Communications and Control Engineering

Yongduan Song
Yujuan Wang

Cooperative Control
of Nonlinear
Networked Systems
Infinite-time and Finite-time Design
Methods
Communications and Control Engineering

Series editors
Alberto Isidori, Roma, Italy
Jan H. van Schuppen, Amsterdam, The Netherlands
Eduardo D. Sontag, Boston, USA
Miroslav Krstic, La Jolla, USA
Communications and Control Engineering is a high-level academic monograph
series publishing research in control and systems theory, control engineering and
communications. It has worldwide distribution to engineers, researchers, educators
(several of the titles in this series find use as advanced textbooks although that is not
their primary purpose), and libraries.
The series reflects the major technological and mathematical advances that have a
great impact in the fields of communication and control. The range of areas to
which control and systems theory is applied is broadening rapidly with particular
growth being noticeable in the fields of finance and biologically-inspired control.
Books in this series generally pull together many related research threads in more
mature areas of the subject than the highly-specialised volumes of Lecture Notes in
Control and Information Sciences. This series’s mathematical and control-theoretic
emphasis is complemented by Advances in Industrial Control which provides a
much more applied, engineering-oriented outlook.
Publishing Ethics: Researchers should conduct their research from research
proposal to publication in line with best practices and codes of conduct of relevant
professional bodies and/or national and international regulatory bodies. For more
details on individual ethics matters please see:
https://www.springer.com/gp/authors-editors/journal-author/journal-author
helpdesk/publishing-ethics/14214

More information about this series at http://www.springer.com/series/61

Yongduan Song Yujuan Wang
•

Cooperative Control
of Nonlinear Networked
Systems
Infinite-time and Finite-time Design Methods

123
Yongduan Song Yujuan Wang
School of Automation School of Automation
Chongqing University Chongqing University
Chongqing, China Chongqing, China

ISSN 0178-5354 ISSN 2197-7119 (electronic)

Communications and Control Engineering
ISBN 978-3-030-04971-3 ISBN 978-3-030-04972-0 (eBook)
https://doi.org/10.1007/978-3-030-04972-0

Library of Congress Control Number: 2018962782

© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family for all the love, support
and understanding.
Yongduan Song

To my family, for all the support, love and

kindness.
Yujuan Wang
Preface

Motivation. Multiple networked dynamical systems, involving non-parametric or

non-vanishing uncertainties under certain communication topologies, have the
potential for many important applications, making it important and interesting to
control such systems. Due to the requirement for a powerful central station with
global information flow, the traditional centralized control has not been the ideal
control framework for the multiple networked systems under limited communica-
tion condition. In this regard, the distributed control, dedicated to achieving a global
common objective by using only local communication in the absence of central
coordination, is more promising due to the advantages such as limited
communication/sensing range, cooperative fashion, flexibility, and scalability. On
the other hand, adaptive control is known as one of the most effective ways to deal
with parametric uncertainties by employing online parameter estimators. As such, it
is necessary to investigate the cooperative adaptive control of multiple networked
nonlinear systems involving non-parametric or non-vanishing uncertainties.
Technical challenge in cooperative control of multiple networked nonlinear
systems with non-parametric/non-vanishing uncertainties lies in two aspects:
(1) how to realize the global cooperative objective in a distributive manner and
(2) how to effectively compensate the non-parametric/non-vanishing uncertainties
and meanwhile achieve the cooperative control objective with high precision and
high rate of error convergence. Although a rich collection of cooperative control
results are available in the literature for nonlinear multi-agent systems (MASs),
most existing works appear to only for the nonlinear MAS with linearly parame-
terized nonlinearities or achieve the so-called cooperative uniformly ultimately
bounded (CUUB) stability result. This is partly due to the constraint of the local
information transformation allowed in the communication network and partly due
to the extra residual terms produced by the non-vanishing uncertainties that cannot
be canceled in the Lyapunov stability analysis (for the MAS with non-vanishing
uncertainties).
Hence, it is crucial to develop control tools in the continuum domain, for the
distributed cooperative control of multiple networked nonlinear systems subject to
non-parametric/non-vanishing uncertainties to achieve a global cooperative

vii
viii Preface

objective with high precision and high rate of error convergence under relatively
weak communication topology condition.
What does the book cover? This book is concerned with the distributed cooper-
ative control of multiple networked nonlinear systems in the presence of unknown
non-parametric nonlinearities and non-vanishing uncertainties under a certain
communication condition. The technical coverage extends to
• Stability analysis such as input-to-state stability (ISS), cooperative uniformly
ultimately bounded (CUUB), globally exponentially stability, finite-time sta-
bility, and finite-time cooperative uniformly ultimately bounded (FT-CUUB),
etc., to analyze and synthesize the multiple networked nonlinear systems;
• Control design techniques and tools including matrix analysis tool (to develop
the distributed controller), core function technique and virtual parameter esti-
mation error technique (to construct the adaptive part of the Lyapunov function
candidate to deal with the non-parametric uncertainties), and Young’s inequality
(to establish various inequalities for the stability analysis), etc.;
• Various control schemes ranging from regular state feedback-based cooperative
adaptive control (to achieve CUUB), fractional state feedback-based distributed
finite-time control (to achieve finite-time stability), fractional state feedback-
based distributed finite-time adaptive control (to achieve FT-CUUB), and so on.
The infinite-time stability theory and finite-time stability theory allow for the
establishment of both qualitative and quantitative properties of the closed-loop
dynamical systems, and the adaptive control theory plays an important role in
coping with system uncertainties by employing online parameter estimators in
numerous dynamical systems. The required communication topologies can be either
undirected or directed, and the convergence time can be endowed with either
infinite time or finite time.
Technical contents of the book are divided into three parts. Part I consists of
Chaps. 1 and 2. Chapter 1 provides an overview of the existing results on coop-
erative control of multiple networked systems from the two aspects: infinite-time
control and finite-time control. Chapter 2 contains a review of algebraic graph
theory, certain matrix analysis theory associated with the graph, and stability
analysis theory and design tools.
Part II, composed of Chaps. 3–5, deals with the infinite-time cooperative lead-
erless consensus control of multiple networked nonlinear systems subject to
non-parametric/non-vanishing uncertainties under some fixed and local communi-
cation conditions. In Chap. 3, we present some key technique methods, including
the virtual parameter estimation error method and the core function technique, to
handle the inherent unknown time-varying control gain and the unknown
non-parametric uncertainties, and also introduce the construction of the distributed
part of the Lyapunov function (it is related to the graph topology) that is important
for the Lyapunov stability analysis and distributed controller design. Chapter 4
shows how to design the distributed adaptive controller for the networked MAS
with non-parametric/non-vanishing uncertainties from first-order to high-order
Preface ix

dynamics under the fixed and undirected topology, and also gives the Lyapunov
stability analysis. In Chap. 5, we study the cooperative leaderless consensus control
problem for networked MAS with first-order and second-order dynamics under the
directed topology condition.
In Part III of the book, which is comprised of Chaps. 6–9, we study the
finite-time adaptive consensus control for multiple networked nonlinear systems
subject to non-parametric/non-vanishing uncertainties linked to each other by a
local communication graph. The idea of fractional state feedback-based distributed
finite-time adaptive control is introduced in Chap. 6. We present some key
inequalities related to the fractional adaptive updated law and some useful lemmas
related to the directed topology condition, all of which are important for the
finite-time stability analysis and finite-time leaderless consensus controller design.
Chapter 7 shows the finite-time controller design and stability analysis for the
networked MAS with non-parametric/non-vanishing uncertainties with second-
order dynamics under the fixed and undirected topology. In Chap. 8, we extend the
results obtained in Chap. 7 to the directed topology. Chapter 8 studies the
finite-time cooperative leaderless consensus control problem for networked MAS
with high-order dynamics under the directed topology condition.
Whom is the book for? The book presents various solutions to cooperative control
problems of multiple networked nonlinear systems on graphs. The researcher,
engineer, or student in cooperative adaptive control of nonlinear MAS will find a
wealth of examples that are solved and presented in full analytical and numerical
detail, graphically illustrated, and with the intuition that helps understand the
results. It should be noted that although the overall system with linearly parame-
terized nonlinearities is nonlinear, such nonlinearity is sufficiently benign and
globally tractable. This is unlike the nonlinearities one encounters in the nonlinear
MAS with non-parametric or non-vanishing uncertainties. This book is intended
primarily as a bridge between the distributed control of MAS with linearly
parameterized nonlinearities and the one with non-parametric uncertainties. The
researchers and scientists in research institutes as well as academics in universities
working on nonlinear systems, adaptive control, and distributed control will find the
book filled with problems of a new kind.
We propose to use mathematical tools such as algebraic graph theory and certain
matrix analysis theory, and potential readers who already had some exposure to
basic systems theory and algebra graph theory can obtain a deeper understanding
of the roles of matrix operators as mathematical machineries for cooperative control
of MAS. This book can be used as an introductory material to fresh graduates (at
bachelor or master's level) joining the control workforce to work on distributed
control of multiple networked systems. The book can also be used as reference
textbooks for new postgraduate students who have the intention to work on this
arena of research. At the same time, many universities have established programs
and courses in this new field, with many cross-faculty and interdiscipline research
going on in this arena as well.
x Preface

We believe that there are a significant number of potential readers for our
proposed book among the professionals working in the engineering fields, which
include control engineers, mechanical engineers, and aerospace engineers who deal
with various mechanics problems with spatially varying coefficients even if control
is not their focus. This is because the methods employed are not from the generic
control catalog but are very structure-specific. Physicists would find the book of
interest for the same reason.
Acknowledgements. We sincerely appreciate David J. Hill, Miroslav Krstic, and
Frank L. Lewis for their valuable comments during the preparation of this book. We
also would like to thank Wei Ren, Yongcan Cao, Frank L. Lewis, Hongwei Zhang,
Kristian Hengster-Movric, Abhijit Das, and Zhihua Qu for their inspiring early
work on cooperative control of MAS. We are thankful to our editors Oliver Jackson
and Balaganesh Sukumar for their interests in our project and their professionalism.
In addition, we acknowledge IEEE and Elsevier for granting us the permission to
reuse materials from our publications copyrighted by these publishers in this book.
Finally, we gratefully acknowledge the National Natural Science Foundation of
China and Ministry of Science and Technology of China, and the Institute of Smart
Engineering in Chongqing University for their support. We dedicate this book to
our families.

Chongqing, China Yongduan Song

January 2019 Yujuan Wang
Contents

Part I Overview and Preliminary

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ......... 3
1.1 Cooperative Control of Multi-agent Systems . . . . . . . ......... 3
1.2 Overview of Recent Research Results in Cooperative
Control of Multi-agent Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Non-finite Time Convergence . . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Increased Convergence Speed . . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 Finite-Time Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1 Algebraic Graph Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Matrix Analysis on Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Matrix and Algebra Theory Background . . . . . . . . . . . . . . . . . . . 28
2.4 Stability Analysis Theory on Cooperative Control Systems . . . . . . 30
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Part II Infinite-Time Cooperative Adaptive Control of Multiple

Networked Nonlinear Systems
3 Lyapunov Analysis for Cooperative Adaptive Consensus Under
Undirected Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1 Leaderless Consensus Formulation . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Core Function and Virtual Parameter Estimation Error . . . . . . . . . 39
3.3 Lyapunov Function Design Under Undirected Topology . . . . . . . . 42
3.3.1 The Distributed Part of Lyapunov Function . . . . . . . . . . . . 42
3.3.2 The Adaptive Part of Lyapunov Function . . . . . . . . . . . . . 44
3.4 Lyapunov Analysis for Adaptive Leaderless Consensus . . . . . . . . 44
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

xi
xii Contents

4 Cooperative Adaptive Consensus Control for Uncertain

Multi-agent Systems with nth-Order Dynamics Under
Undirected Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... 49
4.1 Cooperative Adaptive Control of First-Order Uncertain
Systems Under Undirected Topology . . . . . . . . . . . . . . . . . . . . . . 50
4.1.1 System Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.1.2 Controller Design and Stability Analysis . . . . . . . . . . . . . . 51
4.1.3 Simulation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Second-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Consensus Error Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.2 Controller Design and Stability Analysis . . . . . . . . . . . . . . 58
4.2.3 Simulation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3 High-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3.1 Consensus Error Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3.2 Controller Design and Stability Analysis . . . . . . . . . . . . . . 66
4.4 Discussion on Cooperative Adaptive Consensus of MAS
with Non-parametric Uncertainties . . . . . . . . . . . . . . . . . ....... 69
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... 70
5 Cooperative Adaptive Consensus for Multi-agent Systems
Under Directed Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 73
5.1 Graph Laplacian Potential Under Directed Topology . . . . . . . . .. 74
5.2 Leaderless Consensus of First-Order Systems Under Directed
Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 77
5.2.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . .. 77
5.2.2 Controller Design and Stability Analysis for First-Order
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 78
5.3 Leaderless Consensus of Second-Order Uncertain Systems
Under Directed Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 79
5.3.1 Problem Formulation—Second-Order Uncertain Systems
Under Directed Topology . . . . . . . . . . . . . . . . . . . . . . . .. 80
5.3.2 Leaderless Consensus Control Design and Stability
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 81
5.4 Simulation Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 90
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 94

Part III Finite-Time Cooperative Adaptive Control for Multiple

Networked Nonlinear Systems
6 Finite-Time Leaderless Consensus Control for Systems
with First-Order Uncertain Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 97
6.1 Preliminaries for the Finite-Time Control . . . . . . . . . . . . . . . . . . . 98
6.2 Finite-Time Leaderless Consensus Control Under Undirected
Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Contents xiii

6.2.1 Problem Formulation of Finite-Time Leaderless

Consensus for First-Order Uncertain Systems . . . . . . . . . . 100
6.2.2 Finite-Time Leaderless Consensus Control Design . . . . . . . 101
6.2.3 Finite-Time Leaderless Consensus Stability Analysis . . . . . 102
6.2.4 Comparison with Regular State Feedback-Based
Infinite-Time Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.3 Finite-Time Leaderless Consensus Control Under Directed
Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3.2 Lyapunov Function Design Under Directed Topology . . . . 112
6.3.3 Finite-Time Leaderless Consensus Control Design
and Stability Analysis Under Directed Topology . . . . . . . . 113
6.4 Simulation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7 Finite-Time Consensus for Systems with Second-Order Uncertain
Dynamics Under Undirected Topology . . . . . . . . . . . . . . . . . . . . . . . 121
7.1 Formulation of Finite-Time Consensus for Second-Order
Uncertain Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.2 Useful Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.3 Distributed Adaptive Finite-Time Control for Second-Order
Uncertain Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.3.1 Finite-Time Consensus Control Design . . . . . . . . . . . . . . . 125
7.3.2 Finite-Time Consensus Stability Analysis . . . . . . . . . . . . . 126
7.4 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8 Finite-Time Consensus for Systems with Second-Order
Uncertain Dynamics Under Directed Topology . . . . . . . . . . . . . . . . . 143
8.1 Formulation of Finite-Time Consensus for Second-Order
Uncertain Systems Under Directed Topology . . . . . . . . . . . . . . . . 144
8.1.1 Control Problem Formulation Under Directed
Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.2.1 Useful Lemmas Related to Finite-Time Control . . . . . . . . . 145
8.2.2 Lyapunov Function Design Under Directed Topology . . . . 148
8.3 Distributed Adaptive Finite-Time Control Design
and Stability Analysis Under Directed Topology . . . . . . . . . . . . . 149
8.3.1 Comparison with Regular State Feedback-Based
Infinite-Time Method Under Directed Topology . . . . . . . . 160
xiv Contents

8.4 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

8.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9 Finite-Time Leaderless Consensus Control for Systems
with High-Order Uncertain Dynamics . . . . . . . . . . . . . . . . . . . . . . . 171
9.1 Finite-Time Consensus Formulation for High-Order
Uncertain Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
9.2 Useful Lemmas and Intermediate Results . . . . . . . . . . . . . . . . . . . 173
9.3 Inductive Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.4 The Finite-Time Stability Analysis for the Uncertain
High-Order MAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
9.5 Comparison with Regular State Feedback-Based Infinite-Time
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
9.6 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Abbreviations and Notation

Abbreviations
CUUB Cooperative uniformly ultimately bounded
LMI Linear matrix inequality
LTI Linear time invariant
MIMO Multiple input, multiple output
NMAS Networked multi-agent systems
SISO Single input, single output
TSM Terminal sliding mode

Notation
R Field of real numbers
R Summation
jaj The absolute of a scalar a
kX k The norm of a vector X
max Maximum
min Minimum
sup Supremum, the least upper bound
inf Infimum, the greatest lower bound
8 For all
2 Belongs to
! Tends to
\ð [ Þ Less (greater) than
ðÞ Less (greater) than or equal to
 ðÞ Much less (greater) than
 Kronecker product
RN The N-dimensional Euclidean space
IN The N-dimensional identity matrix
1N ð0N Þ The N-dimensional vector with each entry being 1(0)
y_ The first derivative of y with respect to time

xv
xvi Abbreviations and Notation

€y The second derivative of y with respect to time

yðiÞ The ith derivative of y with respect to time
w.r.t. With respect to
satðÞ The saturation function
sgnðÞ The signum function
diagfa1 ;    ; aN g A diagonal matrix with diagonal elements a1 to aN
diagfA1 ;    ; AN g A diagonal matrix with diagonal elements A1 to AN , here Ai
(i ¼ 1;    ; N) denotes matrix
P[0 A positive definite matrix P
P0 A positive semi-definite matrix P
AT ðX T Þ The transpose of matrix A (of vector X)
ðAÞðrðAÞÞ
r The maximum (minimum) singular value of matrix A
kmax ðPÞðkmin ðPÞÞ The maximum (minimum) eigenvalue of symmetric matrix P
f 1 ðÞ The inverse of function f
 Part I
Overview and Preliminary
Chapter 1
Introduction

This chapter introduces the cooperative control of multi-agent systems (MASs) and
overviews recent research results in distributed cooperative control of networked
MAS. The recent research results in distributed cooperative control of MAS are
roughly categorized as leaderless consensus, leader–follower consensus, distributed
formation control, and distributed containment control. A short discussion is given
to propose several future research directions and problems that deserve further inves-
tigation.

