Proceedings of the European Control Conference 2007 WeA08.

6
Kos, Greece, July 2-5, 2007

Distributed Consensus Control for Networks of Second-Order Agents

With Switching Topology and Time-Delay
Peng Lin, Yingmin Jia, Junping Du and Shiying Yuan

Abstract— In this paper, distributed consensus control is generate stable flocking motion. Wei Ren et al. [15] proposed
investigated for networks of agents with double integrator a second-order protocol and provided sufficient conditions
dynamics. Two network cases are considered, undirected net- for the case of fixed topology. They all did not consider
works with switching topology and undirected networks with
switching topology and time-delay. Two linear consensus pro- the communication time-delay. In practical application, the
tocols with and without time-delay are introduced to solve the disturbance of communication time-delay is unavoidable and
consensus problem, which includes two aspects, the agreement it might cause the network system to diverge or oscillate, so
of the position states and the synchronization of the speed it is significant to investigate the effects of time-delay, but,
states. Moreover, the convergence analysis is proved for both to our knowledge, there has been little work focused on this
cases and in each case sufficient conditions are given. Finally,
numerical simulation examples are included to illustrate the problem for agents with double integrator dynamics.
obtained theoretical results. In this paper, we study the distributed consensus control
in networks of agents with double integrator dynamics for
I. I NTRODUCTION two cases, undirected networks with switching topology, and
Distributed coordination of multiple agents has attracted undirected networks with switching topology and time-delay,
great attention in recent years. This is partly due to recent varying or not. Based on local information, we not only
technological advances in communication and computation, introduce two linear protocols, but also, for each case, give
and a number of new important applications ranging from un- sufficient conditions by a dimension reduction approach. In
manned aerial vehicles, automated highway systems, under- addition, the cases of fixed topology with and without time-
water vehicles to mini-satellites, communication networks, delay are addressed separately in another paper[17], which
etc. Quite a tremendous amount of interesting results have gives sufficient and necessary conditions for directed or
been addressed [1]-[19]. In [1], Vicsek et al. proposed a undirected network.
simple model for phase transition of a group of self-driven The paper is organized as follows. In Section II, we
particles and numerically demonstrated complex dynamics provide some results of graph theory and Kronecker product
of the model. In [2], Jadbabaie et al. provided a theoretical and define the consensus problem, the agreement of the
explanation for the consensus behavior of the Vicsek model position states and the synchronization of the speed states.
using graph theory. Also, Moreau [3] used a set-valued In section III, two linear protocols are introduced. The
Lyapunov approach to study consensus problems with uni- main stability results are derived in section IV. Section V
directional time-dependent communication links. Moreover, gives numerical simulations. Some conclusions are drawn in
Olfati-Saber et al. [4] solved the average-consensus problem section VI.
with directed interconnection graphs or time-delays by a
Lyapunov-based approach. However, most of them assume II. G RAPH THEORY AND CONSENSUS PROBLEM
that each agent has single integrator dynamics. As a matter
A. Graph theory
of fact, a broad class of agents, such as unmanned aerial
vehicles, underwater vehicles and so on, are adjusted to Let G(V , ε , A ) be an undirected graph of order n with the
produce desired motion directly by their accelerations rather set of nodes V = {s1 , · · · , sn }, the set of edges ε ⊆ V × V ,
than by their speeds, and they all have double integrator dy- and a weighted adjacency matrix A = [ai j ] with nonnegative
namics. Thus, it is quite necessary to find suitable protocols elements. The node indexes belong to a finite index set I =
or control laws for such agents. Tanner et al. [8][9] introduced {1, 2, · · · , n}. An edge of G is denoted by ei j = (si , s j ). The
a set of control laws that enable the second-order agents to adjacency elements associated with the edges are positive,
i.e., ei j ∈ ε ⇔ ai j > 0. Moreover, we assume aii = 0 for all i ∈
This work was supported by the NSFC (60374001,60334030), the MOE I. Correspondingly, the graph Laplacian with the undirected
(20030006003) and the COSTIND (A2120061303).
Peng Lin and Yingmin Jia are with the Seventh Research graph is defined as L = [li j ], where lii = ∑ j ai j and li j =
Division, Beihang University (BUAA), Beijing 100083, P.R.China. −ai j ,i = j. The set of neighbors of node si is denoted by
Lin [email protected],[email protected] Ni = {s j ∈ V : (si , s j ) ∈ ε }.
Junping Du is with the Beijing Key Laboratory of Intelligent Telecom-
munications Software and Multimedia, School of Computer Science and A path is a sequence of ordered edges of the form
Technology, Beijing University of Posts and Telecommunications, Beijing (si1 , si2 ), (si2 , si3 ), · · · , where si j ∈ V in an undirected graph.
100876, [email protected] If there is a path from every node to every other node, the
Shiying Yuan is with the School of Electrical and Automa-
tion, Henan Polytechnic University, Jiaozuo 454000, Henan, P.R.China graph is said to be strongly connected. Note that since the
[email protected] graph considered is undirected, it means once ei j is an edge

