[227] A. Ajorlou, A. Momeni, and A. G. Aghdam, “A class of bounded [251] G. Ferrari-Trecate, L. Galbusera, M. P. E. Marciandi, and R. Scattolini,
distributed control strategies for connectivity preservation in multi- “Model predictive control schemes for consensus in multi-agent sys-
agent systems,” IEEE Transactions on Automatic Control, vol. 55, tems with single- and double-integrator dynamics,” IEEE Transactions
no. 12, pp. 2828–2833, 2010. on Automatic Control, vol. 54, no. 11, pp. 2560–2572, 2009.
[252] B. T. Stewart, A. N. Venkat, J. B. Rawlings, S. J. Wright, and
[228] T. Gustavi, D. V. Dimarogonas, M. Egerstedt, and X. Hu, “Sufficient
G. Pannocchia, “Cooperative distributed model predictive control,”
conditions for connectivity maintenance and rendezvous in leader-
Systems and Control Letters, vol. 59, no. 8, pp. 460–469, 2010.
follower networks,” Automatica, vol. 46, no. 1, pp. 133–139, 2010.
[253] J. Cortés, S. Martı́nez, T. Karatas, and F. Bullo, “Coverage control
[229] H. Su, X. Wang, and Z. Lin, “Synchronization of coupled harmonic
for mobile sensing networks,” IEEE Transactions on Robotics and
oscillators in a dynamic proximity network,” Automatica, vol. 45,
Automation, vol. 20, pp. 243–255, 2004.
no. 10, pp. 2286–2291, 2009.
[254] J. Cortés, S. Martı́nez, and F. Bullo, “Spatially distributed coverage
[230] H. Su, X. Wang, and G. Chen, “A connectivity-preserving flocking al- optimization and control with limited-range interactions,” ESAIM:
gorithm for multi-agent systems based only on position measurements,” Control, Optimisation and Calculus of Variations, vol. 11, no. 4, pp.
International Journal of Control, vol. 82, no. 7, pp. 1334–1343, July 691–719, 2005.
2009. [255] C. Gao, J. Cortés, and F. Bullo, “Notes on averaging over acyclic
[231] M. Ji and M. Egerstedt, “Distributed coordination control of multia- digraphs and discrete coverage control,” Automatica, vol. 44, no. 8,
gent systems while preserving connectedness,” IEEE Transactions on pp. 2120–2127, 2008.
Robotics, vol. 23, no. 4, pp. 693–703, 2007. [256] K. Laventalla and J. Cortés, “Coverage control by multi-robot networks
[232] M. M. Zavlanos and G. J. Pappas, “Distributed connectivity control of with limited-range anisotropic sensory,” International Journal of Con-
mobile networks,” IEEE Transactions on Robotics, vol. 24, no. 6, pp. trol, vol. 82, no. 6, pp. 1113–1121, June 2009.
1416–1428, 2008. [257] A. Kwok and S. Martı́nez, “A distributed deterministic annealing
[233] B. Shucker, T. D. Murphey, and J. K. Bennett, “Convergence-preserving algorithm for limited-range sensor coverage,” IEEE Transactions on
switching for topology-dependent decentralized systems,” IEEE Trans- Control Systems Technology, vol. 19, no. 4, pp. 792–804, July 2011.
actions on Robotics, vol. 24, no. 6, pp. 1405–1415, 2008. [258] Y. Wang and I. I. Hussein, “Awareness coverage control over large-
[234] M. M. Zavlanos and G. J. Pappas, “Potential fields for maintaining scale domains with intermittent communications,” IEEE Transactions
connectivity of mobile networks,” IEEE Transactions on Robotics, on Automatic Control, vol. 55, no. 8, pp. 1850–1859, 2010.