1.1 Cooperative Control of Multi-agent Systems

In the past two decades, the control of multiple dynamical systems interconnected
by a communication network has attracted extensive attention among scientific com-
munities due to the benefits obtained by replacing a solo complicated system with
several simple systems. For the control of multiple networked dynamical systems,
two methods are commonly used: the centralized control method and the distributed
control method. As shown in Fig. 1.1, the centralized control method relies on the
assumption that each subsystem of the group has the ability to communicate to a cen-
tral station or share information via a fully connected network. Essentially, the cen-
tralized control method is a direct extension of the traditional single-system control
methodology. As a result, the centralized control method may result in a catastrophic
failure of the overall system due to its single point of failure. Also, real-world com-
munication topologies are not usually fully connected. Therefore, cooperative con-
trol in the presence of real-world communication constraints becomes a significant
challenge. Different from the traditional centralized control method, the distributed
control method, as shown in Fig. 1.2, does not require the existence of a central
station. The distributed control is dedicated to achieving a global common objective
by using only local communication in the absence of central coordination and thus is
more promising due to the advantages such as limited communication/sensing range,
© Springer Nature Switzerland AG 2019 3
Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_1
4 1 Introduction

Fig. 1.1 Centralized control

Fig. 1.2 Distributed control

cooperative fashion, flexibility, and scalability, although it is far more complex than
the centralized one.
By cooperative control system on networks, it is referred to a system that consists
of a group of autonomous subsystems with sensing or communication capabilities,
for example, vehicles with communication devices and robots with camera sensors.
Of fundamental concern for the networked cooperative system is the study of the
interactions and collective behaviors of each subsystem under the influence of the
information flow allowed in the communication network. This communication net-
work can be modeled as a graph with directed edges or links corresponding to the
allowed flow of information among the subsystems, and the subsystems are mod-
eled as the nodes (sometimes called agents) in the graph. The goal of the networked
cooperative system is to achieve a common group behavior by using only local infor-
mation available to each agent from sensing or communication devices. Such local
information may include relative configuration and motion obtained from sensing
or communication among agents, agents’ sensor measurements, and so on. Thus,
a cooperative system has four basic elements: group objective, agents, information
topology, and control protocols governing the motion of agents. Specifically, the
fundamental problem in the networked cooperative control system is the design of
1.1 Cooperative Control of Multi-agent Systems 5

distributed protocols in the sense that the control law of each agent is only allowed to
depend on local information from its neighbors in the networks to guarantee that all
the agents achieve a common group objective or that all the states of agents achieve
a common value.
Cooperative control of networked multiple dynamical systems poses significant
theoretical and practical challenges. First, the research objective is to develop a sys-
tem of subsystems rather than a single system. Second, the communication bandwidth
and connectivity of the team are often limited, and the information exchange among
subsystems may be unreliable. It is also difficult to decide what to communicate and
when and with whom the communication takes place. Third, arbitration between
team goals and individual goals needs to be negotiated. Fourth, the computational
resources of each individual subsystem will often be limited.
In general, the control problems of the cooperative systems can be roughly cate-
gorized based on the following directions:

1. Consensus/agreement/synchronization/rendezvous. Distributed consensus is a

study dedicated to ensuring an agreement between states or output variables
among networked systems, in which distributed controller is designed for each
agent such that the states or output of all agents is eventually driven to an unpre-
scribed common value. This common value may be a constant or may be a time-
varying function which is generally a function of all the agents’ initial states [1].
In many cases, the four words can be used without discrimination.
2. Leader–follower consensus. For the leader–follower consensus problem, a leader
agent acting as a command generator is involved with all the other agents being
follower agents. The leader agent generates the desired reference trajectory and
does not accept information from follower agents, and all the follower agents
attempt to follow the leader agent’s trajectory. The control objective in leader–
follower consensus is to design distributed controller for each follower such that
each follower can synchronize to the leader agent. This problem is also known as
coordinated tracking, synchronization to a leader [2], model reference consensus
[3], leader-following control [4], or pinning control [5].
3. Distributed formation control. Distributed formation control problem refers to
the scenario that all the agents in the cooperative system are required to operate
according to the prescribed trajectory to form a certain geometrical configuration
through local interaction with or without a group reference. The control objective
in the formation control is to stabilize the relative distances/position between the
agents to achieve a pre-specified desired value.
4. Distributed containment control. The distributed containment control problem
arises when there exist multiple leaders. It refers to the scenario that all the
follower agents are forced to enter the convex hull formed by the leader agents
through local interaction among the neighbor follower agents only.

How to achieve cooperative consensus with the least possible topological require-
ment and how to achieve it in a timely and distributive manner are two common
challenges in distributed control of networked cooperative systems. In this sense,
6 1 Introduction

the convergence rate is one of the most important performance indicators for the
cooperative control of networked multi-agent systems. So far, most of the existing
cooperative control results are achieving asymptotical convergence, meaning that the
convergence rate is at best exponential with infinite settling time. That is, it needs
infinite time to achieve cooperative consensus or formation or containment or other
cooperative objectives. It is indicated in [6] that the second smallest eigenvalue of
the Laplacian matrix quantifies the convergence rate of the networked multi-agent
systems. Several researchers have tried to find better interaction graph to get a larger
second smallest eigenvalue such that a better convergence rate can be obtained. To
maximize the second smallest eigenvalue of the interaction graph, the researchers in
[7] try to find the best vertex positional configuration. By utilizing the semi-definite
convex programming, the researchers in [8] design the weights among the agents
to increase the convergence rate. It is worth noting that all these efforts are to find
proper interaction graphs, but not to find proper control schemes to achieve higher
performance. In practice, it is often required that the cooperative control objective
can be reached in finite time. In addition to bearing the benefit of faster rate of conver-
gence, finite-time control systems exhibit the feature of better disturbance rejection
and robustness against uncertainties as shown in [9], making the finite-time coopera-
tive control problem one of the most active research topics. Thus, several researchers
invoke finite-time control schemes for networked multi-agent systems to guarantee
that the cooperative control objective is reached within finite time.

1.2 Overview of Recent Research Results in Cooperative

Control of Multi-agent Systems

In this section, we overview recent research results in distributed cooperative control

of MAS. According to different convergence rates of the cooperative control systems,
we roughly categorize the recent research results into the following three classes:

1. Non-finite time convergence: The non-finite time convergence means that the
convergence rate of the cooperative control systems is at best exponential with
infinite settling time. That is, it needs infinite time to achieve cooperative con-
sensus or formation or containment or other cooperative objectives. The control
results corresponding to the non-finite time convergence include the asymptotical
stability, exponential stability, ultimately uniformly boundedness (UUB).
2. Increased convergence speed: This is corresponding to the case to choose proper
interaction graphs to increase the convergence rate but not to find proper control
schemes to achieve high performance.
3. Finite-time convergence: The finite-time convergence means that the control
objectives of the cooperative control systems are realized in finite time. The finite-
time control systems not only possess faster convergence rate but also possess
better disturbance rejection and robustness against uncertainties.
1.2 Overview of Recent Research Results in Cooperative Control … 7

1.2.1 Non-finite Time Convergence

In this subsection, we overview recent research results in distributed cooperative

control of networked MAS with non-finite time convergence.
The research on cooperative control of MAS has gained considerable attention
from scientific communities. This is partly due to the broad potential applications
of the cooperative control of MAS in various areas, such as rendezvous [10], flock-
ing [11], synchronization [12, 13], and formation control [14–16]. Consensus is
known as one of the most important issues in cooperative control of MAS, which
refers to all agents reaching an agreement on certain variables of interest. Vicsek
et al. proposed in [17] in 1995 a simple but compelling discrete-time model of
n autonomous agents (points or particles) all moving in the plane with the same
speed but with different headings, in which each agent’s heading is updated by
using a local rule based on the average of its own heading plus the headings of
its “neighbors.” This work provided study base for the cooperative consensus stud-
ies. Vicsek et al. in [17] provided simulation results to demonstrate that the nearest
neighbor rule they are studying can cause all agents to eventually move in the same
direction despite the absence of centralized coordination and despite the fact that
each agent’s set of nearest neighbors change with time as the system evolves, on
basis of which Jadbabaie et al. further provided a theoretical explanation for the
observed behavior in [18]. In [6], Olfati-Saber and Murray addressed consensus
problems for networks with directed information flow and provided analytical tools
relying on control theory, algebraic graph theory, and matrix theory, which pro-
vides a theoretical framework to analyze and study the consensus problem of multi-
agent systems. In [12], Moreau studied a simple but compelling model of networked
agents interacting via time-dependent communication links and represented neces-
sary and/or sufficient conditions for the convergence of the individual agents’ states
to a common value. Ren et al. showed in [19, 20] that the consensus under dynam-
ically changing interaction topologies can be achieved asymptotically if the union
of the directed interaction graphs has a spanning tree frequently enough as the sys-
tem evolves, which extends sufficient conditions for the cooperative control systems
to achieve consensus to more general case. In [21], Cortes addressed the problem
of designing (continuous-time) coordination algorithms that make a networked sys-
tem asymptotically agree upon the value of a desired function of the initial state
of the individual agents. Ren and Atkins proposed second-order consensus proto-
cols that take motions of the information states and their derivatives into account,
extending first-order protocols from the literature, and also showed that unlike the
first-order case, having a (directed) spanning tree is a necessary rather than a sufficient
condition for consensus seeking with second-order dynamics in [22, 23]. Yu et al.
studied some necessary and sufficient conditions for second-order consensus: Both
the real and imaginary parts of the eigenvalues of the Laplacian matrix associated
with the interaction graph topology play key roles in reaching consensus in [24]. In
[25], Abdessameud et al. dealt with consensus strategy design for double-integrator
dynamics without velocity measurements and with input constraints. The works in
[26–29] studied the leaderless consensus problem for high-order linear time-invariant
8 1 Introduction

(LTI) multi-agent systems with asymptotical convergence. Yu et al. addressed the

leaderless consensus problem for second-order multi-agent systems with nonlinear
dynamics under directed topologies in [30]. In [31], Wang and Song investigated the
leaderless consensus for a class of second-order multi-agent systems with unknown
time-varying control gains and non-parametric uncertainties as well as the unknown
actuator failures.
It is noted that the aforementioned works are all about the leaderless consensus,
and the final consensus value is some functions that is related to the initial states
of all agents in cooperative system, i.e., χ -consensus [21], including the average
consensus [32–35] as a special case. For more detailed results on consensus control,
the authors can be referred to [36, 37].
In some practical applications, there exists one or even multiple leaders that pro-
vide/provides group reference states or external control instructions for networked
multi-agent systems. Consensus with one leader is often named as consensus track-
ing or coordinated tracking. In [38], Ren studied the multi-vehicle consensus with
a time-varying reference state, in which the cooperative tracking problem for the
first-order multi-agent systems with one time-varying leader has been solved. Qin
et al. considered the consensus problem of multiple second-order vehicles with a
time-varying reference signal under directed communication topology in [39]. In
[40], Cao et al. addressed the distributed discrete-time consensus tracking with a
time-varying reference trajectory and limited communication. Gu et al. investigated
a leader–follower flocking system that can track the desired trajectory led by group
leaders in [41]. By using the polynomial filtering algorithms, a control method to
accelerate the convergence rate for given network matrices was proposed in [42].
The work in [43] addressed the stochastic bounded consensus tracking problems
of leader–follower multi-agent systems, in which the information measured at the
sampling instants from its neighbors or a virtual leader with a time-varying refer-
ence trajectory is available, the measurements are corrupted by random noises, and
the signal sampling process induces a small sampling delay. The second-order con-
sensus of leader–follower multi-agent systems with jointly connected topologies and
time-varying delays was studied in [44]. A distributed consensus protocol was devel-
oped for a class of homogeneous time-varying nonlinear multi-agent systems in the
work [45], in which the agent dynamics were supposed to be in the strict feedback
form and satisfy Lipschitz conditions with time-varying gains. In the work [46], a
coordinated consensus problem with an active leader under variable interconnection
topology was considered, in which the leader’s state not only keeps changing but
also may not be measured. The work in [4] considered a leader-following problem
for multi-agent systems with a switching interconnection topology, in which dis-
tributed observers were designed for second-order follower agents under a common
assumption that the active leader’s velocity cannot be measured in real time. The con-
sensus tracking problems for high-order LTI multi-agent systems under both fixed
and switching interaction topologies were coped with by Ni and Cheng in [47]. The
work [48] solved a distributed coordinated tracking problem via a variable structure
approach. The work [49] investigated the coordinated tracking problem of multiple
networked dynamic systems with both physical connections and virtual connections
1.2 Overview of Recent Research Results in Cooperative Control … 9

in which unknown unidentical nonlinearities and time-varying yet undetectable actu-

ation faults and varying actuation authorities are involved.
If there exist multiple leaders, containment problems requiring that the states of
followers converge to the convex hull formed by the states of leaders arise. Cao
and Ren studied the containment control problem of a group of mobile autonomous
agents with single-integrator dynamics in the presence of both multiple stationary and
dynamic leaders under fixed and switching directed network topologies in [50], and
studied containment control problem for double-integrator dynamics in the presence
of both stationary and dynamic leaders in [51], respectively. Mei et al. [52] inves-
tigated the distributed containment problem for multiple Lagrangian systems in the
presence of parametric uncertainties with multiple stationary leaders under a directed
graph without using the neighbors’ velocity information in the absence of communi-
cation. The containment problems for low-order multi-agent systems have also been
considered in [54–59]. Li et al. [53] discussed the containment control problems for
both continuous-time and discrete-time high-order LTI multi-agent systems under
directed communication topologies by using a dynamic output protocol. Dong et al.
[60] investigated the containment control problem for high-order LTI singular multi-
agent systems with time delays. By classifying the agents into boundary agents and
internal agents, Liu et al. [61] proposed a criterion for the states of internal agents to
converge to a convex combination of the states of the boundary agents.