ISBN 978-3-9524173-8-6 2841

 WeA08.6

of G, e ji is an edge of G as well. As a result, the adjacency state feedback controller for each agent not only to solve
matrix A is a symmetric nonnegative matrix. the agreement of the position states of network but also to
synchronize the speed states of the network.
Lemma 1[4] If the undirected graph G is strongly con-
nected, then the Laplacian L of the graph has the following III. P ROTOCOL AND NETWORK DYNAMICS
properties: In this section, we introduce two linear consensus proto-
(1) rank(L) = n − 1. cols that solve the pv-consensus problem in networks with
(2) Zero is one eigenvalue of L, and 1n is the correspond- switching topology and zero or nonzero time-delay.
ing eigenvector, i.e., L1n = 0. i) Switching topology and zero communication time-delay
(3) The rest n − 1 eigenvalues are all positive and real.
B. Kronecker product
ui (t) = k1 ∑ ai j (x j − xi ) + k2 ∑ ai j (v j − vi ). (A1)
s j ∈Ni s j ∈Ni
Kronecker product will be used as one of the main tools ii) Switching topology and nonzero communication time-
to study the stability of the protocols, so it is necessary delay
to introduce some very basic properties of the Kronecker
product. ui (t) = k1 ∑ ai j (x j (t − τ ) − xi (t − τ ))
For any X0 ,Y0 , Z0 , D0 ∈ R n×n and a0 ∈ R, s j ∈Ni

(1) (a0 X0 ) ⊗Y0 = X0 ⊗ (a0Y0 ) = a0 (X0 ⊗Y0 ), +k2 ∑ ai j (v j (t − τ ) − vi (t − τ )) (A2)

s j ∈Ni
(2) (X0 +Y0 ) ⊗ Z0 = X0 ⊗ Z0 +Y0 ⊗ Z0 ,
Z0 ⊗ (X0 +Y0 ) = Z0 ⊗ X0 + Z0 ⊗Y0 , Let
ξ = [x1 , v1 , x2 , v2 , · · · , vn ]T
(3) (X0 ⊗Y0 )(Z0 ⊗ D0 ) = (X0 Z0 ) ⊗ (Y0 ⊗ D0 ),    
0 1 0 0
(4) (X0 ⊗Y0 )T = X0T ⊗Y0T . A= , B=
0 0 k1 k2
C. Consensus problem in networks Using protocol (A1) the network dynamics can be given
We assume that each agent is a node in an undirected as
graph, G. ξ̇ = Φsw ξ (2)
Let xi be the position state of the ith agent, vi be the speed similarly, using protocol (A2) the network dynamics is
state. Suppose each agent has the dynamics as follows
ξ̇ = In ⊗ Aξ − Lsw ⊗ Bξ (t − τ ) (3)
ẋi = vi
where Lsw is the Laplacian of the graph Gsw , Φsw = In ⊗ A −
mi v̇i = ui (1)
Lsw ⊗ B and sw(t) : [0, +∞) → {1, · · · , N}(N ∈ Z + , denotes
where mi is the mass and ui is the control input. Moreover, the total number of all possible undirected graphs ) is a
we assume m1 = m2 = · · · = mn = 1. switching signal that determines the topology.
Let pv(position-velocity):R 2n+1 → R be a function IV. S TABILITY ANALYSIS OF NETWORK DYNAMICS
of 2n + 1 variables x1 , x2 , · · · , xn , v1 , · · · , vn ,t and x0 =
x(0), v0 = v(0) denote the initial states of the system. The In this section,we will provide the stability analysis in two
pv−consensus problem in a dynamic graph is distributed way cases: 1) undirected networks with switching topology and
to calculate pv(x0 , v0 ) by applying inputs ui that only depend zero communication time-delay; 2) undirected networks with
on the states of node si and its neighbors. switching topology and nonzero communication time-delay.
We say protocol ui asymptotically solves the A. Networks with switching topology and zero communica-
pv−consensus problem if and only if the states of tion time-delay
agents satisfy Lemma 2 Assume the communication topology Gsw is
lim (xi − x j ) = 0 kept strongly connected. Consider an undirected network of
t→+∞

