Distributed Flocking of Second-order Multi-agent
Systems with Global Connectivity Maintenance
Yutian Mao , Lihua Dou, Hao Fang and Jie Chen, Senior Member, IEEE, Tao Cai
Abstract—This paper investigates the problem of connectivity-
preserving flocking of multiple autonomous agents with second-
order dynamics. First, the inverse power iteration algorithm
is formulated in a completely distributed manner to estimate
the algebraic connectivity, i.e., the second smallest eigenvalue of
the group Laplacian, as well as the corresponding eigenvector.
Furthermore, distributed gradient-based flocking algorithms that
exploit decentralized eigenvalue/eigenvector estimation are devel-
oped both to steer the agent group to the desired flocking motion
and to maintain the global connectivity of the underlying network
during maneuvers. Different from the common potential/tension
function method which keeps certain fixed edges all the time,
the algorithm proposed in this paper guarantees the global
connectivity which allows any existing edge to be broken, thus
gives more freedom of motions for the agents. Finally, nontrivial
simulations are performed to demonstrate the correctness and
effectiveness of the theoretical results.
I. INTRODUCTION
Recently, cooperative flocking has emerged as a robust way
of addressing a wide variety of spatially distributed tasks.
During the last decade, considerable effort has been made in
studying emergent group flocking behavior among researchers
in various fields. Reynolds proposed the boids model in
[1], in which three heuristic rules of separation, cohesion
and alignment were introduced. Vicsek et al. presented a
simple model of self-propelled particles which can achieve the
velocity alignment [2]. Jadbabaie et al. provided the theoretical
explanation for the observations of Vicsek ˛a
´
rs model based on
nearest neighbor rules [3]. Olfati-Saber presented a compre-
hensive theoretical framework of distributed flocking control
with a virtual leader [4]. Tanner et al. introduced local control
laws via artificial potential fields to obtain stable flocking
motion in both fixed and switching networks [5]. However,
the aforementioned works do not address the important issue
of control over connected underlying networks, which is vital
for network-wide collaboration and the success of the entire
mission.
As a common property, most recent results have strongly
relied on network connectivity. It is often required that agents
form a connected multi-hop communication network in which
the information exchange and sharing can be realized by
message relaying and forwarding between neighbors to ensure
Y. Mao, L. Dou, H. Fang, J. Chen and Tao Cai are with School of
Automation and Key Laboratory of Complex System Intelligent Control and
Decision (Ministry of Education), Beijing Institute of Technology, Beijing
100081, P.R.China, virtual123@126.com.
This work was supported by Projects of Major International (Regional)
Joint Research Program (No.61120106010), National Science Foundation for
Distinguished Young Scholars of China (No.60925011) and National Natural
Science Foundation of China (No.61175112).
reliable and efficient motion coordination within the under-
lying network. It is also demonstrated in [6] that network
connectivity fundamentally impacts the convergence rate, the
time-delay stability, and the robustness of consensus. As
connectedness is most commonly assumed rather than proved,
the problem of network connectivity maintenance is clearly
motivated.
In the recent literature, several works have taken connec-
tivity maintenance into consideration via both distributed and
centralized control strategies. Spanos and Murray introduced
a measure of local connectedness by utilizing the geometric
connectivity robustness [8]. In [9] and [10], the edges of graphs
were assigned to appropriate nonlinear weights to preserve
dynamic network connectivity. Potential fields for flocking
with connectivity maintenance were devised by Zavlanos et
al. in [12], the graph Laplacian and its spectral properties
were utilized to translate the connectivity condition into dif-
ferentiable constraints on motion of the individual agent. In
[13] and [17], market-based auctions combined with gossip
algorithms yielded hybrid control protocols for connectivity
preserving link addition and deletion. Bounded connectivity-
preserving coordinated control protocols were studied in [15]
and [16] to deal with rendezvous and formation control
problems respectively.
To the best of our knowledge, most of the above control
strategies can only guarantee local connectivity which main-
tain every single link initially created between any pair of
neighboring agents. Although connectivity maintenance can be
formally proven along this line, these methods often impose
so restrictive constraints on the system behavior that can be
avoided by considering global connectivity. Specifically, to
guarantee global connectivity, single links are allowed to be
deleted or added, as long as the overall graph is connected. If
necessary, redundant links can be removed and new ones can
be introduced. Moreover, maintaining all the edges of topology
helps to preserve connectivity but may hinder the realization
of flocking configuration. In many situations, the actual initial
connected topology may be highly intricate and can not be
spread out to have the same length for all the edges, which
is expected to form the α-Lattice flocking structure [4]. To
overcome these drawbacks, in this paper, a novel decentralized
flocking control strategy of agent group with global connec-
tivity maintenance is proposed. The contributions of this paper
are two folds:
(i) A novel inverse iteration algorithm is introduced for
estimation and computation of λ
2
as well as its cor-
responding eigenvector v
2
in a completely distributed
2013 American Control Conference (ACC)
Washington, DC, USA, June 17-19, 2013
978-1-4799-0178-4/$31.00 ©2013 AACC 976