1.2.2 Increased Convergence Speed

To achieve a faster convergence rate in consensus control of networked MAS, two

main methods have been proposed: The first one is to use the physical communi-
cation network as an information flow graph but with the optimized weights for all
information flow edges, and the second one is to introduce a few shortcut (non-local)
multi-hop edges to the information flow network without physically changing or
adding any edges in the underlying communication network. The most interesting
and famous one in the later category is small-world networks that can dramatically
enhance the algebraic connectivity of regular complex networks. As shown in [6], the
second smallest eigenvalue of the Laplacian matrix quantifies the convergence rate.
Several researchers have tried to find better interaction graph to get a larger second
smallest eigenvalue such as to get better convergence rate. On the basis of the work
[6], Olfati-Saber further demonstrated in [62] that without adding new links or nodes
to the network it was still possible to dramatically increase the algebraic connectiv-
ity of a regular complex network by 1000 times or more. However, Nosrati et al.
showed the fact in [63] that, “the small-world network construction has a negligible
effect on the time-delay robustness of the consensus protocol over a initial regular
network,” is not true. Xiao et al. [64] showed how problem structure can be exploited
to speed up interior-point methods for solving the fastest distributed linear iteration
problem for networks with up to a thousand or so edges. Kim et al. [65] posed the
problem of finding the best vertex positional configuration to maximize the second
10 1 Introduction

smallest eigenvalue of the associated Laplacian. In order to achieve a fast consensus

seeking, Jin and Murray [66] proposed the multi-hop relay protocols, where each
agent can expand its knowledge by employing multi-hop paths in the network, and
demonstrated that multi-hop relay protocols can enlarge the algebraic connectivity
without physically changing the network topology. Yang et al. [67] analyzed the rela-
tionship between the second smallest eigenvalue of the associated Laplacian and the
convergence rate of the consensus control algorithm, and found that the convergence
rate of the consensus control algorithm on a completely regular lattice can be greatly
enhanced by just randomly rewiring a very small number of links in the network.

1.2.3 Finite-Time Convergence

It is worth noting that all these solutions mentioned in Sect. 1.2.2 are to choose
proper interaction graphs to achieve a faster convergence, but not to find proper
control schemes to achieve a higher performance. It is often required in practice
that the cooperative control objective is reached in finite time. In addition to bear-
ing the benefit of faster rate of convergence, finite-time control systems exhibit the
feature of better disturbance rejection and robustness against uncertainties as shown
in [9], enticing increasing research attention from control community. Thus, sev-
eral researchers seek finite-time control algorithms to guarantee that the cooperative
control objective is reached within finite time.
Weiss and Infante [68] gave the definition of finite-time stability of systems in 1965
and extended the finite-time stability theory to the non-autonomous systems under
the influence of external forces in [69], which were the embryo of finite-time stability
theory. However, these results obtained in [68, 69] are only about the analysis on the
finite-time stability, and there are not the analysis on how to design the finite-time
controller for systems. In general, according to the continuousness of the control input
signals, the finite-time control methods in the existing literatures are classified into
three categories: discontinuous finite-time control, continuous but non-smooth finite-
time control, and continuous and smooth finite-time control. The discontinuous finite-
time control methods mainly include signum function feedback-based control and
terminal sliding mode (TSM)-based control; the continuous but non-smooth finite-
time control methods mainly include single fractional state feedback-based control,
homogeneous finite-time control, and adding power integrator-based control; the
continuous and smooth finite-time control methods mainly include the time-varying
gain-based regular state feedback control proposed by Song and Wang in [70, 71].
Signum function feedback-based control: The signum function feedback-based
control is a class of discontinuous control method based on the signs of system states.
Take the simple first-order linear system as an example,

ẋ = u, (1.1)
1.2 Overview of Recent Research Results in Cooperative Control … 11

where x and u denote the system state and control input, respectively. The signum
function feedback-based control is of the basic form,

u = −csgn(x) (1.2)

where c > 0 is a freely chosen finite constant parameter. This control is based on the
signum function, in which infinite bandwidth is required and the control action is
discontinuous. Under such discontinuous control, the system states converge to the
origin in a finite time T , satisfying

T ≤ x(t0 )/c. (1.3)

The signum function-based control can ensure the finite-time stability of systems
and also possesses better disturbance rejection; however, the discontinuous terms in
control input might cause the notorious chatting phenomenon during system oper-
ation. Cortes [72] discussed the application of the discontinuous control method
based on signum function to consensus problems in first-order multi-agent systems
and showed how the proposed non-smooth gradient flows achieve consensus in finite
time. Chen et al. [73] investigated the finite-time consensus problem for first-order lin-
ear multi-agent systems under specific directed topology using the signum protocol.
Li and Qu [74] extended the signum function-based control to first-order nonlinear
multi-agent systems under the general setting of directed and switching topologies to
achieve finite-time consensus. Upon using the signum-based control method, Zhang
et al. [75] studied the finite-time consensus tracking problem of networked nonlinear
multi-agent systems subject to unknown external disturbances under the directed
topology that satisfies detailed balanced condition. Franceschelli et al. [76] proposed
a decentralized consensus algorithm for a network of continuous-time integrators
subject to unknown-but-bounded time-varying disturbances by using the discontinu-
ous local interaction rule, in which it was proven that the agents achieve an approxi-
mated consensus condition by attenuating the destabilizing effect of the disturbances
after a finite transient time. Wen et al. [77] proposed a class of discontinuous con-
trol protocols that only utilize the position information of each agent by using the
signum-based control method and solved the finite-time coordinated tracking con-
trol problem. Meng and Lin [78] studied the finite-time consensus tracking control
problem for multiple double-integrator networked systems involving a time-varying
velocity leader and unknown bounded external disturbances, in which a discontinu-
ous consensus tracking control protocol upon using a distributed finite-time observer
is proposed. Based on signum function control method, Fu and Wang [79] addressed
the finite-time consensus tracking problem for high-order integrator networked sys-
tems with bounded matched uncertainties.
Terminal sliding mode (TSM) control: Sliding mode control is a special kind of
variable structure control proposed by Emelyanov and Taran in [80]. Sliding mode
control is known to drive the system states to the sliding surface in finite time, and
further, it is indicated that the sliding mode control has strong robustness and thus
provides an effective control method for the nonlinear control systems. Yu and Man
12 1 Introduction

introduced the terminal sliding mode (TSM) control method in [81, 82], in which
the terminal sliding mode (TSM) was defined by the first-order dynamics

ẋ + βx q/ p = 0 (1.4)

where x is a scalar variable and β > 0 and p > q > 0 are odd integers such that,
for any real number x, x q/ p returns a real number. For system (1.4), the trajectories
converge to an equilibrium state in a finite time T , satisfying
p
T = |x(0)|( p−q)/ p . (1.5)
β( p − q)

Further, the dynamics (1.4) can be extended to higher-dimensional terminal sliding

mode case by introducing the hierarchical sliding mode structure as
q / p1
s1 =ṡ0 + β1 s0 1
q /p
s2 =ṡ1 + β2 s1 2 2
.. .. ..
. . .
q / pn−1
sn−1 =ṡn−2 + βn−1 sn−2
n−1
(1.6)

where s0 = x, βi > 0, and pi > qi are positive odd integers, and the finite conver-
gence time T s for system (1.6) satisfies


n−1
pn−i
T s = t1s + |sn−i−1 (tis )|( pn−i −qn−i )/ pn−i (1.7)
i=1
βn−i ( pn−i − qn−i )

where tis is the reaching time of the terminal sliding mode sn−i = 0 for i =
1, . . . , n − 1. The TSM control can ensure the finite-time convergence of systems
and is robust to input disturbances [83]. However, this method may cause the singu-
larity problem around the equilibrium. Khoo et al. [84] studied the integral terminal
sliding mode control of networked multi-robot systems, in which an integral ter-
minal sliding mode surface for a class of first-order systems was proposed and the
finite-time coordinated consensus tracking of multi-robot networked systems can
be achieved on this integral terminal sliding mode surface. Khoo et al. [85] further
studied the finite-time consensus tracking control for leader–follower multi-agent
systems dominated by second-order dynamics and proved that the global finite-time
consensus tracking can be achieved on the terminal sliding mode surface by switch-
ing control law. Zou and Kumar [86] applied the terminal sliding mode control to
the attitude coordination of a group of spacecraft in the presence of external dis-
turbances, in which a fast terminal sliding manifold was presented and a robust
control term based on the hyperbolic tangent function was employed to suppress the
bounded external disturbances. By using sliding mode control theory, Chen et al.
[87] investigated the finite-time consensus tracking problem for a class of networked
1.2 Overview of Recent Research Results in Cooperative Control … 13

Euler–Lagrange systems with a leader–follower structure, where the leader has an

active dynamics and only a subset of followers have access to the leader system.
Ghasemi and Nersesov [88] considered the finite-time coordination for a group of
agents described by fully actuated Euler–Lagrange dynamics along with a leader
agent by using a sliding mode control approach. With the construction of a nonlinear
tracking control protocol via non-singular terminal sliding mode scheme, He et al.
[89] investigated the finite-time cooperative attitude synchronization and tracking
control for multiple rigid spacecraft with a time-varying leader whose attitude is
represented by modified Rodriguez parameters. By introducing decentralized slid-
ing mode estimators, Cao et al. [90] proposed a framework to achieve finite-time
decentralized formation tracking of multiple autonomous vehicles. Upon using the
second-order twisting sliding mode control system, Kamble et al. studied the leader–
follower finite-time formation control of networked multi-agent systems with input
disturbances in [91]. Li et al. [92] investigated the robust finite-time coordinated
tracking problem for a group of second-order multi-agent systems with nonlinear
dynamics under directed topology by using the fast terminal sliding mode control
method. Zhao and Hua constructed a nonlinear distributed consensus tracking proto-
col via non-singular terminal sliding mode scheme in [93] to address the finite-time
consensus tracking problem for second-order multi-agent systems composed of a
leader with bounded input signal and n followers with bounded disturbances. Yang
et al. [94] proposed a distributed robust control method for synchronized tracking of
networked Euler–Lagrange systems, in which a disturbance observer was introduced
to compensate for the low-passed-coupled uncertainties and a sliding mode control
term was employed to handle the uncertainties that the disturbance observer cannot
compensate for sufficiently.
The above-mentioned two control methods are discontinuous, and the discontin-
uous control terms would lead to chatting in control systems, which is not desired in
the actual control. Thus, it is highly desired to develop continuous finite-time control
methods for systems to avoid chatting.
Most of the existing works on continuous finite-time control are based on the
theory of finite-time stability proposed in [9, 95, 96]. Following the terminology of
[9, 95, 96], the global finite-time stability can be introduced in mathematical terms
as follows.

Definition 1.1 Consider the system

ẋ = f (x, t), f (0, t) = 0, x ∈ Rn , (1.8)

where f : U0 × R+ → Rn is continuous w. r. t. x on an open neighborhood U0 of the

origin x = 0. The equilibrium x = 0 of the system is said to be (locally) finite-time
stable if it is Lyapunov stable and finite-time convergent in a neighborhood U ⊆ U0
of the origin. By “finite-time convergence,” we mean that if there is a finite settling
time function T : U/{0} → (0, ∞), such that ∀x0 ∈ U , every solution x(t; 0, x0 ) of
system (1.8) with x0 as the initial condition is defined, and x(t; 0, x0 ) ∈ U/{0} for
14 1 Introduction

t ∈ [0, T (x0 )); limt→T (x0 ) x(t; 0, x0 ) = 0. The origin is said to be a globally finite-
time stable equilibrium if U = Rn .

Lemma 1.1 Consider system (1.8). Suppose that there exists a continuously dif-
ferentiable function V (x) : Û → R (Û ⊂ U0 ⊂ Rn ), finite real number c > 0 and
α ∈ (0, 1), such that
(i) V (x) is positive and definite on Û .
(ii) V̇ (x) + cV (x)α is negative semi-definite (along the trajectory) on Û .
Then, the origin of system (1.8) is finite-time stable. Moreover, the finite settling time
T satisfies

1
T ≤ V (x)1−α (1.9)
c(1 − α)

for all x in some open neighborhood of the origin.

Single fractional-order state feedback-based control. By employing the the-

ory of finite-time stability proposed in [95, 96], Wang and Xiao [97] discussed
finite-time state consensus problems for multi-agent systems with single-integrator
dynamics under undirected topology and directed topology that contains a span-
ning tree, respectively, and presented one framework for constructing effective dis-
tributed finite-time control algorithms based on fractional state feedback. Xiao et al.
[98] extended the finite-time consensus results obtained in [97] to finite-time for-
mation control for single-integrator multi-agent systems. On the basis of these two
works mentioned above, Xiao et al. [99] further proposed a general class of nonlin-
ear protocols in the form of continuous state feedbacks for finite-time consensus in
integrator-like networked dynamic agents with directional link failure, which, with
adjustable terms, were applicable in a wide range of situations, such as with input
saturation restrictions and with convergence rate constraints. Upon using the single
fractional state feedback-based method, Cao and Ren [100] analyzed the finite-time
convergence of a nonlinear but continuous consensus algorithm for first-order multi-
agent systems with unknown inherent nonlinear dynamics under directed switching
topology graph that had a directed spanning tree at each time interval. Zuo and Lin
[101, 102] constructed a class of global continuous time-invariant consensus proto-
cols based on single fractional-order state feedback for each single-integrator agent
in networked multi-agent systems under undirected topology and directed topol-
ogy, respectively, with the settling time being upper bounded for arbitrary initial
conditions.
Homogeneous finite-time control. The finite-time stability of homogeneous sys-
tems was proposed by Bhat in [103]. It is stated in [103] that a homogeneous system
is finite-time stable if and only if it is asymptotically stable and has a negative
degree of homogeneity, and a finite-time stable homogeneous system has a smooth
homogeneous Lyapunov function that satisfies a finite-time differential inequality.
Hong et al. [104–107] further deepened this theory and applied it to finite-time con-
trol of homogeneous systems. This approach is mainly based on the properties of
1.2 Overview of Recent Research Results in Cooperative Control … 15

homogeneous systems. In the following, we introduce some concepts about homo-

geneous systems which can be found, for example, in [103, 104, 108, 109].

Definition 1.2 Dilation Δ(r ε

1 ,...,rn )
is mapping dependent on dilation coefficients
(r1 , . . . , rn ), which assigns to every real number ε > 0 a diffeomorphism

Δ(r
ε
1 ,...,rn )
(x1 , . . . , xn ) = (εr1 x1 , . . . , εrn xn ) (1.10)

where x1 , . . . , xn are suitable coordinates on Rn and r1 , . . . , rn are positive real

numbers.
A scalar function ϕ(x1 , . . . , xn ) is called homogeneous of degree σ ∈ R w. r. t.
the dilation Δ(r ε
1 ,...,rn )
if there exists σ ∈ R such that

ϕ(Δ(r
ε
1 ,...,rn )
(x1 , . . . , xn )) = εσ ϕ(x1 , . . . , xn ). (1.11)

A vector field f (x) = [ f 1 (x), . . . , f n (x)]T is called homogenous of degree k ∈ R

w. r. t. the dilation Δ(r
ε
1 ,...,rn )
if there exists k ∈ R such that

f i (Δ(r
ε
1 ,...,rn )
(x1 , . . . , xn )) = εk+ri f i (x1 , . . . , xn ), i = 1, . . . , n. (1.12)

System ẋ = f (x) is called homogeneous if its vector field f is homogeneous.

Lemma 1.2 Suppose that system ẋ = f (x) is homogeneous of degree k < 0 w. r. t.

dilation Δ(r
ε
1 ,...,rn )
, in which f is continuous and x = 0 is its asymptotically stable
equilibrium. Then, the equilibrium of this system ẋ = f (x) is globally finite-time
stable.

Lemma 1.3 Suppose that system (1.8) is homogeneous of degree k < 0 w. r. t. dila-
tion Δ(r
ε
1 ,...,rn )
, in which f is continuous and x = 0 is its asymptotically stable equi-
librium. Then, the equilibrium of system (1.8) is globally finite-time stable. Moreover,
if f˜(x) is a continuous vector field defined on Rn such that

f i (εr1 x1 , . . . , εrn xn )
lim = 0, i = 1, . . . , n, f˜ = ( f˜1 , . . . , f˜n )T (1.13)
ε→0 εk+ri

uniformly for x = 1, then the zero solution of

ẋ = f (x) + f˜(x), f˜(0) = 0 (1.14)

is (locally) finite-time stable.

By merging the theories of semi-stability and finite-time stability, Hui et al.