and agents with switching topology and zero time-delay. The

lim (vi − v j ) = 0, linear system (2) is equivalent to
t→+∞
δ̇ (t) = Φsw δ (t) (4)
for all i, j ∈ I. Furthermore, if the speeds satisfy  T
1 where δ (t) = ξ (t)−1n ⊗ α β with α = n ∑i=1 xi (0)+
1 n

n∑
lim vi = v j (0), β t and β = n Σi=1 vi (0).
1
t→+∞
j
Proof: Notice that the Laplacian
 Lsw of the graph Gsw
we say protocol ui asymptotically solves a pv−average
is symmetric, and hence 1Tn ⊗ 0 1 Φsw = 02n , which
consensus problem.
implies ∑ni=1 v̇i = 0. Thus, α is the average position of the
In this paper, our work focuses on the distributed solution
agents at time t, β is the average speed, and
of pv−consensus problem. To solve such problem is a  
challenging task. It needs one to find suitable distributed 1Tn ⊗ 1 0 δ (t) = 0,

2842
 WeA08.6

 
1Tn ⊗ 0 1 δ (t) = 0. where
   
0 2 2k1 k1 + k2
Consequently, (2) can be rewritten as H = In ⊗ − Lsw ⊗
 T 2 2 k1 + k2 2k2
ξ̇ (t) = δ̇ (t) + 1n ⊗ β 0 Since the graph is undirected and strongly connected, by
 T
= In ⊗ Aδ (t) − L ⊗ Bδ (t) + 1n ⊗ β 0 . Lemma 1, there exists an orthogonal matrix W satisfying

Hence, system (4) is equivalent to (2).  E = W T LswW = diag{0, λ2 , · · · , λn }

Lemma 3 Consider the matrix and
⎛ ⎞ λ2 > 0, λ3 > 0, · · · , λn > 0.
n − 1 −1 · · · −1
⎜ −1 n − 1 · · · −1 ⎟
⎜ ⎟ Then, we have
Ā = ⎜ . .. .. .. ⎟.
⎝ .. . . . ⎠ M = (W T ⊗ I2 )H(W ⊗ I2 )
−1 −1 ··· n−1    
0 2 2k1 k1 + k2
= In ⊗ −E ⊗
The following statements hold. 2 2 k1 + k2 2k2
 
(1) The eigenvalues of Ā are n with multiplicity n − 1 and 0 2
= diag{ , Γ2 , · · · , Γn }
zero with multiplicity 1. 2 2
(2) The vectors 1Tn and 1n are the left and the right  
−2k1 λi 2 − (k1 + k2 )λi
eigenvectors of Ā associated with the zero eigenvalue. where Γi = ,
2 − (k1 + k2 )λi 2 − 2k2 λi
(3) There exists an orthogonal matrix U such that i = 2, · · · , n.
 
nIn−1 0(n−1)×1 Considering the eigenpolynomial of M, we have
U T ĀU =
01×(n−1) 0 det(M − sI2n ) = det(C̄ − sI2 )Πni=2 φi
 