vol. 23, no. 4, pp. 812–816, 2007. [259] J. Choi and R. Horowitz, “Learning coverage control of mobile sensing
[235] Z. Sun and J. Huang, “A note on connectivity of multi-agent systems agents in one-dimensional stochastic environments,” IEEE Transactions
with proximity graphs and linear feedback protocol,” Automatica, on Automatic Control, vol. 55, no. 3, pp. 804–809, 2010.
vol. 45, no. 8, pp. 1953–1956, 2009. [260] J. Cortés, “Coverage optimization and spatial load balancing by robotic
[236] Y. Kim, “Bisection algorithm of increasing algebraic connectivity by sensor networks,” IEEE Transactions on Automatic Control, vol. 55,
adding an edge,” IEEE Transactions on Automatic Control, vol. 55, no. 3, pp. 749–754, 2010.
no. 1, pp. 170–174, 2010. [261] A. Kwok and S. Martı́nez, “Unicycle coverage control via hybrid
[237] R. Carli, A. Chiuso, L. Schenato, and S. Zampieri, “Optimal synchro- modeling,” IEEE Transactions on Automatic Control, vol. 55, no. 2,
nization for networks of noisy double integrators,” IEEE Transactions pp. 528–532, 2010.
on Automatic Control, vol. 56, no. 5, pp. 1146–1152, 2011. [262] Z. Jin, T. Shima, and C. J. Schumacher, “Optimal scheduling for
[238] Y. Kim, D.-W. Gu, and I. Postlethwaite, “Spectral radius minimization refueling multiple autonomous aerial vehicles,” IEEE Transactions on
for optimal average consensus and output feedback stabilization,” Robotics, vol. 22, no. 4, pp. 682–693, 2006.
Automatica, vol. 45, no. 6, pp. 1379–1386, 2009. [263] J. Finke, K. M. Passino, and A. G. Sparks, “Stable task load balancing
[239] M. Rabbat and R. Nowak, “Distributed optimization in sensor net- strategies for cooperative control of networked autonomous air vehi-
works,” in ACM Information Processing in Sensor Networks, Berkeley, cles,” IEEE Transactions on Control Systems Technology, vol. 14, no. 5,
CA, April 2004, pp. 20–27. pp. 789–803, September 2006.
[240] A. Nedic and A. Ozdaglar, “Distributed subgradient methods for multi- [264] K. L. Moore and D. Lucarelli, “Decentralized adaptive scheduling using
agent optimization,” IEEE Transactions on Automatic Control, vol. 54, consensus variables,” International Journal of Robust and Nonlinear
no. 1, pp. 48–61, 2009. Control, vol. 17, no. 11, pp. 921–940, July 2007.
 20

[265] H.-L. Choi, L. Brunet, and J. P. How, “Consensus-based decentralized [288] M. V. Subbotin and R. S. Smith, “Design of distributed decentralized
auctions for robust task allocation,” IEEE Transactions on Robotics, estimators for formations with fixed and stochastic communication
vol. 25, no. 4, pp. 912–926, 2009. topologies,” Automatica, vol. 45, no. 11, pp. 2491–2501, 2009.
[266] S. A. Reveliotis and E. Roszkowska, “Conflict resolution in free- [289] H. Hirai and F. Miyazaki, “Dynamic coordination between robots: Self-
ranging multivehicle systems: A resource allocation paradigm,” IEEE organized timing selection in a juggling-like ball-passing task,” IEEE
Transactions on Robotics, vol. 27, no. 2, pp. 283–296, 2011. Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics,
[267] H. Sayyaadi and M. Moarref, “A distributed algorithm for proportional vol. 36, no. 4, pp. 738–754, August 2006.
task allocation in networks of mobile agents,” IEEE Transactions on [290] F. Kunwar and B. Benhabib, “Rendezvous-guidance trajectory plan-
Automatic Control, vol. 56, no. 2, pp. 405–410, 2011. ning for robotic dynamic obstacle avoidance and interception,” IEEE
[268] D. A. Anisi, P. Ögren, and X. Hu, “Cooperative minimum time Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics,
surveillance with multiple ground vehicles,” IEEE Transactions on vol. 36, no. 6, pp. 1432–1441, December 2006.