[110] developed a rigorous framework for finite-time semi-stability, in which suf-
ficient and necessary conditions for finite-time semi-stability of homogeneous
systems were addressed by exploiting the fact that a homogeneous system is finite-
time semi-stable if and only if it is semi-stable and has a negative degree of
16 1 Introduction

homogeneity. Upon using the finite-time control technique and homogenous sys-
tems theory, Du et al. [111] addressed the finite-time synchronization problem for
a class of nonlinear leader–follower multi-agent systems with second-order dynam-
ics by constructing a finite-time state feedback controller and a finite-time observer.
With the aid of homogeneous theory, Zhang and Yang [112] studied the finite-time
consensus control problems of second-order multi-agent systems with one and mul-
tiple leaders under a directed graph and established some sufficient conditions for
the achievement of the finite-time consensus tracking. Based on the matrix the-
ory, graph theory, homogeneity with dilation, and LaSalle’s invariance principle,
Guan et al. [113] investigated the finite-time consensus tracking control problem
for second-order networked multi-agent systems under fixed and switched topolo-
gies, and designed finite-time consensus protocols by the pinning control technique
without the assumption that the topology graph is connected or the leader is glob-
ally reachable. Based on homogeneous theory, Lu et al. [114] considered finite-time
consensus tracking control problems for multi-agent systems with a virtual leader
under fixed and switching topologies, respectively, in which an improved consen-
sus tracking protocol was proposed in that the chattering phenomenon occurred in
non-Lipschitz dynamical systems can be eliminated by introducing a saturation func-
tion to replace the original sign function. By using tools from homogeneous theory,
Zhao et al. [115] addressed the finite-time consensus tracking problem of networked
multiple agents with Euler–Lagrange dynamics. Based on the homogeneous control
method, finite-time position consensus and collision avoidance problems were inves-
tigated for multiple autonomous underwater vehicle (multi-AUV) systems by Li and
Wang in [116]. Lu et al. [117] considered the finite-time consensus tracking prob-
lems for double-integrator multi-agent systems with bounded control input by using
homogeneous theory. By applying the homogeneous theory to stability analysis, the
work in [118] by Zhang et al. was devoted to the distributed finite-time observers for
multi-agent systems with input saturations and without velocity measurements. The
work in [119] by Yu et al. was concerned with the finite-time consensus problem
of distributed agents having nonidentical unknown nonlinear dynamics, to a leader
agent that also has unknown nonlinear control input signal, in which homogeneous
Lyapunov functions and homogeneous vector fields were introduced in the stability
analysis although the whole system is not homogeneous.
Adding power integrator-based control. Adding a power integrator, as a feed-
back design tool, was first introduced by Caron in [120] for the stabilization problem
and then was applied extensively by Lin and Qian [121–123] to solve the problem
of stabilization for uncertain nonlinear systems with unmatched uncertainties. Based
on adding a power integrator method, Li et al. [124] addressed the finite-time con-
sensus control problems for linear multi-agent systems with second-order dynamics
and unknown bounded disturbances in two cases: without a leader and with one
leader, and showed that the disturbance rejection property of the closed-loop system
can be enhanced by adjusting the fractional power in the continuous non-smooth
finite-time consensus controller. However, the communication topology considered
in [124] is restricted to be an undirected connected graph, and the effect of commu-
nication delays is not considered. By integrating the non-smooth finite-time control
1.2 Overview of Recent Research Results in Cooperative Control … 17

technique and adding a power integrator method, Du et al. [125] investigated the
robust consensus control problem for second-order multi-agent systems subject to
external disturbances for more general cases: (i) The communication topology is a
directed graph which is more general than that of [124]; (ii) there exists communi-
cation delay between the agents which is not considered in [124]. On the basis of the
work in [124], Du, Li, and Lin, by applying the adding a power integrator technique
and a finite-time observer, further studied the finite-time formation control of multi-
ple second-order agents via dynamic output feedback under the assumption that the
velocities of all agents cannot be measured. Wang et al. [127] further extended the
finite-time consensus tracking method proposed in [124] to the finite-time contain-
ment control for second-order linear multi-agent systems with multiple leaders under
undirected topology. With the adding a power integrator method, Huang et al. [128]
considered the finite-time leaderless consensus for a group of nonlinear mechanical
systems under undirected topology with a backstepping-like adaptive control scheme,
in which the nonlinearities in the system are linearly parameterized. Upon using the
non-smooth finite-time control method and adding a power integrator technique, Ou,
Du, and Li further addressed the finite-time consensus tracking control problem for
multiple non-holonomic wheeled mobile robots in dynamic model in [129]. With
the aid of the adding a power integrator technique, Wang et al. [130] investigated
the problem of achieving rotating formation and containment simultaneously via
finite-time control schemes for second-order linear multi-agent systems, in which
the desired rotating formation structure is time-varying. By integrating the fractional
state feedback method and adding a power integrator technique, Wang et al. [131]
considered the finite-time leaderless consensus problem for a group of nonlinear
mechanical systems under single-way directed interaction topology and unknown
actuation failures, in which the resultant control gain of the system is unknown and
time-varying, making the control impact on the system uncertain and the finite-time
control synthesis non-trivial. In [132], Wang et al. further extended the finite-time
leaderless consensus control methods for second-order nonlinear multi-agent sys-
tems with non-parametric uncertainties to high-order case under directed topology
by using the adding a power integrator technique associated with high-order case.
Based on fractional power feedback of local error and the adding a power integrator
technique, Wang et al. proposed a finite-time distributed adaptive control solution to
the formation–containment problem for multiple networked nonlinear systems with
uncertainties and directed communication constraints in [133].
Time-varying gain-based regular state feedback control. The time-varying
gain-based regular state feedback finite-time control was first introduced by Song et
al. in [70] for the robust regulation of normal-form nonlinear systems in prescribed
finite time and then was applied extensively in [134] to solve the pre-specified finite-
time leader-following control problem of high-order linear multi-agent systems under
directed graphs. This finite-time control method is based on the time-varying control
gain together with regular state feedback, and thus the control input signal is not
only continuous but also C 1 smooth. Particularly, the convergence time is not only
18 1 Introduction

finite, but can be pre-specified by the designer due to the independence of any other
design parameters and initial conditions. For more detailed results on the prescribed
finite-time control, the authors can be referred to [71].

1.3 Notes

As finite-time convergence is of special interest to many important applications that

involve multiple agent systems (MASs) [95], the past few years have witnessed
sustained growing interest in finite-time control of MAS, leading to fruitful results on
finite-time consensus and/or containment of MAS in the literatures; see [68–133]. It
is noted that, however, most existing finite-time control methods cannot guarantee the
convergence within pre-specified finite time in that the actual converging time period
is not uniform nor pre-specifiable. This is because the finite time T ∗ is determined by
T ∗ ≤ Vγ (t(1−α)
0)
1−α
[96], with γ > 0 and 0 < α < 1 being design parameters, and V (t0 )
being the Lyapunov function of system initial states, from which it is seen that the
finite time T ∗ depends on both the initial conditions and a set of design parameters,
and thus cannot be pre-specified arbitrarily. In other words, for a prescribed T ∗ , it is
non-trivial to determine the corresponding value for the crucial parameters α, γ , etc.
Furthermore, several other constraints are imposed in the existing finite-time con-
trol methods. For instance, in [72–79], signum function-based discontinuous control
has to be used to achieve finite-time consensus of MAS under certain topology con-
ditions, which makes the control action discontinuous. The methods suggested in
[124–133] involve the fractional power state feedback, rendering non-smooth con-
trol action. So far, as summarized in the above analysis, most existing finite-time
control schemes for MAS are either discontinuous or non-smooth, and the problem
of smooth distributed control for networked multiple systems (especially for the
nonlinear multi-agent systems with unknown uncertainties) featured with uniform
pre-specified finite settling time has remained almost untouched, which represents
an interesting topic for future research.

References

1. Ren, W., Beard, R., Atkins, E.: Information consensus in multivehicle cooperative control.
IEEE Control Syst. Mag. 27(2), 71–82 (2007)
2. Wang, W., Slotine, J.: A theoretical study of different leader roles in networks. IEEE Trans.
Autom. Control 51(7), 1156–1161 (2006)
3. Ren, W., Moore, K., Chen, Y.: High-order and model reference consensus algorithms in
cooperative control of multivehicle systems. ASME J. Dyn. Syst. Meas. Control 129(5), 678–
688 (2007)
4. Hong, Y., Chen, G., Bushnell, L.: Distributed observers design for leader-following control
of multi-agent networks. Automatica 44(3), 846–850 (2008)
5. Wang, X., Chen, G.: Pinning control of scale-free dynamical networks. Physica A 310(3–4),
521–531 (2002)
References 19

6. Olfati-Saber, R., Murray, R.: Consensus problems in networks of agents with switching topol-
ogy and time-delays. IEEE Trans. Autom. Control 49(9), 1520–1533 (2004)
7. Kim, Y., Mesbahi, M.: On maximizing the second smallest eigen-value of a state-dependent
graph laplacian. IEEE Trans. Autom. Control 51(1), 116–120 (2006)
8. Xiao, L., Boyd, S.: Fast linear iterations for distributed averaging. Syst. Control Lett. 53,
65–78 (2006)
9. Bhat, S., Bernstein, D.: Finite-time stability of continuous autonomus systems. SIAM J.
Control Optim. 38(3), 751–766 (2000)
10. Ren, W.: Collective motion from consensus with cartesian coordinate coupling. IEEE Trans.
Autom. Control 54(6), 1330–1335 (2009)
11. Olfati-Saber, R.: Flocking for multi-agent dynamic systems: algorithms and theory. IEEE
Trans. Autom. Control 51(3), 401–420 (2006)
12. Moreau, L.: Stability of multiagent systems with time-dependent communication links. IEEE
Trans. Autom. Control 50(2), 169–182 (2005)
13. Das, A., Lewis, F.L.: Distributed adaptive control for synchronization of unknown nonlinear
networked systems. Automatica 46(12), 2014–2021 (2010)
14. Fax, J.A., Murry, R.M.: Information flow and cooperative control of vehicle formations. IEEE
Trans. Autom. Control 49(9), 1465–1476 (2004)
15. Lafferriere, G., Williams, A., Caughman, J., Veerman, J.J.P.: Decentralized control of vehicle
formations. Syst. Control Lett. 54(9), 899–910 (2005)
16. Ma, C.Q., Zhang, J.F.: On formability of linear continuous-time multi-agent systems. J. Syst.
Sci. Complex. 25(1), 13–29 (2012)
17. Vicsek, T., Czirok, A., Jacob, E.B., Cohen, I., Schochet, O.: Novel type of phase transitions
in a system of self-driven particles. Phys. Rev. Lett. 75, 1226–1229 (1995)
18. Jadbabaie, A., Lin, J., Morse, A.S.: Coordination of groups of mobile autonomous agents
using nearest neighbor rules. IEEE Trans. Autom. Control 48(6), 988–1001 (2003)
19. Ren, W., Beard, R.W.: Consensus seeking in multiagent systems under dynamically changing
interaction topologies. IEEE Trans. Autom. Control 50(5), 655–661 (2005)
20. Ren, W., Beard, R.W., Atkins, E.M.: Information consensus in multivehicle cooperative con-
trol. IEEE Control Syst. Mag. 27(2), 71–82 (2007)
21. Corts, J.: Distributed algorithms for reaching consensus on general functions. Automatica
44(3), 726–737 (2008)
22. Ren, W., Atkins, E.M.: Distributed multi-vehicle coordinated control via local information
exchange. Int. J. Robust Nonlinear Control 17(10–11), 1002–1033 (2007)
23. Ren, W.: On consensus algorithms for double-integrator dynamics. IEEE Trans. Autom. Con-
trol 53(6), 1503–1509 (2008)
24. Yu, W., Chen, G., Cao, M.: Some necessary and sufficient conditions for second-order con-
sensus in multi-agent dynamical systems. IEEE Trans. Autom. Control 46(6), 1089–1095
(2010)
25. Abdessameud, A., Tayebi, A.: On consensus algorithms for double-integrator dynamics with-
out velocity measurements and with input constraints. Syst. Control Lett. 59(12), 812–821
(2010)
26. Wang, J.H., Cheng, D.Z.: Consensus of multi-agent systems with higher order dynamics. In:
Proceedings of 26th Chinese Control Conference, pp. 76–765 (2007)
27. Jiang, F.C., Wang, L., Xie, G.M.: Consensus of high-order dynamic multi-agent systems with
switching topology and time-varying delays. J. Control Theory Appl. 8(1), 52–60 (2010)
28. Xi, J.X., Cai, N., Zhong, Y.S.: Consensus problems for high-order linear time-invariant swarm
systems. Physica A 389(24), 5619–5627 (2010)
29. Dong, X.W., Xi, J.X., Shi, Z.Y., Zhong, Y.S.: Practical consensus for high-order linear time-
invariant swarm systems with interaction uncertainties, time-varying delays and external dis-
turbances. Int. J. Syst. Sci. 44(10), 1843–1856 (2013)
30. Yu, W., Chen, G., Cao, M., et al.: Second-order consensus for multiagent systems with directed
topologies and nonlinear dynamics. IEEE Trans. Syst. Man Cybern. 40(3), 881–891 (2010)
20 1 Introduction

31. Wang, Y.J., Song, Y.D., Lewis, F.L.: Robust adaptive fault-tolerant control of multi-agent
systems with uncertain non-identical dynamics and undetectable actuation failures. IEEE
Trans. Ind. Electron. 62(6), 3978–3988 (2015)
32. Qin, J.H., Gao, H.J., Zheng, W.X.: On average consensus in directed networks of agents with
switching topology and time delay. Int. J. Syst. Sci. 42(12), 1947–1956 (2011)
33. Tian, Y.P., Liu, C.L.: Consensus of multi-agent systems with diverse input and communication
delays. IEEE Trans. Autom. Control 53(9), 2122–2128 (2008)
34. Wu, Y.J., Wang, L.: Average consensus of continuous-time multi-agent systems with quantized
communication. Int. J. Robust Nonlinear Control 24(18), 334–3371 (2014)
35. Shang, Y.L.: Average consensus in multi-agent systems with uncertain topologies and multiple
time-varying delays. Linear Algebr. Appl. 459, 411–429 (2014)
36. Murray, R.M.: Recent research in cooperative control of multi-vehicle systems. J. Dyn. Syst.
Meas. Control 129(5), 571–583 (2007)
37. Cao, Y.C., Yu, W.W., Ren, W., Chen, G.R.: An overview of recent progress in the study of
distributed multi-agent coordination. IEEE Trans. Ind. Inf. 9(1), 427–438 (2013)
38. Ren, W.: Multi-vehicle consensus with a time-varying reference state. Syst. Control Lett.
56(7–8), 474–483 (2007)
39. Qin, J., Zheng, W.X., Gao, H.: Consensus of multiple second-order vehicles with a time-
varying reference signal under directed topology. Automatica 47(9), 1983–1991 (2011)
40. Cao, Y.C., Ren, W., Li, Y.: Distributed discrete-time coordinated tracking with a time-varying
reference state and limited communication. Automatica 45(5), 474–483 (2009)
41. Gu, D.B., Wang, Z.Y.: Leader-follower flocking: algorithms and experiments. IEEE Trans.
Control Syst. Tech. 17(5), 1211–1219 (2009)
42. Kokiopoulou, E., Frossard, P.: Polynomial filtering for fast convergence in distributed con-
sensus. IEEE Trans. Signal Process. 57(1), 342–354 (2009)
43. Wu, Z.H., Peng, L., Xie, L.B., et al.: Stochastic bounded consensus tracking of leader-follower
multi-agent systems with measurement noises based on sampled-data with small sampling
delay. Physica A 392(4), 918–928 (2013)
44. Qi, B., Lou, K., Miao, S., et al.: Second-order consensus of leader-following multi-agent
systems with jointly connected topologies and time-varying delays. Arabian J. Sci. Eng.
39(2), 143–1440 (2014)
45. Zhang, X.F., Liu, L., Feng, G.: Leader-follower consensus of time-varying nonlinear multi-
agent systems. Automatica 52, 8–14 (2015)
46. Hong, Y., Hu, J., Gao, L.: Tracking control for multi-agent consensus with an active leader
and variable topology. Automatica 42(7), 1177–1182 (2006)
47. Ni, W., Cheng, D.Z.: Leader-following consensus of multi-agent systems under fixed and
switching topologies. Syst. Control Lett. 59(3–4), 209–217 (2010)
48. Cao, Y., Ren, W.: Distributed coordinated tracking with reduced interaction via a variable
structure approach. IEEE Trans. Autom. Control 57(1), 33–48 (2011)
49. Wang, Y.J., Song, Y.D., Gao, H., Lewis, F.L.: Distributed fault-tolerant control of virtuallyand
physically interconnected systems with application to high-speed trains undertraction/braking
failures. IEEE Trans. Intell. Transp. Syst. 17(2), 535–545 (2016)
50. Cao, Y., Ren, W.: Containment control with multiple stationary or dynamic leaders under a
directed interaction graph. In: CDC 2009, Shanghai, China, pp. 3014–3019 (2009)
51. Cao, Y., Stuart, D., Ren, W., et al.: Distributed containment control for multiple autonomous
vehicles with double-integrator dynamics: algorithms and experiments. IEEE Trans. Control
Syst. Technol. 19(4), 929–938 (2011)
52. Mei, J., Ren, W., Ma, G.: Distributed adaptive coordination for multiple Lagrangian systems
under a directed graph without using neighbors velocity information. Automatica 49(6), 1723–
1731 (2013)
53. Li, Z., Ren, W., Liu, X., et al.: Distributed containment control of multi-agent systems with
general linear dynamics in the presence of multiple leaders. Int. J. Robust Nonlinear Control
23(5), 534–547 (2013)
References 21

