BOUNDED CONSENSUS TRACKING FOR SAMPLED-DATA SECOND-ORDER MAS 253
Because of the existing communication failures, packet dropouts and unstable communication,
it is necessary to study the coordinated tracking problem in a stochastic setting, where the interac-
tion graph evolves according to some random distributions. Although there have been some papers
concerning the stochastic consensus of MASs (see [12, 13] and the references therein for details),
few results on coordinated tracking with Markovian switching topology are available in the existing
literature. Recently, Zhao et al. [1] studied discrete-time coordination tracking problem for single-
integrator MAS with Markovian switching topology and showed the ultimate bound of the tracking
errors. However, the system model investigated in [1] is single-integrator MAS, and the switching
is assumed to be equiprobable among all the subgraphs. This is one of the important reasons for us
to write this paper.
On the other hand, all the aforementioned papers concerning the coordination tracking prob-
lem are based on a common assumption: Each agent can measure the leader’s information accu-
rately. Obviously, this assumption is ideal for real communication channels because the discrepancy
between the mathematical model and the actual dynamics of the system in operation is unavoidable.
Furthermore, it may adversely affect the stability and consensus or other performances of a control
system. As far as we know, few papers have investigated this kind of uncertainties. This is another
important reason for us to write this paper.
This paper will focus on the bounded consensus tracking of second-order MASs with fixed topol-
ogy and Markovian switching topologies in a sampling setting, respectively. Similar to [1], only
some agents are assumed to have the access to the leader because of the low communication cost.
The leader’s acceleration is assumed to be not measured, and all the agents can only obtain its
approximation. By virtue of matrix analysis and perturbation theory, we present necessary and
sufficient conditions for boundedness of tracking error system and show the ultimate bound of track-
ing errors under fixed and Markovian switching topology, respectively. Our paper is different from
[1] in system model, communication topology constraints, control protocol, and the accuracy of the
leader’s information. On the other hand, Zhang and Tian [12] and Chen and Li [13] investigated
the consensus problem of MAS with Markovian switching topology, not the bounded consensus
tracking problem. Their system models neglected the uncertainties between the actual value and
the approximative value. How to deal with the uncertainties and the Markovian switching topology
becomes a key problem in our article.
Notations: We use standard notations throughout this paper. R, C,andZ
C
denote the sets of real
numbers, complex numbers, and positive integers, respectively. I
n
represents the identity matrix of
dimension n,andI denotes the identity matrix of an appropriate dimension. 1
n
is a vector with
all entries equal to 1. diag¹
1
, ,
n
º represents a diagonal matrix with
i
, i D 1, , n on its
diagonal. Let diag¹A
1
, , A
n
º denote the diagonal matrix with diagonal block A
i
, i D 1, 2, , n.
./ and det./ represent the spectral radius and determinant of a matrix, respectively. kxk
1
, kxk
2
and kAk
1
, kAk
2
denote the inf-norm and 2-norm of vector x and A, respectively. Given X.k/ 2 R
p
,
define kX.k/k
E
, kEŒX.k/X
T
.k/k
2
,whereEŒ is the mathematical expectation. Let ƒ.A/
denote all the eigenvalues of matrix A.Fors 2 C, jsj, Re.s/,andI m.s/ denote its modulus, real
and imaginary part, respectively.
2. PROBLEM FORMULATIONS AND PRELIMINARIES
In this section, some basic knowledge on graph theory, problem formulations, some definitions, and
lemmas are given as the preliminaries of this paper.
2.1. Graph theory
Let G D .V, E , A/ be a weighted directed graph of order n with the set of nodes V D
¹v
1
, v
2
, , v
n
º, set of edges E V V , and the nonsymmetric weighted adjacency matrix
A D Œa
ij
2 R
nn
with real adjacency elements a
ij
. The node indexes belong to a finite index
set ` WD¹1, 2, , nº. An edge of G is denoted by e
ij
D .v
j
, v
i
/. The adjacency elements associated
with the edges of the graph are nonzero, that is, e
ij
2 E if and only if a
ij
¤ 0. Moreover, we assume
a
ii
D 0 for all i 2 `. The set of neighbors of node v
i
is denoted by N
i
D¹v
j
2 V W .v
j
, v
i
/ 2 Eº.
L.G/ DŒl
ij
is the Laplacian matrix of topology G and is defined by
Copyright © 2013 John Wiley & Sons, Ltd. Int. J. Robust. Nonlinear Control 2015; 25:252–268
DOI: 10.1002/rnc