Automatic Control, vol. 55, no. 12, pp. 2679–2691, 2010. [291] G. Antonelli and S. Chiaverini, “Kinematic control of platoons of
[269] D. Kingston, R. W. Beard, and R. S. Holt, “Decentralized perimeter autonomous vehicles,” IEEE Transactions on Robotics, vol. 22, no. 6,
surveillance using a team of UAVs,” IEEE Transactions on Robotics, pp. 1285–1292, 2006.
vol. 24, no. 6, pp. 1394–1404, 2008. [292] G. Baldassarre, V. Trianni, M. Bonani, F. Mondada, M. Dorigo, and
S. Nolfi, “Self-organized coordinated motion in groups of physically
[270] B. J. Moore and K. M. Passino, “Decentralized redistribution for
connected robots,” IEEE Transactions on Systems, Man, and Cybernet-
cooperative patrol,” International Journal of Robust and Nonlinear
ics, Part B: Cybernetics, vol. 37, no. 1, pp. 224–239, February 2007.
Control, vol. 18, no. 2, pp. 165–195, January 2008.
[293] W. Ren, H. Chao, W. Bourgeous, N. Sorensen, and Y. Chen, “Experi-
[271] A. E. Gil, K. M. Passino, and J. B. Cruz, “Stable cooperative surveil-
mental validation of consensus algorithms for multivehicle cooperative
lance with information flow constraints,” IEEE Transactions on Control
control,” IEEE Transactions on Control Systems Technology, vol. 16,
Systems Technology, vol. 16, no. 5, pp. 856–868, September 2008.
no. 4, pp. 745–752, July 2008.
[272] M. Peasgood, C. M. Clark, and J. McPhee, “A complete and scalable [294] R. O’Grady, A. L. Christensen, and M. Dorigo, “Swarmorph: Multi-
strategy for coordinating multiple robots within roadmaps,” IEEE robot morphogenesis using directional self-assembly,” IEEE Transac-
Transactions on Robotics, vol. 24, no. 2, pp. 283–292, 2008. tions on Robotics, vol. 25, no. 3, pp. 738–743, 2009.
[273] Y. Yang, M. M. Polycarpou, and A. A. Minai, “Multi-UAV cooperative [295] Z. Cao, M. Tan, L. Li, N. Gu, and S. Wang, “Cooperative hunting by
search using an opportunistic learning method,” ASME Journal of distributed mobile robots based on local interaction,” IEEE Transac-
Dynamic Systems, Measurement, and Control, vol. 129, no. 5, pp. 716– tions on Robotics, vol. 22, no. 2, pp. 403–407, 2006.
728, 2007. [296] S. D. Bopardikar, F. Bullo, and J. P. Hespanha, “A cooperative
[274] J. Finke and K. M. Passino, “Stable cooperative vehicle distributions homicidal chauffeur game,” Automatica, vol. 45, no. 7, pp. 1771–1777,
for surveillance,” ASME Journal of Dynamic Systems, Measurement, 2009.
and Control, vol. 129, no. 5, pp. 597–608, 2007. [297] ——, “On discrete-time pursuit-evasion games with sensing limita-
[275] Z. Tang and Ümit Özgüner, “Cooperative sensor deployment for multi- tions,” IEEE Transactions on Robotics, vol. 24, no. 6, pp. 1429–1439,
target monitoring,” International Journal of Robust and Nonlinear 2008.