and the last column is √1nn . Let B̄ be the Laplacian of any 0 2
undirected graph with n nodes, then where C̄ = , φi = s2 + (−2 + 2k1 λi + 2k2 λi )s +
2 2
  4k2 λi + 2k1 k2 λi2 − 4 − (k12 + k22 )λi2 .
∗ 0(n−1)×1
U T B̄U = . Now, let us find sufficient conditions such that the zeros
01×(n−1) 0
of φi are all located on the open LHP. Set
Remark 1 Note that 1Tn B̄ = 0Tn , B̄1n = 0n , and the last
column of U is √1nn , it is easily obtained that −2 + 2k1 λi + 2k2 λi > 0
  4k2 λi + 2k1 k2 λi2 − 4 − (k12 + k22 )λi2 > 0
∗ 0(n−1)×1
U B̄U =
T
.
01×(n−1) 0 Namely,
k1 λi + k2 λi > 1 (6)
Theorem 1 Consider an undirected network of agents √ √
with switching communication topology Gsw that is kept k2 λi − 4k2 λi − 4 k2 λi + 4k2 λi − 4
strongly connected. Then, protocol (A1) globally asymptoti- < k1 < (7)
λi λi
cally solves the pv-average consensus problem if k2 λmin > 1,
Obviously, a sufficient condition for (6)(7) is k2 λmin >
c1 < k1 < c2 , where
√ 1, c1 < k1 < c2 .
k2 λi − 4k2 λi − 4 Recall that  
c1 = max { }, 1Tn ⊗ 1 0 δ (t) = 0
Lsw ,λi =0 λi
√  
k2 λi + 4k2 λi − 4 1Tn ⊗ 0 1 δ (t) = 0
c2 = min { },
Lsw ,λi =0 λi and due to the fact that the first column of W is the
c1 < c2 , λmin = min λi . eigenvector of Lsw associated with eigenvalue 0, i.e. √1nn , we
λi =0 have
Proof: Define a common Lyapunov function for system V̇ = δ T (t)[(W ⊗ I2 )M(W T ⊗ I2 )]δ (t)
(2) as follows
= δ̄ T (t)M δ̄ (t)
V = δ T (t)Pδ (t) (5)  T
  δ̄ = (W T ⊗ I2 )δ (t) = 0 0 ∗ · · · ∗
2 1
where P = In ⊗ .
1 1 Since W T ⊗ I2 is an orthogonal matrix, then δ̄ (t) = 0 if
Calculating V̇ , we get and only if δ (t) = 0. Thus, if k2 λmin > 1, c1 < k1 < c2 and
δ (t) = 0 are satisfied, then V̇ < 0. So,
V̇ = δ T (t)(PΦsw + ΦTsw P)δ (t)
= δ T (t)H δ (t) lim δ (t) = 02n ,
t→+∞

2843
 WeA08.6

namely, we find that

1 n 1 n −2δ T (t)P(Lsw ⊗ B)δ (t − τ )

lim [xi −
t→+∞
∑
n i=1
xi (0) − ∑ vi (0)t] = 0
n i=1 = −2δ T (t)P(Lsw ⊗ B)δ (t)
t
n + 2((Lsw ⊗ B)T PT δ (t))T δ̇ (s)ds
1 t−τ
lim vi = ∑ vi (0) t
t→+∞ n i=1
≤ −2δ T (t)P(Lsw ⊗ B)δ (t) + β δ̇ T (s)δ̇ (s)ds
t−τ
This completes the proof.  τ T
+ δ (t)P(Lsw
2
⊗ BBT )Pδ (t)
β
Corollary 1 Assume the communication topology Gsw is
kept strongly connected. Consider an undirected network of Hence we have
agents with switching topology and zero time-delay. Given τ T
protocol (A1) with k1 = k2 , pv-average consensus can be V̇ ≤ 2δ T (t)PΦsw δ (t) + δ (t)P(Lsw2
⊗ BBT )Pδ (t)
β
achieved if k1 > maxLsw ,λi =0 { λ1i }. + δ T (t)Qδ (t) − δ T (t − τ )Qδ (t − τ )
 
