3780 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 64, NO. 9, SEPTEMBER 2019
Bounds on Delay Consensus Margin of Second-Order Multiagent
Systems With Robust Position and Velocity Feedback Protocol
Dan Ma
, Senior Member, IEEE, Rui Tian , Adil Zulfiqar , Jie Chen , Fellow, IEEE,
and Tianyou Chai
, Fellow, IEEE
Abstract—This paper studies the delay consensus margin and
its bounds for second-order multiagent systems to achieve robust
consensus with respect to uncertain delays varying within a range.
This paper attempts to answer the question: What is the largest
delay range within which a control protocol is able to achieve and
maintain the consensus? We consider second-order agents with
unstable poles, which communicate over an undirected network
topology, and derive explicit bounds on the delay consensus mar-
gin. The results show that the consensuability robustness of such
unstable agents depends on the pole locations of the agents, as
well as on the eigenratio of the network graph.
Index Terms—Delay consensus margin (DCM), multiagent sys-
tems (MASs), second-order agents, time delay.
I. INTRODUCTION
In a multi-agent system (MAS), agents are to exchange information
over a communication network, which, invariably, is prone to trans-
mission delays, due to, for example, communication congestions and
transmission bandwidth. It is known that time delay generally degrades
a system’s performance, and this remains so, if not more acute, for a
MAS. As such, communication delays in MASs must be accounted for.
Over the past decade, MAS consensus problems with time delays have
been well studied; see, e.g., [1], [3], [4], [6]–[11], [13], and [14]. Var-
ious agent models have been considered, including single-integrator
agents, double-integrator agents [1], [3], [6], [7], high-order agents
[13], [14], and, furthermore, nonlinear agents. Upper bounds on ho-
mogenous delays were obtained, which provide a range of delay that a
MAS of single-integrator agents can achieve consensus by a constant
feedback protocol over a fixed undirected network topology. Other sim-
ilar results can be found in, e.g., [1], [9]–[11], and [22], which also seek
to ascertain delay ranges to ensure consensus and to a certain degree to
achieve performance of first-order MASs.
Manuscript received April 13, 2018; revised April 22, 2018 and
September 11, 2018; accepted November 20, 2018. Date of publication
November 29, 2018; date of current version August 28, 2019. This work
was supported in part by National Natural Science Foundation of China
under Grant 61603079 and Grant 61773098 and in part by the Hong
Kong RGC under Project CityU 11201514 and Project CityU 111613.
Recommended by Associate Editor G. Gu. (Corresponding author: Dan
Ma.)
D. Ma and R. Tian are with the College of Information Science and
Engineering, State Key Laboratory of Synthetical Automation for Process
Industries, Northeastern University, Shenyang 110819, China (e-mail:,
madan@mail.neu.edu.cn; 364368795@qq.com).
A. Zulfiqar and J. Chen are with the Department of Electronic Engi-
neering, City University of Hong Kong, Hong Kong (e-mail:, azulfiqar2-
c@my.cityu.edu.hk; jichen@cityu.edu.hk).
T. Chai is with the State Key Laboratory of Synthetical Automation for
Process Industries, Northeastern University, Shenyang 110819, China
(e-mail:,tychai@mail.neu.edu.cn).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAC.2018.2884154
The consensus problem for second-order agents proves harder. As
expected, the conditions sufficient for the consensus of first-order
MASs are hardly adequate to ensure second-order MASs to achieve
the consensus [2]. There has since ensued a multitude of works on
MASs of second-order agents, of which double-integrators provide a
notable case of study. In [3], consensus conditions were given for the
double-integrator MASs with a constant communication delay over
a fixed network topology. With given consensus protocols, neces-
sary and sufficient consensus conditions were given in [6]–[8] and
[23]. Furthermore, robust consensuability problems were studied in
[4], [7], and [10], resulting in conditions that ensure double-integrator
MASs to maintain consensus despite that the delay may vary within
a range. Robust consensus problems were addressed in, e.g., [15] and
[16], with respect to time-varying graph topologies and time-varying
delays.
In this paper, we focus on the robust consensus problem of more
general s econd-order MASs under an undirected network, in which the
agents may have strictly unstable poles, and the control protocol con-
sists of position and velocity feedback. Consensus analysis problems
for such second-order MASs have been previously studied in, e.g.,
[6] and [8], where the allowable delay range for consensus is deter-
mined for given position and velocity feedback. Our work goes beyond
to address the synthesis of optimal position and velocity protocols to
maximize the delay range possible. Motivated by [5] and [12], we in-
troduce the notion of delay consensus margin (DCM) to characterize
the maximal delay range and derive explicit bounds on the DCM. The
upper bounds give the range beyond which no such protocol may ex-
ist to achieve robust consensus for all possible delay values within the
range, while within the lower bounds, robust consensus is guaranteed to
be achievable. The results consequently provide conditions for robust
consensuability of general second-order MASs. The bounds, obtained
by optimizing the consensus protocol, show that the robust consensua-
bility conditions depend critically on the pole locations of the agents
and on the eigenratio of the network graph, whereas the latter is known
to be a measure of network connectivity.
Partial results of this paper are reported in its abridged conference
version [20]. The present paper provides complete proofs of the tech-
nical results along with numerical examples unavailable from [20].
II. P
ROBLEM FORMULATION
A. Rudiments of Algebraic Graph Theory
We begin by collecting from [21] some basic elements of algebraic
graph theory. A graph of order N can be represented as G =(V, E, A),
where V = {1,...,N} is the node set with each node representing
an agent, E⊂V×Vis an edge set of paired nodes, and A =[a
ij
]
is an N ×N adjacency matrix of the graph G. If an edge (i, j) ∈E,
the jth node can obtain information from the ith node. The node i is
called a neighbor of node j. The set of neighbors of node i is denoted
as N
i
{j|(j, i) ∈E}. The graph G is said to be undirected if for all
m, n ∈V, (m, n) ∈Eimplies that (n, m) ∈E.Apath from node
n
1
to node n
k
is a sequence of nodes n
1
,...,n
k
, such that for each i,
1 ≤ i ≤ k − 1, (n
i
,n
i +1
) is an edge. A graph is said to be connected
if there exists a path from any node to any other node and complete if
0018-9286 © 2018 IEEE. Personal use is per mitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications
standards/publications/rights/index.html for more information.