54. Ji, M., Ferrari-Trecate, G., Egerstedt, M., Buffa, A.: Containment control in mobile networks.
IEEE Trans. Autom. Control 53(8), 1972–1975 (2008)
55. Meng, Z.Y., Ren, W., You, Z.: Distributed finite-time attitude containment control for multiple
rigid bodies. Automatica 46(12), 2092–2099 (2010)
56. Notarstefano, G., Egerstedt, M., Haque, M.: Containment in leader-follower networks with
switching communication topologies. Automatica 47(5), 1035–1040 (2011)
57. Cao, Y.C., Ren, W., Egerstedt, M.: Distributed containment control with multiple stationary
or dynamic leaders in fixed and switching directed networks. Automatica 48(8), 1586–1597
(2012)
58. Liu, H.Y., Xie, G.M., Wang, L.: Necessary and sufficient conditions for containment control
of networked multi-agent systems. Automatica 48(7), 1415–1422 (2012)
59. Lou, Y.C., Hong, Y.G.: Target containment control of multi-agent systems with random switch-
ing interconnection topologies. Automatica 48(5), 879–885 (2012)
60. Dong, X.W., Xi, J.X., Lu, G., Zhong, Y.S.: Containment analysis and design for high-order lin-
ear time-invariant singular swarm systems with time delays. Int. J. Robust Nonlinear Control
24(7), 1189–1204 (2014)
61. Liu, H.Y., Xie, G.M., Wang, L.: Containment of linear multi-agent systems under general
interaction topologies. Syst. Control Lett. 61(4), 528–534 (2012)
62. Olfati-Saber, R.: Ultrafast consensus in small-world networks. In: ACC 2005, Portland, OR,
USA, pp. 2374–2378 (2005)
63. Nosrati, S., Shafiee, M.: Time-delay dependent stability robustness of small-world protocols
for fast distributed consensus seeking. In: The 5th International Symposium on Modeling and
Optimization in Mobile, Ad Hoc and Wireless Networks and Workshops (2007)
64. Xiao, L., Boyd, S.: Fast linear iterations for distributed averaging. Syst. Control Lett. 53(1),
65–78 (2004)
65. Kira, Y., Mesbahi, M.: On maximizing the second smallest eigenvalue of a state-dependent
graph laplacian. IEEE Trans. Autom. Control 51(1), 116–120 (2006)
66. Jin, Z.P., Murray, R.M.: Multi-hop relay protocols for fast consensus seeking. In: CDC 2006,
CA, USA, pp. 1001–1006 (2006)
67. Yang, W., Wang, X.F.: Consensus filters on small world networks. In: Proceedings of the 6th
World Congress on Intelligent Control and Automation, Dalian, China, pp. 1217–1221 (2006)
68. Weiss, L., Infante, E.F.: On the stability of systems defined over a finite time interval. Proc.
Natl. Acad. Sci. U.S.A. 1965, 44–48 (1965)
69. Weiss, L., Infante, E.F.: Finite time stability under perturbing forces and on product spaces.
IEEE Trans. Autom. Control 12(1), 54–59 (1967)
70. Song, Y.D., Wang, Y.J., Holloway, J., Krstic, M.: Time-varying feedback for finite-time robust
regulation of normal-form nonlinear systems. In: The 55th Conference on Decision and Con-
trol (CDC), pp. 3837–3842 (2016)
71. Song, Y.D., Wang, Y.J., Holloway, J., Krstic, M.: Time-Varying feedback for robust regulation
of normal-form nonlinear systems in prescribed finite time. Automatica 2017(83), 243–251
(2017)
72. Corts, J.: Finite-time convergent gradient flows with applications to network consensus. Auto-
matica 42(11), 1993–2000 (2006)
73. Chen, G., Lewis, F.L., Xie, L.H.: Finite-time distributed consensus via binary control proto-
cols. Automatica 47(9), 1962–1968 (2011)
74. Li, C.Y., Qu, Z.H.: Distributed finite-time consensus of nonlinear systems under switching
topologies. Automatica 50(6), 162–1631 (2014)
75. Zhang, Y., Yang, Y., Zhao, Y., Wen, G.: Distributed finite-time tracking control for nonlinear
multi-agent systems subject to external disturbances. Int. J. Control 86(1), 29–40 (2013)
76. Franceschelli, M., Guia, A., Pisano, A., Usai, E.: Finite-time consensus for switching network
topologies. Nonlinear Anal. Hybrid Syst. 10, 83–93 (2013)
77. Wen, G., Rahmani, A., Yu, Y.: Consensus tracking for multi-agent systems with nonlinear
dynamics under fixed communication topologies. In: Proceeding of the World Congress on
Engineering and Computer Science, San Francisco, USA (2011)
22 1 Introduction

78. Meng, Z., Lin, Z.: On distributed finite-time observer design and finite-time coordinated
tracking of multiple double integrator systems via local interactions. Int. J. Robust Nonlinear
Control 24(16), 2473–2489 (2014)
79. Fu, J., Wang, J.: Observer-based finite-time coordinated tracking for high-order integrator
systems with matched uncertainties under directed communication graphs. In: Proceeding of
the 11th IEEE International Conference on Control and Automation, pp. 880–885 (2014)
80. Emelyanov, S.V., Taranm, V.A.: On a class of variable structure control systems. In: Proceeding
of USSR Academy of Sciences, Energy and Automation, Moskov, p. 3 (1962)
81. Yu, X., Ledwich, G., Man, Z.: Terminal sliding mode control of linear systems. IEEE Trans.
Autom. Control (1994)
82. Yu, X., Man, Z.: Model reference adaptive control systems with terminal sliding modes. Int.
J. Control 64(6), 1165–1176 (1996)
83. Xu, X., Wang, J.: Finite-time consensus tracking for second-order multiagent systems. Asian
J. Control 15(4), 1246–1250 (2013)
84. Khoo, S., Xie, L., Man, Z.: Integral terminal sliding mode cooperative control of multi-robot
networks. In: IEEE/ASME International Conference Advanced Intelligent Mechatronics, Sin-
gapore, pp. 969–973 (2009)
85. Khoo, S., Xie, L., Man, Z.: Robust finite-time consensus tracking algorithm for multirobot
systems. IEEE/ASME Trans. Mechatron. 14(2), 219–228 (2009)
86. Zou, A., Kumar, K.D.: Distributed attitude coordination control for spacecraft formation
flying. IEEE Trans. Aerosp. Electron. Syst. 48(2), 1329–1346 (2012)
87. Chen, G., Yue, Y.L., Song, Y.D.: Finite-time cooperative-tracking control for networked Euler-
Lagrange systems. IET Control Theory Appl. 7(11), 1487–1497 (2013)
88. Ghasemi, M., Nersesov, S.G.: Finite-time coordination in multiagent systems using sliding
mode control approach. Automatica 50(4), 1209–1216 (2014)
89. He, X., Wang, Q., Yu, W.: Finite-time distributed cooperative attitude tracking control for
multiple rigid spacecraft. Appl. Math. Comput. 256, 724–734 (2015)
90. Cao, Y., Ren, W., Meng, Z.: Decentralized finite-time sliding mode estimators and their
applications in decentralized finite-time formation tracking. Syst. Control Lett. 59(9), 522–
529 (2010)
91. Kamble, U.B., Chandle, J.O., Lahire, V.S., Langde, R.D.: Second order twisting sliding mode
control of multi-agent network with input disturbance. In: Proceeding of the 2013 4th Inter-
national Conference on Computing, Communications and Networking Technologies (2013)
92. Li, H., Liao, X., Chen, G.: Leader-following finite-time consensus in second-order multi-agent
networks with nonlinear dynamics. Int. J. Control Autom. Syst. 11(2), 422–426 (2013)
93. Zhao, L., Hua, C.: Finite-time consensus tracking of second-order multi-agent systems via
nonsingular TSM. Nonlinear Dyn. 75(1–2), 311–318 (2014)
94. Yang, Z., Shibuya, Y., Qin, P.: Distributed robust control for synchronized tracking of net-
worked Euler-Lagrange systems. Int. J. Syst. Sci. 46(4), 720–732 (2015)
95. Haimo, V.T.: Finite time controllers. SIAM J. Contr. Optim. 24, 760–770 (1986)
96. Bhat, S.P., Bernstein, D.S.: Continuous finite-time stabilization of the translational and rotional
double integrators. IEEE Trans. Autom. Control 43(5), 678–682 (1998)
97. Wang, L., Xiao, F.: Finite-time consensus problems for networks of dynamic agents. IEEE
Trans. Autom. Control 55(4), 950–955 (2010)
98. Xiao, F., Wang, L., Chen, J., et al.: Finite-time formation control for multi-agent systems.
Automatica 45(11), 2605–2611 (2009)
99. Xiao, F., Wang, L., Chen, T.: Finite-time consensus in networks of integrator-like dynamic
agents with directional link failure. IEEE Trans. Autom. Control 59(3), 756–762 (2014)
100. Cao, Y., Ren, W.: Finite-time consensus for multi-agent networks with unknown inherent
nonlinear dynamics. Automatica 50(10), 2648–2656 (2014)
101. Zuo, Z., Tie, L.: A new class of finite-time nonlinear consensus protocols for multi-agent
systems. Int. J. Control 87(2), 363–370 (2014)
102. Zuo, Z., Tie, L.: Distributed robust finite-time nonlinear consensus protocols for multi-agent
systems. Int. J. Control 47(6), 1366–1375 (2016)
References 23

103. Bhat, S.P., Bernstein, D.S.: Finite time stability of homogeneous systems. In: In Proceedings
in American Control Conference, pp. 2513–2514 (1997)
104. Hong, Y.G., Huang, J., Xu, Y.S.: On an output feedback finite-time stabilization problem.
IEEE Trans. Autom. Control 46(2), 305–309 (2001)
105. Hong, Y.G.: Finite-time stabilization and stabilizability of a class of controllable systems.
Syst. Control Lett. 46, 231–236 (2002)
106. Hong, Y.G., Jiang, Z.P.: Finite-time stabilization of nonlinear systems with parametric and
dynamic uncertainties. IEEE Trans. Autom. Control 51(12), 1950–1956 (2006)
107. Hong, Y.G., Wang, J., Cheng, D.Z.: Adaptive finite-time control of nonlinear systems with
parametric uncertainty. IEEE Trans. Autom. Control 51(5), 858–862 (2006)
108. Hermes, H.: Homogeneous coordinates and continuous asymptotically stabilizing feedback
controls. In: Elaydi, S. (ed.) Differential Equations, Stability and Control, pp. 249–260. Marcel
Dekker, New York (1991)
109. Rosier, L.: Homogeneous Lyapunov function for homogeneous continuous vector field. Syst.
Control Lett. 19(4), 467–473 (1992)
110. Hui, Q., Haddad, W.M., Bhat, S.P.: Finite-time semistability and consensus for nonlinear
dynamical networks. IEEE Trans. Autom. Control 53(8), 1887–1990 (2008)
111. Du, H., He, Y., Cheng, Y.: Finite-time synchronization of a class of second-order nonlinear
multi-agent systems using output feedback control. IEEE Trans. Circuits Syst. I 61(6), 1778–
1788 (2014)
112. Zhang, Y., Yang, Y.: Finite-time consensus of second-order leader-following multi-agent sys-
tems without velocity measurements. Phys. Lett. A 337(3–4), 243–249 (2013)
113. Guan, Z., Sun, F., Wang, Y., Li, T.: Finite-time consensus for leader-following second-order
multi-agent networks. IEEE Trans. Circuits Syst. I 59(11), 2646–2654 (2012)
114. Lu, X., Lu, R., Chen, S., Lu, J.: Finite-time distributed tracking control for multi-agent systems
with a virtual leader. IEEE Trans. Circuits Syst. I 60(2), 352–362 (2013)
115. Zhao, Y., Duan, Z., Wen, G.: Distributed finite-time tracking of multiple Euler-Lagrange
systems without velocity measurements. Int. J. Robust Nonlinear Control 25(11), 1688–1703
(2015)
116. Li, S., Wang, X.: Finite-time consensus and collision avoidance control algorithms for multiple
AUVs. Automatica 49(11), 3359–3367 (2013)
117. Lu, X., Chen, S., Lu, J.: Finite-time tracking for double-integrator multi-agent systems with
bounded control input. IET Control Theory Appl. 7(11), 1562–1573 (2013)
118. Zhang, B., Jia, Y., Matsno, F.: Finite-time observers for multi-agent systems without velocity
measurements and with input saturations. Syst. Control Lett. 68, 86–94 (2014)
119. Yu, H., Shen, Y., Xia, X.: Adaptive finite-time consensus in multi-agent networks. Syst.
Control Lett. 62(10), 880–889 (2013)
120. Caron, J.M.: Adding an integrator for the stabilization problem. Syst. Control Lett. 17, 89–104
(1991)
121. Lin, W., Qian, C.: Adding one power integrator: a tool for global stabilization of high-order
lower-triangular systems. Syst. Control Lett. 39(5), 339–351 (2000)
122. Lin, W., Qian, C.: Adaptive control of nonlinearly paramterized systems: the smooth feedback
case. IEEE Trans. Autom. Control 47(8), 1249–1266 (2002)
123. Lin, W., Qian, C.: Adaptive control of nonlinearly paramterized systems: a nonsmooth feed-
back framwork. IEEE Trans. Autom. Control 47(5), 757–774 (2002)
124. Li, S., Du, H., Lin, X.: Finite-time consensus algorithm for multi-agent systems with double-
integrator dynamics. Automatica 47(8), 1706–1712 (2011)
125. Du, H., Li, S., Shi, P.: Robust consensus algorithm for sencond-order multi-agent systems
with external disturbances. Int. J. Control 85(12), 1913–1928 (2012)
126. Du, H., Li, S., Lin, X.: Finite-time formaiton control of multi-agent systems via dynamic
output feedback. Int. J. Robust Nonlinear Control 23, 1609–1628 (2013)
127. Wang, X., Li, S., Shi, P.: Distributed finite-time containment control for double-integrator
multiagent systems. IEEE Trans. Cybern. 44(9), 1518–1528 (2014)
24 1 Introduction

128. Huang, J., Wen, C., Wang, W., Song, Y.D.: Adaptive finite-time consensus control of a group
of uncertain nonlinear mechanical systems. Automatica 51, 292–301 (2015)
129. Ou, M., Du, H., Li, S.: Finite-time tracking contrl of multiple noholonomic mobile robots. J.
Frankl. Inst. 349, 2834–2860 (2012)
130. Wang, Y.J., Song, Y.D., Krstic, M.: Collectively rotating formation and containment deploy-
ment of multi-agent systems: a polar coordinate based finite time approach. IEEE Trans.
Cybern. 47, 2161–2172 (2017)
131. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Fault-tolerant finite time consensus for multiple
uncertain nonlinear mechanical systems under single-way directed communication interac-
tions and actuation failures. Automatica 63, 374–383 (2016)
132. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Adaptive finite time coordinated consensus for
high-order multi-agent systems: adjustable fraction power feedback approach. Inf. Sci. 372,
392–406 (2016)
133. Wang, Y.J., Song, Y.D., Ren, W.: Distributed adaptive finite time approach for formation-
containment control of networked nonlinear systems under directed topology. IEEE Trans.
Neural Netw. Learn. Syst. https://doi.org/10.1109/TNNLS.2017.2714187 (2017)
134. Wang, Y.J., Song, Y.D.: Leader-following control of high-order multi-agent systems under
directed graphs: pre-specified finite time approach. Automatica 87, 113–120 (2018)
Chapter 2
Preliminaries

This chapter introduces algebraic graph theory, matrix analysis theory on graphs,
stability analysis theory on cooperative control systems, and the theory of finite-time
stability analysis.