Control, vol. 18, no. 2, pp. 196–217, January 2008. [298] A. Kolling and S. Carpin, “Pursuit-evasion on trees by robot teams,”
[276] F. Zhang and N. E. Leonard, “Cooperative filters and control for IEEE Transactions on Robotics, vol. 26, no. 1, pp. 32–47, 2010.
cooperative exploration,” IEEE Transactions on Automatic Control, [299] X. Liang and Y. Xiao, “Studying bio-inspired coalition formation of
vol. 55, no. 3, pp. 650–663, 2010. robots for detecting intrusions using game theory,” IEEE Transactions
[277] P. Yang, R. Freeman, G. Gordon, K. Lynch, S. Srinivasa, and R. Suk- on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 40, no. 3,
thankar, “Decentralized estimation and control of graph connectivity pp. 683–693, June 2010.
for mobile sensor networks,” Automatica, vol. 46, no. 2, pp. 390–396, [300] D. Gu, “A differential game approach to formation control,” IEEE
2010. Transactions on Control Systems Technology, vol. 16, no. 1, pp. 85–93,
[278] J. Choi, S. Oh, and R. Horowitz, “Distributed learning and cooperative January 2008.
control for multi-agent systems,” Automatica, vol. 45, no. 12, pp. 2802– [301] D. Bauso, L. Giarre, and R. Pesenti, “Consensus for networks with
2814, 2009. unknown but bounded disturbances,” SIAM Journal on Control and
[279] K. M. Lynch, I. B. Schwartz, P. Yang, and R. A. Freeman, “Decen- Optimization, vol. 48, no. 3, pp. 1756–1770, 2009.
tralized environmental modeling by mobile sensor networks,” IEEE [302] ——, “Consensus in noncooperative dynamic games: A multiretailer
Transactions on Robotics, vol. 24, no. 3, pp. 710–724, 2008. inventory application,” IEEE Transactions on Automatic Control,
[280] P. Yang, R. A. Freeman, and K. M. Lynch, “Multi-agent coordination by vol. 53, no. 4, pp. 998–1003, 2008.
decentralized estimation and control,” IEEE Transactions on Automatic [303] P. Vrancx, K. Verbeeck, and A. Nowe, “Decentralized learning in
Control, vol. 53, no. 11, pp. 2480–2496, 2008. markov games,” IEEE Transactions on Systems, Man, and Cybernetics,
Part B: Cybernetics, vol. 38, no. 4, pp. 976–981, August 2008.
[281] R. S. Smith and F. Y. Hadaegh, “Closed-loop dynamics of cooperative
[304] M. Huang, P. E. Caines, and R. P. Malhame, “Large-population
vehicle formations with parallel estimators and communication,” IEEE
cost-coupled LQG problems with nonuniform agents: Individual-mass
Transactions on Automatic Control, vol. 52, no. 8, pp. 1404–1414,
behavior and decentralized ǫ-nash equilibria,” IEEE Transactions on
2007.
Automatic Control, vol. 52, no. 9, pp. 1560–1571, 2007.
[282] S. S. Stanković, M. S. Stanković, and D. M. Stipanović, “Consensus
based overlapping decentralized estimation with missing observations
and communication faults,” Automatica, vol. 45, no. 6, pp. 1397–1406,
2009.
[283] G. C. Calafiore and F. Abrate, “Distributed linear estimation over sensor
networks,” International Journal of Control, vol. 82, no. 5, pp. 868–
882, May 2009.
[284] M. A. Demetriou and I. I. Hussein, “Estimation of spatially distributed
processes using mobile spatially distributed sensor network,” SIAM
Journal on Control and Optimization, vol. 48, no. 1, pp. 266–291,
2009.
[285] N. Marechal, J.-M. Gorce, and J.-B. Pierrot, “Joint estimation and
gossip averaging for sensor network applications,” IEEE Transactions
on Automatic Control, vol. 55, no. 5, pp. 1208–1213, 2010.
[286] S. S. Stanković, M. S. Stanković, and D. M. Stipanović, “Decentralized
parameter estimation by consensus based stochastic approximation,”
IEEE Transactions on Automatic Control, vol. 56, no. 3, pp. 531–543,
2011.
[287] D. Yu, “Estimating the topology of complex dynamical networks by
steady state control: Generality and limitation,” Automatica, vol. 46,
no. 12, pp. 2035–2040, 2010.