0 0
B. Networks with switching topology and nonzero commu- + β τδ (t)In ⊗
T
δ (t) (10)
0 1
nication time-delay
+ β τδ T (t − τ )(Lsw
2
⊗ BT B)δ (t − τ )
Lemma 4 Assume the communication topology Gsw is
kept strongly connected. Consider an undirected network of A sufficient condition for V̇ < 0 is
agents with switching topology and time-delay. The system  
0 0
(3) is equivalent to δ T
(t)U2U2T (PΦsw + ΦTsw P + Q + β τ In ⊗
0 1
δ̇ (t) = In ⊗ Aδ (t) − Lsw ⊗ Bδ (t − τ ) (8) τ
+ P(Lsw
2
⊗ BBT )P)U2U2T δ (t) < 0 (11.a)
 T β
where δ (t) = ξ (t)−1n ⊗ α β with α = 1n ∑ni=1 xi (0)+
β t and β = n Σi=1 vi (0).
1 δ T (t − τ )U2U2T (β τ Lsw
2
⊗ BT B − Q)U2U2T δ (t − τ ) < 0
(11.b)
Proof: Its proof is very similar to that of Lemma 2 and where U2 = U ⊗ I2 and U has been defined in Lemma 3.
hence omitted.   
Note that 1n ⊗
T
1 0 δ (t) = 0, 1n ⊗
T
0 1 δ (t) = 0,
Theorem 2 Consider an undirected network of agents with and the last column of U is √1nn . Thus, it follows that
fixed time-delay and switching communication topology Gsw  T
that is kept strongly connected. Given protocol (A2), pv- U2T δ (t) = ∗ ∗··· 0 0 ,
average consensus can be achieved if k2 λmin > 1, c1 < k1 < c2
and τ < d, where d is an estimation for the upper bound of also, by Lemma 3,
time-delay and c1 , c2 , λmin have been defined in Theorem 1. U2T (PΦsw + ΦTsw P)U2
 
Proof: Define a Lyapunov function for system (8) as 0 2
= In ⊗ +U T LswU ⊗ Ha
follows 2 2
⎛ ⎞
t T 0(2n−2)×2
V = δ T (t)Pδ (t) + δ T (s)Qδ (s)ds = ⎝ 0 2 ⎠
t−τ 02×(2n−2)
0 t 2 2
+ β δ̇ T (s)δ̇ (s)dsd θ (9) U2T P(Lsw
2
⊗ BBT )PU2
−τ t+θ  2 
  k1 + k22 k12 + k22
2 1 = (U LswU) ⊗
T 2
where P = In ⊗ , β > 0, and Q is a positive-semi- k12 + k22 k12 + k22
1 1  
T1 02×(2n−2)
definite matrix. =
Calculating V̇ , we get 0(2n−2)×2 02×2
U2T P(Lsw
2
⊗ BT B)PU2
V̇ = 2δ (t)P(In ⊗ A)δ (t) + δ (t)Qδ (t)
T T
= (U T Lsw 2
U) ⊗ (Hb HbT )
t  
− δ T (t − τ )Qδ (t − τ ) − β δ̇ T (θ )δ̇ (θ )d θ T2 02×(2n−2)
t−τ =
0(2n−2)×2 02×2
+ β τ δ̇ T (t)δ̇ (t) − 2δ T (t)P(Lsw ⊗ B)δ (t − τ )    
2k1 k1 + k2 0 2k1 + k2
t Ha = , Hb = .
Since δ (t − τ ) = δ (t) − t−τ δ̇ (s)ds, for any x, y ∈ R n , k1 + k2 2k2 0 k1 + k2
From the proof of Theorem 1, all the eigenvalues of T are
2xT y ≤ xT x + yT y located on the LHP.

2844
 WeA08.6

Choose semi-positive matrix Q satisfying Proof of Theorem 3 In fact, this theorem can be proven
  following the lines of the proof of Theorem 2. The main
1
Q(1n ⊗ ) = 02n ideas used in that proof essentially remain valid even if the
0
  delay is time varying in the equations.
0
Q(1n ⊗ ) = 02n , Remark 3 Note that the cases studied in Theorems 1,2,3
1
are under arbitrary switching, and all the results will hold
and it follows that even if sw(t) switches arbitrarily fast in a small time interval.
 
Q̄ 0(2n−2)×2 And it is possible to relax the condition “arbitrary switching”
U2T QU2 =
02×(2n−2) 02×2 to improve the conservativeness.
 