2.1 Algebraic Graph Theory

For a group of agents interacting with each other through communication network,
it is natural to model the interaction among agents by a directed weighted graph.
Suppose that there are N agents in the group. A directed graph with N nodes (agents)
is represented by G = (ι, ε), in which ι = {ι1 , ι2 , . . . , ι N } denotes a nonempty finite
set of nodes (agents) and ε ⊆ ι × ι denotes the set of edges between two distinct
nodes. A weighted graph denotes that a graph associates a weight with every edge
in the graph. A directed edge is represented by an ordered pair of distinct nodes
εi j = (ιi , ι j ), where ιi is called the parent (terminal) node and ι j is called the child
(initial) node, meaning that node ιi can receive information from node ι j , but not
necessarily vice versa. We assume the graph is simple, i.e., no self loops ((ιi , ιi ) ∈ / ε,
∀i ∈ {1, . . . , N }) and no multiple edges between the same pairs of nodes. Node ι j is
an in-neighbor of node ιi if ιi can obtain information from ι j , i.e., (ι j , ιi ) ∈ ε. The set
of in-neighbors of node ιi is defined as Ni = {ι j ∈ ι|(ι j , ιi ) ∈ ε}. In an undirected
graph, the pairs of nodes are unordered, in which the edge (ιi , ι j ) corresponds to
the edges (ιi , ι j ) and (ι j , ιi ) in the directed graph. An undirected graph is actually a
special case of a directed graph. A directed path is denoted by a sequence of ordered
edges (ιi1 , ιi2 ), (ιi2 , ιi3 ),. . ., where ιik ∈ ι. A cycle refers to a directed path starting
and ending at the same node. A directed graph is strongly connected if there is a
directed path from every node to any other distinct node. For an undirected graph,
it is said to be connected if there is an undirected path between any pair of distinct
nodes, and further, if there exists an edge between any pair of distinct nodes, this
undirected graph is said to be fully connected. A directed graph is complete if there
is an edge from every node to any other node. A root is a node that has no parent and
© Springer Nature Switzerland AG 2019 25
Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_2
26 2 Preliminaries

Fig. 2.1 Different network

graph topologies. Notice that
graph c contains a directed
spanning tree with 1 being
the root node but is not
strongly connected because
nodes 2, 3, 4, and 5 do not
have directed paths to all
other nodes

has directed paths to all other nodes in a directed graph. A directed tree is a directed
graph where every node has exactly one parent except for the root. A directed tree
has no cycle because every edge is oriented away from the root. In the undirected
graph, a tree refers to a graph where each pair of nodes is connected by exactly one
undirected path. A directed spanning tree is a directed tree formed by graph edges
that connects all the nodes of the graph. In a graph if a subset of the edges forms a
directed tree, the graph is said to have a spanning tree. A graph may have multiple
spanning trees. If a graph is strongly connected (Fig. 2.1a, b), then it contains at
least one spanning tree (Fig. 2.1c). The in-degree of ιi is the number of nodes ιi can
obtain information from whom. The out-degree of ιi is the number of nodes who can
obtain information from ιi . If the in-degree of a node ιi (i = 1, . . . , N ) is equal to
its out-degree, the node is balanced. If all nodes ιi in a graph are balanced, the graph
is balanced.

2.2 Matrix Analysis on Graphs

The graph structure and properties can be studied by examining the properties of the
graph topology of the graph. This is known as algebraic graph theory [1, 2].
2.2 Matrix Analysis on Graphs 27

Denote by ai j the weight for the edge εi j ∈ ε, then the graph topology of the
network G can be represented by a weighted adjacency matrix A = [ai j ], where
εi j ∈ ε ⇔ ai j > 0, and otherwise, ai j = 0. Self-edges are not allowed, that is, aii = 0
for all i, unless otherwise indicated. If the weight is not relevant, then ai j is set equal
to 1 if εi j ∈ ε. In an undirected graph, the adjacency matrix A is symmetric, i.e.,
A = A T.
weighted in-degree of node i is denoted by the i-th row sum of A , i.e.,
The
di = Nj=1 ai j , and the weighted out-degree of node ιi is denoted by the i-th column
  
sum of A, i.e., di0 = Nj=1 a ji . Note that Nj=1 ai j = j∈N i ai j . For a graph, if
di = di0 for all i, the graph is weight balanced. For brevity, we sometimes simply
refer to node ιi as node i, and refer simply to in-degree, out-degree, and the balanced
property, without the qualifier ‘weight’.

Definition 2.1 ([3]) The weighted Laplacian graph matrix is denoted by

L = [li j ] = D − A ∈ R N ×N ,

where D is the diagonal in-degree matrix defined as

D = diag(d1 , . . . , d N ) ∈ R N ×N .

The Laplacian matrix is of extreme importance in the study of dynamical multi-

agent systems on graphs. For an undirected graph, L is symmetric and is called
the Laplacian matrix. For a directed graph, L is not necessarily symmetric and
is sometimes called the nonsymmetric Laplacian matrix [4] or directed Laplacian
matrix [5].

Lemma 2.1 ([6]) Let L be the nonsymmetric Laplacian matrix (Laplacian matrix)
of the directed graph G (undirected graph G ). Then L has at least one zero eigen-
value and all its nonzero eigenvalues have positive real parts (are positive). Further-
more, L has a simple zero eigenvalue and all its other eigenvalues have positive
real parts (are positive) if and only if G has a directed spanning tree (is connected).
In addition, L 1 N = 0 N and there exists a nonnegative vector p ∈ R N satisfying
pT L = 01×N and pT 1 N = 1. N N
For an undirected graph G , it holds that x T L x = 21 i=1 j=1 ai j (x i − x j ) ,
2

where x = [x1 , . . . , x N ] .
T

Lemma 2.2 ([6]) Let L be the nonsymmetric Laplacian matrix of the directed
graph G . If G is balanced, then x T L x ≥ 0. If G is strongly connected and balanced
(containing undirected and connected graph as a special case), then L x = 0 N or
x T L x = 0 if and only if xi = x j for all i, j ∈ {1, . . . , N }.
Let λ1 (L ) ≤ λ2 (L ) ≤ · · · ≤ λ N (L ) be the N eigenvalues of L associated an
undirected graph G , so that λ1 (L ) = 0 and λ2 (L ) denotes the algebraic connec-
tivity, which is positive if and only if G is connected according to Lemma 2.1. The
algebraic connectivity quantifies the convergence rate of consensus algorithms [6].
28 2 Preliminaries

Lemma 2.3 ([7]) Suppose that z = [z 1T , . . . , z TN ]T with z i ∈ Rm . Let A and L

be the adjacency matrix and nonsymmetric Laplacian matrix associated with the
directed graph G of order N , respectively. Then the following five conditions are
equivalent:
(i) The directed graph G has a directed spanning tree;
(ii) The rank of L is N − 1;
(iii) L has a simple zero eigenvalue with an associated eigenvector 1 N and all other
eigenvalues have positive real parts;
(iv) (L ⊗ Im )z = 0 N m if and only if z 1 = · · · = z N ;
(v) Consensus is reached for the closed-loop system ż = −(L ⊗ Im )z.
For the leader–follower MAS containing 1 leader and N followers, where a leader
is an agent without in-neighbors and a follower is an agent that has at least one in-
neighbor, let G¯ be the corresponding directed graph between the 1 leader, labeled as
agent 0, and N followers, labeled as agents or followers 1 to N . Let A¯ and L¯ be
the adjacency matrix and the nonsymmetric Laplacian matrix (Laplacian matrix)
associated with G¯, respectively. In addition, let G , A , and L be, respectively,
the directed graph between the N followers, the adjacency matrix, and the non-
symmetric
 Laplacian
 matrix (Laplacian matrix) associated with G . In such case,
0 0
A¯ = 1×N
, where ā = [a10 , . . . , a N 0 ]T and ai0 > 0 (i = 1, . . . , N ) if agent
ā A
 
¯ 0 01×N
0 is an in-neighbor of agent i and ai0 = 0 otherwise, and L = , where
ā H
H = L + diag{a10 , . . . , a N 0 }.
Lemma 2.4 ([6]) Let G and H be defined the same as in the above, if G is directed
(undirected), all eigenvalues of H have positive real parts (H is symmetric positive
definite) if and only if the leader in the directed graph G has directed paths to all
followers.

2.3 Matrix and Algebra Theory Background

Definition 2.2 ([8]) A Hurwitz matrix or a stability matrix is a matrix A ∈ R N ×N

satisfying that all its eigenvalues have negative real parts.
Definition 2.3 ([9]) The M-matrix is a matrix of the form

B = α I N − C,

where α ∈ R+ , C ∈ R N ×N is semi-positive definite, satisfying ρ(C) ≤ α (ρ(C) is

the modulus of C). Further, if ρ(C) < α, the matrix B is a nonsingular M-matrix.
Lemma 2.5 ([9]) Let Z N ×N = {B = [bi j ] ∈ R N ×N |bi j ≤ 0, i = j}. A matrix B ∈
Z N ×N is a nonsingular M-matrix if and only if B −1 exists and B −1 ≥ 0.
2.3 Matrix and Algebra Theory Background 29

Lemma 2.6 ([10] Gershgorin’s disk theorem) Let A = [ai j ] ∈ R N ×N , and


N
Ri (A) = |ai j |, i = 1, . . . , N ,
j=1, j =i

be the deleted absolute row sums of A. Then all eigenvalues of A are located in the
union of N disks

∪iN {z ∈ C : |z − ai i| ≤ Ri (A)}.

Furthermore, if a union of k of these N disks forms a connected region that is

disjoint from all of the remaining N − k disks, then in this region there are precisely
k eigenvalues of A.

Definition 2.4 ([11]) Let A = [ai j ] ∈ Rm×n and B = [bi j ] ∈ R p×q . The Kronecker
product of A and B is
⎡ ⎤
a11 B a12 B · · · a1n B
⎢ a21 B a22 B · · · a2n B ⎥
⎢ ⎥
A⊗B =⎢ . .. .. .. ⎥ ∈ R
mp×nq
.
⎣ .. . . . ⎦
am1 B am2 B · · · amn B

Lemma 2.7 ([11]) Let A, B, C, and D be matrices with appropriate dimensions

and k ∈ R, then the following equalities hold.
(1) k(A ⊗ B) = (k A) ⊗ B = A ⊗ (k B);
(2) A ⊗ (B + C) = A ⊗ B + A ⊗ C; (B + C) ⊗ A = B ⊗ A + C ⊗ A;
(3) (A + B) ⊗ (C + D) = A ⊗ C + A ⊗ D + B ⊗ C + B ⊗ D;
(4) (A ⊗ B) ⊗ C = A ⊗ (B ⊗ C) = A ⊗ B ⊗ C;
(5) (A ⊗ B)T = A T ⊗ B T ;
(6) (A ⊗ B)(C ⊗ D) = AC ⊗ B D.

Lemma 2.8 ([6] Holder inequality) Let 1 ≤ p, q ≤ ∞ with 1

p
+ 1
q
= 1. Then for
all vectors f ∈ Rk and g ∈ Rk , it holds that

| f T g| ≤ f p g q.

Lemma 2.9 ([12]) For any x, y ∈ R N and any symmetric positive-definite matrix
A ∈ R N ×N , it holds that

2x T y ≤ x T A−1 x + y Ay.

Lemma 2.10 ([10] Rayleigh-Ritz theorem) Let A ∈ R N ×N be a symmetric matrix.

Then for all x ∈ R N , it holds that
30 2 Preliminaries

λmin (A)x T x ≤ x T Ax ≤ λmax (A)x T x,

x T Ax x T Ax x T Ax
where λmin (A) = min x =0 N xT x
= min x T x=1 xT x
, λmax (A) = maxx =0 N xT x
=
T
maxx T x=1 xx TAx
x
.

Lemma 2.11 ([13] Young’s inequality) Let a, b ≥ 0 and c > 0, then it holds that
1
a 1+c cb1+ c
ab ≤ + .
1+c 1+c

Lemma 2.12 ([14]) Let a1 , a2 > 0 and 0 < c < 1, then it holds that

(a1 + a2 )c ≤ a1c + a2c .

Lemma 2.13 ([14]) Let a1 , a2 , . . . , an ∈ R+ and 0 < p < 2, then the following
inequality holds,
p p
(a12 + a22 + · · · + an2 ) p ≤ (a1 + a2 + · · · + anp )2 .

Lemma 2.14 ([15]) For any ε ∈ R+ and x ∈ R, it holds that

x
0 ≤ |x| − x tanh( ) ≤ kε, k = 0.2785.
ε
Lemma 2.15 ([16]) If h = h 2 / h 1 ≥ 1, where h 1 , h 2 are positive odd integers, then
it holds that |x − y|h ≤ 2h−1 |x h − y h |.

Lemma 2.16 ([16]) If 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are odd integers, then
|x h − y h | ≤ 21−h |x − y|h .

Lemma 2.17 ([16]) If c, d > 0, then for any x, y ∈ R, |x|c |y|d ≤ c/(c + d)|x|c+d +
d/(c + d)|y|c+d .

Lemma
N  N If 0 h< h ≤1−h
2.18 ([17])  N for xi h∈ R (i = 1, 2, . . . , N ), it holds that
1, then
( i=1 |xi |)h ≤ i=1 |xi | ≤ N ( i=1 |xi |) .

2.4 Stability Analysis Theory on Cooperative

Control Systems

In the study of dynamical systems, different kinds of stability problems arise. This
section is mainly concerned with the stability theory of networked cooperative control
systems.
To investigate the stability of networked cooperative systems, some basic defini-
tions and concepts are introduced as follows.
2.4 Stability Analysis Theory on Cooperative Control Systems 31

Consider a networked cooperative system consisting of N subsystems,

ẋi = f i (x, t), i = 1, . . . , N (2.1)

where xi ∈ Rm denotes the state of subsystem i (i = 1, . . . , N ), x = [x1 , . . . , x N ]T ∈

Rm N , f i : D × R+ → Rm (D ⊂ Rm N is a domain contains the origin) is locally Lip-
schitz in x and piecewise continuous in t on D × R+ .
For the networked cooperative systems without a leader, the neighborhood error
is defined as

ei = ai j (xi − x j ), i = 1, . . . , N . (2.2)
i∈N i

For the networked cooperative system with N followers and 1 leader, labeled 0,
the neighborhood error is defined as

ei = ai j (xi − x j ) + ai0 (xi − x0 ), i = 1, . . . , N , (2.3)
i∈N i

and the tracking error (or disagreement variable) is defined as

i = xi − x0 , i = 1, . . . , N . (2.4)

Denote by e = [e1 , . . . , e N ]T and  = [1 , . . . ,  N ]T .

In the following, we define the concepts of stability, asymptotic stability, exponen-
tial stability, and uniform ultimately boundedness for networked cooperative system
(2.1), which extend the standard concepts of stability for nonautonomous system in
[8].

Definition 2.5 The networked cooperative system (2.1) with (without) a leader is
• stable (Lyapunov stable) if, for each ε > 0, there exists δ(ε, t0 ) > 0 such that

e(t0 ) < δ ⇒ e(t) < ε, ∀ t ≥ t0 ≥ 0. (2.5)

• uniformly stable if, for each ε > 0, there exists δ(ε) > 0 (independent of t0 ), such
that (2.5) is satisfied.
• unstable if it is not stable.
• asymptotically stable if it is stable and there exists a finite constant c = c(t0 ) > 0
such that

lim e(t) = 0, ∀ e(t0 ) < c. (2.6)

t→∞

• uniformly asymptotically stable if it is uniformly stable and there exists a finite

constant c > 0 (independent of t0 ), such that
32 2 Preliminaries

lim e(t) = 0, ∀ e(t0 ) < c (2.7)

t→∞

• uniformly in t0 . That is, for each η > 0, there exists T = T (η) > 0 such that

e(t) < η, ∀t ≥ t0 + T (η), ∀ e(t0 ) < c. (2.8)

• globally uniformly asymptotically stable if it is uniformly stable, and δ(ε) can

be chosen such that limε→∞ δ(ε) = ∞, and, for every finite η > 0 and c > 0, there
exists T = T (η, c) > 0 such that

e(t) < η, ∀t ≥ t0 + T (η, c), ∀ e(t0 ) < c. (2.9)

Definition 2.6 The networked cooperative system (2.1) with (without) a leader is
• exponentially stable, if there exist finite constants c, k, λ > 0 such that

e(t) ≤ k e(t0 ) exp−λ(t−t0 ) , ∀ e(t0 ) < c. (2.10)

• globally exponentially stable if (2.10) is satisfied for any given initial state x(t0 ).
Definition 2.7 The networked cooperative system (2.1) with (without) a leader is
• uniformly bounded if there exists a finite constant c > 0 (independent of t0 ), and
for any a ∈ (0, c), there exists β = β(a) > 0 (independent of t0 ), such that

e(t0 ) < a ⇒ e(t) < β, ∀ t ≥ t0 ≥ 0. (2.11)

• globally uniformly bounded if (2.11) is satisfied for arbitrarily large a.

• uniformly ultimately bounded (UUB) with ultimate bound b if there exist finite con-
stants b, c > 0 (independent of t0 ), and for any a ∈ (0, c), there exists T = T (a, b) ≥
0 (independent of t0 ), such that

e(t0 ) < a ⇒ e(t) < b, ∀ t ≥ t0 + T. (2.12)

• globally UUB if (2.12) is satisfied for arbitrarily large a.

Definition 2.8 The networked cooperative system (2.1) with (without) a leader is
• locally finite-time stable if it is Lyapunov stable, and for any x(t0 ) ∈ D0 (D0 ⊂ D)
there exists a settling time t0 < T < ∞ such that

lim e(t) = 0,
t→T
e(t) = 0, ∀t ≥ T. (2.13)

• globally finite-time stable if (2.13) holds for any given initial state x(t0 ).
In the above, we have introduced the concepts of stability of networked cooperative
systems. Our task now is to introduce the Lyapunov analysis method to determine
stability.
2.4 Stability Analysis Theory on Cooperative Control Systems 33

Definition 2.9 ([8]) A continuously differentiable (C 1 ) function V : D → R is said

to be a Lyapunov function if it satisfies

V (0) = 0 and V (x) > 0 in D − {0} (2.14)

V̇ (x) ≤ 0. (2.15)

Lemma 2.19 ([8]) Consider system (2.1). If there exists a C 1 Lyapunov function
V (x) : D → R+ such that

V̇ (x) < 0 (2.16)

for x ∈ D − {0}, where V̇ (x) is the derivative along the system trajectory, i.e.,
V̇ (x) = ∂∂Vx (x) f (x), then V (x) → 0 as t → ∞, and system (2.1) is asymptotically
stable.

Lemma 2.20 ([8]) Consider system (2.1). If there exist a C 1 Lyapunov function
V (x) : D → R+ , and some real constants 0 < c, d < ∞, such that

V̇ (x) ≤ −cV (x) + d (2.17)

then system (2.1) is uniformly ultimately bounded (UUB).

Lemma 2.21 ([18]) Consider system (2.1). If there exists a C 1 Lyapunov function
V (x) : D → R+ , and some constants 0 < c < ∞ and 0 < α < 1, such that

V̇ (x) ≤ −cV (x)α (2.18)

then system (2.1) is finite-time stable for any given x(t0 ) ∈ D0 ⊂ D, in which the
finite settling time T ∗ satisfies

V (x0 )1−α
T∗ ≤ . (2.19)
c(1 − α)

Lemma 2.22 ([19]) Consider system (2.1). If there exist a C 1 Lyapunov function
V (x) : D → R+ , and also some positive constants c, d, and 0 < α < 1, such that

V̇ (x) ≤ −cV (x)α + d, (2.20)

then system (2.1) is stable within a finite settling time T ∗ satisfying

V (x0 )1−α
T∗ ≤ , (2.21)
cθ0 (1 − α)
34 2 Preliminaries

for any given x(t0 ) ∈ D0 ⊂ D, where 0 < θ0 < 1 is a real constant. Moreover, when
t ≥ T ∗ , the state trajectory convergent to a compact set,

d
Θ = x V (x) ≤ 1 . (2.22)
c(1 − θ0 ) α

References

1. Diestel, R.: Graph Theory. Springer, New York (2000)

2. Godsil, C., Royle, G.F.: Algebraic Graph Theory. Springer, New York (2001)
3. Olfati-Saber, R., Murray, R.: Consensus problems in networks of agents with switching topol-
ogy and time-delays. IEEE Trans. Autom. Control 49, 1520–1533 (2004)
4. Agaev, R., Chebotarev, P.: On the spectra of nonsymmetric Laplacian matrices. Linear Algebra
Appl. 399, 157–178 (2005)
5. Lafferriere, G., Williams, A., Caughman, J., Veerman, J.J.P.: Decentralized control of vehicle
formations. Syst. Control Lett. 54(9), 899–910 (2005)
6. Ren, W., Cao, Y.: Distributed Coordination of Multi-agent Networks: Emergent Problems,
Models, and Issues. Springer, London (2010)
7. Ren, W., Beard, R.W.: Distributed Consensus in Multi-vehicle Cooperative Control. Commu-
nications and Control Engineering. Springer, London (2008)
8. Khalil, H.K.: Nonlinear Systems, 3rd edn. Prentice Hall, Upper Saddle River (2002)
9. Alefeld, G., Schneider, N.: On square roots of M-matrices. Linear Algebra Appl. 42, 119–132
(1982)
10. Horn, R.A., Johnson, C.R.: Matrix Analysis. Cambridge University Press, New York (1985)
11. Horn, R.A., Johnson, C.R.: Topics in Matrix Analysis. Cambridge University Press, Cambridge
(1991)
12. Peng, K., Yang, Y.: Leader-following consensus problem with a varying-velocity leader and
time-varying delays. Phys. A 388(2–3), 193–208 (2009)
13. Hong, Y.G., Jiang, Z.P.: Finite-time stabilization of nonlinear systems with parametric and
dynamic uncertainties. IEEE Trans. Autom. Control 51(12), 1950–1956 (2006)
14. Yu, S., Yu, X., Shirinzadeh, B., Man, Z.: Continuous finite-time control for robotic manipulators
with terminal sliding mode. Automatica 41, 1957–1964 (2005)
15. Polycarpout, M.M., Ioannout, P.A.: A robust adaptive nonllinear control design. In: American
Control Conference, pp. 1365–1369 (1993)
16. Qian, C., Lin, W.: Non-Lipschitz continuous stabilizer for nonlinear systems with uncontrol-
lable unstable linearization. Syst. Control Lett. 42(3), 185–200 (2001)
17. Hardy, G., Littlewood, J., Polya, G.: Inequalities. Cambridge University Press, Cambridge
(1952)
18. Bhat, S., Bernstein, D.: Continuous finite-time stabilization of the translational and rotional
double integrators. IEEE Trans. Autom. Control 43(5), 678–682 (1998)
19. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Fault-tolerant finite time consensus for multiple
uncertain nonlinear mechanical systems under single-way directed communication interactions
and actuation failures. Automatica 63, 374–383 (2016)
 Part II
Infinite-Time Cooperative
Adaptive Control of Multiple Networked
Nonlinear Systems
Chapter 3
Lyapunov Analysis for Cooperative
Adaptive Consensus Under Undirected
Graph

For the distributed adaptive cooperative control of nonlinear multi-agent systems

subject to unknown time-varying gain and non-parametric uncertainties, it requires
the use of special Lyapunov functions that depend on the graph Laplacian and para-
metric estimation error in a certain way. The technical difficulties in the cooperative
adaptive controller design and stability analysis for such uncertain systems mainly
arise from: (1) the time-varying yet unknown control gain; (2) the non-parametric
uncertainties; and (3) the local communication constraints. One of the key approaches
to solve these difficulties is to select special Lyapunov functions for the cooperative
adaptive control design that depends on the graph topology and parameter estimation
error in special ways. In this chapter, we will show how to construct the Lyapunov
function that is related to the graph topology and parameter estimation error. In addi-
tion, some key concepts needed for the distributed adaptive controller design and
stability analysis on graph are introduced, such as the virtual parameter estimation
error technique; the core function concept; and distributed Lyapunov function design
dependent of the graph topology.
In Sect. 3.1, we formulate the leaderless consensus problem on a group of
networked dynamical systems subject to unknown time-varying gain and non-
parametric/non-vanishing uncertainties. Section 3.2 introduces the concepts of the
core function and the virtual parameter estimation error. Section 3.3 addresses the
problem of how to construct the Lyapunov function for the stability analysis of coop-
erative MAS in the presence of unknown time-varying gain and non-parametric/non-
vanishing uncertainties under fixed and undirected connected graph topology. The
constructed Lyapunov function mainly depends on the graph topology and the

Part of Sect. 3.2 has been reproduced from Song, Y. D., Huang, X. C., Wen, C. Y.: Tracking control
for a class of unknown nonsquare MIMO nonaffine systems: A deep-rooted information based robust
adaptive approach. IEEE Transactions on Automatic Control, vol. 61, no. 10, pp. 3227–3233, 2016
© 2016, IEEE, reprinted with permission.

© Springer Nature Switzerland AG 2019 37

Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_3
38 3 Lyapunov Analysis for Cooperative Adaptive Consensus Under Undirected Graph

parameter estimation errors, which consists of two parts: a distributed Lyapunov

function part constituted by the local neighborhood error and graph Laplacian matrix,
and an adaptive Lyapunov function part dependent of the parameter estimation errors.
Section 3.4 presents the basic results and key ideas of the adaptive distributed leader-
less consensus controller design and stability analysis for the continuous-time MAS
subject to unknown and non-parametric uncertainties under the undirected topology
condition. The importance of the key techniques described earlier, including the core
function, the virtual parameter estimation error, and the Lyapunov function design
is shown in Sect. 3.4. These studies open the way for many well-known results from
systems theory to be extended to the case of multi-agent systems on graphs.

3.1 Leaderless Consensus Formulation

Autonomous control of networked MAS has become an interesting topic for research
due to the potential for many important applications. Cooperative consensus is known
as the basic issue in the cooperative control of MAS, which is a study dedicated to
ensuring an agreement between states or output variables among networked MAS.
Consider a group of N networked agents subject to non-parametric uncertainties
modeled by

ẋi (t) = gi (·)u i (t) + f i (z i (t), t) + f di (z i (t), t) (3.1)

where xi ∈ R and u i ∈ R are the state and control input of the ith (i = 1, 2, . . . , N )
agent, respectively, gi is the control gain, f i denotes the system nonlinearities, f di
denotes
 the non-vanishing uncertainties including the external disturbance, and z i =
 x .
j∈N i i j
For ease of notation, it is assumed that the agents’ states are scalars (xi ∈ R), and
the case for the vector states (xi ∈ Rm , m > 1) can be derived similarly by using
standard methods involving the Kronecker products employed in various chapters in
the book.
It is assumed that the nonlinear dynamics f i (·) (i = 1, 2, . . . , N ) are unknown
and thus cannot be available for the controller design. Furthermore, f i (·) (i =
1, 2, . . . , N ) considered in this chapter are non-parametric uncertainties; that is,
f i (·) (i = 1, 2, . . . , N ) cannot be linearly parametric decomposition or cannot be
approximated by neural networks. For the control gain gi (·), it is assumed that the
control gain is unknown and possibly time-varying.

Remark 3.1 Note that each node has its own distinct dynamics, and both the node
nonlinearities f i (·) and the node non-vanishing uncertainties f di (·) are unknown and
non-identical. In addition, f i (·) and f d i(·) associated with the ith agent are not only
related to the state of the ith agent itself but also influenced by its neighbors; that
is, the coupling effects among the neighboring agents are considered in this model.
Furthermore, the impact of the parameter variation of the agent is explicitly reflected
3.1 Leaderless Consensus Formulation 39

in the model. Such variation literally leads to unknown and time-varying control
gains in the model, thus making the control design and stability analysis much more
involved as compared to the case of known and constant control gains.

The focus on this chapter is the controller design and stability analysis for the
cooperative leaderless consensus problem of nonlinear MAS (3.1), in which both the
unknown time-varying control gain and the non-parametric/non-vanishing modeling
uncertainties are simultaneously addressed.
For each agent in the networked system, only local state information relative to
its neighbors is obtainable, which means that, for node i, only local neighborhood
synchronization error is available for controller design. Thus, we introduce the local
neighborhood state error for node i as

ei = ai j (xi − x j ), i = 1, . . . , N . (3.2)
j∈N i

Let E = [e1 , . . . , e N ]T , X = [x1 , . . . , x N ]T such that E = L X .

3.2 Core Function and Virtual Parameter Estimation Error

In this section, we introduce the concepts of core function and virtual parameter
estimation error, both of which are crucial for the controller design and stability
analysis of the cooperative control of MAS with unknown time-varying control gain
and non-parametric uncertainties.
Note that for the system with constant control gain and with nonlinearities that can
be linearly parameterized or can be approximated by neural networks, there exists
some unknown finite constant, pi , and some known scalar function, φi (xi ), related
to the states variables such that the nonlinear term can be linearly parametric decom-
posed as f i (·) = pi φi (·), or there exists some “optimal” vector Wi ∈ Rl , activation
function vector φ(xi ) ∈ Rl , and bounded error ε(xi ) ∈ R (l denotes the number of
neural nodes), such that for xi in some compact set, f i (·) can be approximated by
neural networks as f i (·) = WiT φ(·) + ε(·). For such nonlinear systems, one can per-
form the standard adaptive control method by constructing the parameter estimation
error of the normal form •˜ = • − •ˆ , where •ˆ denotes the estimation of the unknown
constant parameter •, to obtain the system stability results. However, for the system
that involves time-varying control gain as well as non-parametric uncertainties, the
control problem becomes more challenging.
In the following, we introduce the concepts of core function and virtual parameter
estimation error, with aid of both of which the obstacles caused by the unknown time-
varying control gain and non-parametric uncertainties in the controller design and
stability analysis for cooperative control of MAS can be circumvented without the
need for linearization or approximation to the original systems.
40 3 Lyapunov Analysis for Cooperative Adaptive Consensus Under Undirected Graph

To proceed, we should specify the assumptions on the unknown time-varying con-

trol gain and the system non-parametric/non-vanishing uncertainties. The following
assumptions on system (3.1) are imposed.

Assumption 3.1 For system (3.1), it is assumed that the control gain gi (i =
1, 2, . . . , N ) is unknown, time-varying, and bounded away from zero; that is, there
exist unknown finite constants g i and ḡi such that 0 < g i ≤ |gi (·)| ≤ ḡi < ∞, and gi
is sign-definite (w. l. o. g. here sgn(gi ) = +1), which means that the control direction
is definite.

Assumption 3.2 For the non-parametric uncertainty fi (·) (i = 1, 2, . . . , N ), certain

crude structural information on non-parametric nonlinearity fi (·) is available to allow
an unknown constant c f i ≥ 0 and a known scalar function ϕi (z i ) to be extracted,
such that | f i (·)| ≤ c f i ϕi (·) for t ∈ [t0 , ∞), where ϕi (·) is bounded for any xi (i =
1, 2, . . . , N ). The non-vanishing uncertainty f di (·) (i = 1, 2, . . . , N ) is bounded by
some unknown finite constants θi ≥ 0 such that | f di (·)| ≤ θi < ∞.