0 0
δ T
(t)U2U2T (PΦsw + ΦTsw P + β τ In ⊗
0 1 V. S IMULATION RESULTS
τ
+Q + P(Lsw 2
⊗ BBT )P)U2U2T δ (t)
β In this section, we will present some numerical simulations
 
T̄ 0(2n−2)×2 to illustrate the theoretical results obtained in the previous
= δ̄ T (t) δ̄ (t)
02×(2n−2) Z sections. These simulations are performed with six agents,
whose initial conditions are set randomly. Fig.1 denotes
δ T (t − τ )U2U2T (β τ Lsw
2
⊗ BT B − Q)U2U2T δ (t − τ ) topology structures and the switching course. Here, we
  assume that the weight of each edge is 1. And by simple
T¯1 0(2n−2)×2
= δ̄ T (t − τ ) δ̄ (t − τ ) computation, we get the smallest nonzero eigenvalue of the
02×(2n−2) 02×2
 T graph Laplacians in Fig.1 is 2.5858. Fig.2 and Fig.3 show
δ̄ (t) = (U T ⊗ I2 )δ (t) = ∗ · · · ∗ 0 0 , the state trajectories for the cases of switching topology with
and without time-delay, respectively.
T̄ , T̄1 ∈ R (2n−2)×(2n−2) , Z ∈ R 2×2
It is obvious that (11) holds if and only if
1 2 3 1 2 3
T̄ + T̄ T < 0, T̄1 + T̄1T < 0 (12)
Then, an upper bound of time-delay d can be obtained 6 5 4 6 5 4

by solving (12). Otherwise, if the inequality (12) holds, then

V̇ < 0. So,
Gb

lim δ (t) = 02n ,

Gc

t→+∞ 1 2 3

namely,
1 n 1 n 6 5 4
lim [xi −
t→+∞
∑
n i=1
xi (0) − ∑ vi (0)t] = 0
n i=1 Gd

1 n Fig.1
lim vi =
t→+∞
∑ vi (0)
n i=1 A Switching topology with zero communication time-
This completes the proof.  delay

Remark 2 In Theorem 2, an upper bound of time-delay k1 = k 2 = 1

d can be obtained through the following two steps. 70

Step 1) Caculate T, T1 , T2 and set τ = 0, β = β0 , and step

size h = h0 , where β0 and h0 are given constants.
60

Step 2) Based on LMI(linear matrix inequality) approach, 50

search a positive matrix Q̄ that satisfies the matrix inequality 40

(12). If Q̄ can be found, then set τ = τ + h0 and restart step 30

2). Or else, stop and τ is an upper bound of time-delay.
20

Theorem 3 Assume the communication topology Gsw

10
is strongly connected. Consider an undirected network of
agents with switching topology and time-varying time-delay. 0

Given protocol (A2), pv-average consensus can be achieved −10

if 0 ≤ τ < d, τ̇ ≤ d1 , k2 λmin > 1, and c1 < k1 < c2 , where d

−20
is an estimation for the upper-bound of time-delay, and d1 0 5 10 15

is a constant given as d1 < 1. Fig.2(a) Position trajectories of the network

2845
 WeA08.6

20
guarantee the convergence of the protocols under arbitrary
15
switching. In particular, for the delay-dependent case, it was
10
proved that an upper bound for time-delay can be easily
5 obtained by solving a set of matrix inequalities even if the
0 time-delay is time-varying. The cases of the fixed topology
−5
with zero or nonzero time-delay are addressed separately in
another paper[17].
−10

−15
R EFERENCES
−20
[1] T. Vicsek, A. Cziroók, E. Ben-Jacob, O. Cohen, and I. Shochet, “Novel
type of phase transition in a system of self-deriven particles,” Physical
−25 Review Letters, vol. 75, no. 6, pp. 1226-1229, August 1995.
[2] A. Jadbabaie, J. Lin, and A. S. Morse, “Coordination of groups
−30
0 5 10 15 of mobile autonomous agents using nearest neighbor rules,” IEEE
Fig.2(b) Velocity trajectories of the network Transactions on Automatic Control, vol. 48, no. 6, pp. 988 - 1001,
2003.
[3] L. Moreau, “Stability of multi-agent systems with time-dependent
B Switching case with nonzero communication time-delay communication links. IEEE Transactions on Automatic Control”, vol.
50, no.2, pp.169-182, 2005.
k1 = k2 = 4, τ = 0.025 [4] R. O. Saber and R. M. Murray, “Consensus problems in networks of
70 agents with switching topology and time-delays,” IEEE Transactions
on Automatic Control, vol. 49, no. 9, pp. 1520-1533, 2004.
60
[5] –, “Agreement problems in networks with directed graphs and switch-
50 ing topology,” in Proceedings of the 42nd IEEE Conference on
Decision and Control, 2003, pp. 4126-4132.
40
[6] –, “Consensus protocols for networks of dynamic agents,” in Proceed-
30
ings of the 22nd American Control Conference, June 2003, pp. 951-
956.
20 [7] A. Fax and R. M. Murray, “Information flow and cooperative control
of vehicle formations,” IEEE Transactions on Automatic Control, vol.
10
49, no.9, pp. 1465-1476, Sept. 2004.
0 [8] H. Tanner, A. Jadbabaie, and G. Pappas, “Stable flocking of mobile
agents, part I : Fixed topology,” in Proceedings of the 42nd IEEE
−10
Conference on Decision and Control, 2003, pp. 2010-2015.
−20 [9] –, “Stable flocking of mobile agents, part II : Dynamic topology,” in
Proceedings of the 42nd IEEE Conference on Decision and Control,
−30
0 10 20 30 40 50 60 70 80 90 100 2003, pp. 2016-2021.
[10] W. Ren and R. W. Beard, and E. M. Atkin, “A survey of consensus
Fig.3(a) Position trajectories of the network problems in multi-agent coordination,” in Proceedings of the 22nd
8 American Control Conference, June 2005, pp. 1859-1864.
[11] Y. Hong, J. Hu and L. Gao,“Tracking control for multi-agent consensus
6 with an active leader and variable topology”, Automatica, vol. 42, no.7,
pp.1177-1182, 2006.
4
[12] W. Ren and R. W. Beard, “Consensus seeking in multi-agent systems
2
under dynamically changing interaction topologies,” IEEE Transac-
tions on Automatic Control, vol. 50, no. 5, pp. 655-661, May 2005.
0 [13] Y. Liu, K. M. Passino, and M. M. Polycarpou, “Stability analysis
of M-dimensional asynchronous swarms with a fixed communication
−2 topology,” IEEE Transactions on Automatic Control, vol. 48, no.1, pp.
76-95, Jan. 2003.
−4
[14] C.W. Reynolds, “Flocks, herds, and schools: a distributed behavioral
−6
model”, in: ACM SIGGRAPH’87 Conference Proceedings, Computer
Graphics 21 (1987) 25-34.
−8 [15] Wei Ren, Ella Atkins, “Distributed multi-vehicle coordinated control
via local information exchange”, International Journal of Robust and
−10
0 10 20 30 40 50 60 70 80 90 100 Nolinear Control, Published online, 8 Nov. 2006.
[16] Peng Lin, Yingmin Jia, Junping Du, “Average consensus in networks
Fig.3(b) Velocity trajectories of the network of agents with both switching topology and time-delay”, submitted.
[17] Peng Lin and Yingmin Jia, “Distributed Consensus Control for Net-
In addition, according to [17,Theorem 2], we can get the works of Second-Order Agents With Fixed Topology and Time-Delay,”
largest tolerable time-delay τ = 0.0637 with fixed topology to appear in proceedings of the 26th Chinese Control Conference.
for all topology structures shown in Fig.1. [18] G. Xie, L. Wang, “Consensus Control for a Class of Networks of
Dynamic Agents: Fixed Topology”, IEEE Conference on Decision and
Control, 2005, 96-101.
VI. CONCLUSIONS [19] G. Xie, L. Wang, “Consensus Control for a Class of Networks of
Dynamic Agents: switching Topology”, in Proceedings of the 25th
Distributed consensus control has been considered in American Control Conference, June 2006, pp. 1382 - 1387.
networks of agents with double integrator dynamics for [20] C. Godsil and G. Royle, Algebraic Graph Theory. New York: Springer-
two cases, undirected networks with switching topology and Verlag, 2001, vol. 207, Graduate Texts in Mathematics.
[21] R. A. Horn and C. R. Johnson, Matrix Analysis. Cambridge, U.K.:
undirected networks with switching topology and time-delay. Cambridge Univ. Press, 1987.
Two linear consensus protocols with and without time-delay [22] L.Yu. Robust control: An LMI approach, Tsinghua University Press,
were introduced. Sufficient conditions were derived which 2002.(In Chinese)

2846