Remark 3.2 (1) Assumption 3.2 is related to the extraction of the deep-rooted infor-
mation from the non-parametric nonlinearities f i (·), where the constant c f i , although
associated with the system parameters, bears no physical meaning, thus named as
virtual parameter. (2) We call ϕi (z i ) as “core function” [1] for it contains the deep-
rooted information of the system that is easily computable. For any practical system,
the core function (which is non-unique) can be readily obtained with certain crude
structural information on f i (·), as seen from the later part. (3) To extract the core
function ϕi (z i ) from the f i (·), one can perform upper normalization on f i (·) [2].
Although this might render the virtual parameter overestimated, excessive control
effort is avoided because the corresponding control algorithm does not use such
parameter directly, as seen later.

Remark 3.3 By using the concept of deep-rooted (core) function, non-parametric

uncertainties and time-varying parameters in the system can be gracefully handled.
For instance, for the lumped uncertain function of the form
−|hxi | 2
f i (·) = ρi (t) cos(axi ) + xi sin(bt + d xi2 ) + (xi ζi (t) + ei ) (3.3)

where |ρi (t)| ≤ ρ̄ < ∞ and |ζi (t)| ≤ ζ̄ < ∞ are two unknown and time-varying
parameters, and a, b, d, and h are unknown constants. Clearly, it is impossible to
carry out parametric decomposition to separate the parameters ρi (t), a, b, d, ζi (t), and
h from f i (·). However, it is effortless to obtain the deep-rooted function of the form
ϕi (z i ) = 1 + |xi | + xi2 independent of those parameters, such that f i (·) ≤ c f i ϕi (z i ),
with c f i being the virtual parameter defined by c f i = max{ρ̄ + 1, 2ζ̄ + 1, ζ̄ 2 }. Note
that ϕi (z i ) is readily computable and non-unique.

It is interesting to note that the virtual parametric decomposition of the lumped

uncertainty f i (·) as imposed in Assumption 3.2 does exist if f i (·) can be dealt with
by NN-based method as stated in the following fact.
3.2 Core Function and Virtual Parameter Estimation Error 41

Fact 1: If the lumped uncertainty f i (·) can be approximated by neural networks,

then Assumption 3.2 holds naturally.
Proof of Fact 1: According to the universal approximation capability property [3],
there exist some “optimal” vector Wi ∈ Rl , activation function vector φi (z i ) ∈ Rl ,
and bounded error function εi (z i ) ∈ R (l denotes the number of neural nodes), such
that for z i in some compact set,

f i (z i ) = WiT φi (z i ) + εi (z i ) (3.4)

it then follows that

| f i (z i )| ≤ Wi φi (z i ) + |εi (z i )| ≤ Wi φi (z i ) + ε̄i = c f i ϕi (z i ) (3.5)

where ε̄i is an unknown finite constant satisfying |εi (z i )| ≤ ε̄i , c f i = max{Wi , ε̄i }
is the virtual parameter needed to be estimated and ϕi (z i ) = 1 + φi (z i ) is the core
function which is computable.
The significant implications of Fact 1 are threefold: (1) Assumption 3.2 is reason-
able and justifiable; (2) any nonlinear systems with uncertainties that can be dealt
with by NN-based technique can also be handled by the proposed method; (3) if little
information on f i (·) is available to extract the deep-rooted function from the lumped
uncertainties, one can use the combination of the well-known basis functions as the
core function as defined above.
Also, there exist certain situations that might pose challenges to NN-based approx-
imation technique, as noted in the following fact. However, such case can be handled
by the core function method gracefully.
Fact 2: If f i (z i ) is unbounded, it is logically impossible to globally approximate
such unbounded function with a bounded function of the form WiT φi (z i ) + εi (z i )
anymore. In fact, any unbounded function cannot be represented (approximated) by
a bounded function over the entire domain of interest. Instead, it is only possible
to approximate such unbounded function with WiT φi (z i ) + εi (z i ) in which Wi is
unknown bounded but φi (z i ) and εi (z i ) are bounded only if z i is bounded.
In the following, we introduce the concept of the virtual parameter estimation
error [4]. Upon using g (the lower bound of the control gain gi ), the virtual parameter
estimation error is constructed as

•˜ = • − g •ˆ . (3.6)

Note that such defined parameter estimation error is in contrast to the regular error
of the form •˜ = • − •ˆ . This error is named as virtual parameter estimation error. By
introducing it into a novel chosen Lyapunov function, the difficulty associated with
the unknown and time-varying control gain can be handled effectively.
42 3 Lyapunov Analysis for Cooperative Adaptive Consensus Under Undirected Graph

3.3 Lyapunov Function Design Under Undirected Topology

The cooperative adaptive control design for nonlinear networked multi-agent sys-
tems with unknown uncertainties requires the use of special Lyapunov function that
depends on the graph topology and the parameter estimation error in a certain way. In
the design of admissible adaptive controllers, only distributed control protocols and
distributed adaptive laws are permitted. It is not straightforward to develop adaptive
laws for cooperative MAS on graphs that only require information from that agent
and its neighbors. The key to solving this problem is to select special Lyapunov
function for adaptive control design dependent of the graph topology and parameter
estimation errors in specific ways.
It is noted that the Lyapunov function is essential for the Lyapunov stability analy-
sis of the adaptive leaderless consensus control of the nonlinear MAS with unknown
time-varying gain and non-parametric uncertainties. In this section, we show how to
construct the Lyapunov function based on the graph Laplacian, neighborhood error,
and parametric estimation error. The Lyapunov function in the adaptive cooperative
control of nonlinear MAS usually consists of two parts: the distributed part related
to the neighborhood error and the Laplacian matrix, and the adaptive part dependent
of the parameter estimation error.

3.3.1 The Distributed Part of Lyapunov Function

In this subsection, we show how to construct the distributed part of the Lyapunov
function.
We first introduce the definition related to uniform ultimate boundedness (UUB)
for cooperative control systems.
Definition 3.1 ([5, 6]) For the error E(t) ∈ R N , if there exists a compact set Ω ∈
R N with the property that {0 N } ⊂ Ω, so that ∀E(t0 ) ∈ Ω, there exist a finite positive
constant C and a finite time t f (C, X (t0 )), both independent of t0 > 0, such that
for any t ≥ t0 + t f , E(t) ≤ C, then we say E(t) is cooperative uniform ultimate
boundedness (CUUB).
Suppose that the network communication among the N agents is modeled by a
directed graph G , with the weighted adjacency matrix A = [ai j ] and the Laplacian
matrix L (see Chap. 2 for the details on graph theory).
The following results of graph theory are useful for establishing leaderless con-
sensus stability results of the MAS in this chapter.
Lemma 3.1 ([7]) For an undirected and connected graph G , the Laplacian L of
G has only one zero eigenvalue associated with the eigenvector 1 N , i.e., L 1 N = 0,
and all the other eigenvalues of L are positive real constants. If 1TN X = 0, then
X T L X ≥ λ2 (L )X T X , where λ2 (L ) denotes the second minimum eigenvalue of
L.
3.3 Lyapunov Function Design Under Undirected Topology 43

 N([7]) The vector L X , with X = [x1 , . . . , x N ] ∈ R , is a column stack

T N
Lemma 3.2
vector of j=1 ai j (xi − x j ), i = 1, . . . , N . If graph G is undirected (and hence L
N N
is symmetric), then X T L X = 21 i=1 j=1 ai j (x i − x j ) . If the undirected graph
2

G is connected, then L X = 0 N or X L X = 0 if and only if xi = x j , for all i, j =

1, . . . , N .

Lemma 3.3 ([8]) The Laplacian matrix L associated with an undirected connected
graph G is symmetric and semi-positive definite, that is, L = L T ∈ R N ×N and
L ≥ 0. Furthermore, the null space N (L ) = {X |X T L X = 0} = span{1 N }.

Proof We first show that L is symmetric and semi-positive definite. Since the graph
G is undirected, it is straightforward that the adjacency matrix A is symmetric,
implying that L is symmetric. According to Lemma 3.1, the Laplacian matrix L
associated with an undirected and connected graph has only one zero eigenvalue
with all its other eigenvalues being positive real constants; thus, L is semi-positive
definite, and we also derive that the null space of L is N (L ) = span{1 N }. According
to Fact 8.15.2 in [9], if L = L T ∈ R N ×N satisfying L ≥ 0 or L ≤ 0, then the null
space is N (L ) = {X |X T L X = 0}. Therefore, the null space is equal to N (L ) =
{X |X T L X = 0} = span{1 N }. 

According to Lemma 3.2 or 3.3, the leaderless consensus (CUUB consensus)

is achieved if and only if E = L X = 0 (E = L X  ≤ C) or X T L X = 0
(X T L X  ≤ C) according to Definition 3.1. Thus, the objective of the leader-
less consensus (CUUB consensus) is to make E = 0 (E ≤ C) or X T L X = 0
(X T L X  ≤ C).
According to the above analysis, we next construct the first part of Lyapunov
function candidate.
Before moving on, let us first introduce two matrices: Λ̃ = diag{a, λ2 , . . . , λ N }
and Λ = U NT Λ̃−1 U N , where a is an arbitrarily positive constant, and U N is an orthogo-
nal matrix such that L = U NT diag{0, λ2 , . . . , λ N }U N = U NT Λ0 U N with
Λ0 = diag{0, λ2 , . . . , λ N } provided by Lemma 3.1.
Such matrix Λ is a symmetric and positive definite matrix, with which we construct
the first part of Lyapunov function as

V1 = E T ΛE. (3.7)

By recalling that E = L X , it then holds that

V1 = X T L T (U NT Λ̃−1 U N )L X
= X T (U NT Λ0 U N )T (U NT Λ̃−1 U N )(U NT Λ0 U N )X
= X T U NT Λ0 Λ̃−1 Λ0 U N X
= X T U NT Λ0 U N X
= X T L X. (3.8)
44 3 Lyapunov Analysis for Cooperative Adaptive Consensus Under Undirected Graph

It is seen from (3.7) that if V1 → 0 (V1 ∈ L ∞ ), then E → 0 (E ∈ L ∞ ). The dis-

tributed controller can be designed to make the distributed Lyapunov function con-
structed as in (3.7) to converge to zero (be bounded) such that the neighborhood error
E converges to zero (is bounded), which further implies that the leaderless consensus
(CUUB consensus) is achieved by Lemma 3.2 or 3.3 according to Definition 3.1.

3.3.2 The Adaptive Part of Lyapunov Function

In this subsection, we show how to construct the adaptive part of Lyapunov function
available for the cooperative adaptive control of nonlinear MAS with unknown time-
varying control gain and non-parametric uncertainties.
With the aid of the virtual parameter estimation error introduced in Sect. 3.2, we
construct the adaptive part of the Lyapunov function candidate as

1 T −1 1
V2 = •˜ Γ •˜ = (• − g •ˆ )T Γ −1 (• − g •ˆ ) (3.9)
2g 2g

in which Γ = diag{σ1 , . . . , σ N } > 0 is a design parameter matrix. The derivative of

the adaptive part of the Lyapunov function candidate is taken as

V̇2 = (• − g •ˆ )T Γ −1 (−•˙ˆ ). (3.10)

With such defined adaptive part of Lyapunov function, the unknown time-varying
gain and non-parametric uncertainties can be handled gracefully without the need
for linearization or approximation to the original systems.
The next section shows how the graph Laplacian matrix and virtual parameter
estimation error works in the Lyapunov stability analysis for the distributed leaderless
consensus control of nonlinear MAS with unknown non-parameter uncertainties.

3.4 Lyapunov Analysis for Adaptive Leaderless Consensus

In this section, we take the Lyapunov stability analysis for adaptive leaderless consen-
sus of system (3.1) subject to unknown time-varying control gain and non-parametric
uncertainties as an example to show how the Lyapunov function works in the Lya-
punov stability analysis.
The distributed adaptive controller based on the signum function is designed as
follows:

u i = −ki ai j (xi − x j ) + u ci , i = 1, . . . , N , (3.11)
j∈N i
3.4 Lyapunov Analysis for Adaptive Leaderless Consensus 45

where u ci is a compensation control term

u ci = −sgn(ei )ĉ f i ϕi − sgn(ei )θ̂i , (3.12)

with the updated law

ĉ˙ f i = γ1i ϕi |ei |, θ̂˙i = γ2i |ei |, (3.13)

and ki , γ1i , γ2i > 0 are finite design constant parameters, ĉ f i and θ̂i are the estima-
tions of the virtual parameters c f i and θi (c f i and θi are defined as in Assumption
3.2), respectively, and ϕi (·) is the readily computable scalar function as given in
Assumption 3.2.
Next, we provide the Lyapunov stability analysis to show how the distributed part
of Lyapunov function and adaptive part of Lyapunov function work, respectively.

Theorem 3.1 With the distributed control law given in (3.11) together with the
compensation control term (3.12) and the adaptive law (3.13), system (3.1) under
Assumptions 3.1–3.2 and an undirected connected graph topology G achieves asymp-
totically stable consensus. Further, the internal signals including the control input
signal, the neighborhood error, and parameter estimation error in the system remain
uniformly bounded.

Proof The proof consists of three steps.

Step 1. Introducing the Laplacian matrix into the distributed part of Lyapunov
function.
We first construct the following distributed Lyapunov function as

1 T
V1 = E ΛE (3.14)
2

where Λ = U NT Λ̃−1 U N is given in Sect. 3.3.1. By recalling that E T ΛE = X T L X

derived in (3.8), we take the time derivative of V1 along (3.1) and get


N
V̇1 = X T L Ẋ = E T Ẋ = ei (gi u i + f i + f di ). (3.15)
i=1

By performing the control law proposed in (3.11) and the compensation term (3.12),
(3.15) then becomes


N
V̇1 = ei [gi (−ki ei + u ci ) + f i + f di ]
i=1


N  
=− ki gi ei2 + ei gi (−sgn(ei )ĉ f i ϕi − sgn(ei )θ̂i ) + f i + f di
i=1
46 3 Lyapunov Analysis for Cooperative Adaptive Consensus Under Undirected Graph


N 
N
≤− ki gi ei2 + [−g ĉ f i ϕi |ei | − g θ̂i |ei | + c f i ϕi |ei | + θi |ei |]
i=1 i=1


N N 
 
=− ki gi ei2 + (c f i − g ĉ f i )ϕi |ei | + (θi − g θ̂i )|ei | . (3.16)
i=1 i=1

Step 2. Introducing the virtual parameter estimation error into the adaptive part
of Lyapunov function.
Note that (3.16) involves the parameter estimation error of the form •˜ = • − g •ˆ ;
we then introduce such virtual parameter estimation error into the adaptive part of
Lyapunov function candidate as

1 T −1 1 T −1
V2 = C̃ f Γ1 C̃ f + θ̃ Γ2 θ̃ (3.17)
2g 2g

with
C̃ f = C f − g Ĉ f , θ̃ = θ − g θ̂ (3.18)

where C̃ f = [c̃ f 1 · · · c̃ f N ]T ∈ R N , C f = [c f 1 · · · c f N ]T ∈ R N , Ĉ f = [ĉ f 1 · · ·

ĉ f N ]T ∈ R N , θ̃ = [θ̃1 · · · θ̃ N ]T ∈ R N , θ = [θ1 · · · θ N ]T ∈ R N , θ̂ = [θ̂1 · · · θ̂ N ]T ∈
R N , Γ1 = diag{γ1i } ∈ R N ×N , and Γ2 = diag{γ2i } ∈ R N ×N . Such treatment allows
the unknown time-varying parameters involved in the system to be processed grace-
fully, as can be seen shortly.
Upon using the updated law (3.13), the time derivative of V2 is taken as

V̇2 = C̃ Tf Γ1−1 (−Ĉ˙ f ) + θ̃ T Γ2−1 (−θ̂˙ )

   
 ĉ˙ f i  θ̂˙i
N N
= c̃ f i − + θ̃i −
γ1i γ2i
i=1 i=1


N 
N
= (c f i − g ĉ f i )ϕi |ei | + (θi − g θ̂i )|ei |. (3.19)
i=1 i=1

Step 3. Constructing the Lyapunov function candidate.

We choose the Lyapunov function candidate as

V = V1 + V2 . (3.20)

By combining (3.16) and (3.19), we then get the derivative of V as


N
V̇ =V̇1 + V̇2 ≤ − ki gi ei2 ≤ −kgE2 ≤ 0 (3.21)
i=1
3.4 Lyapunov Analysis for Adaptive Leaderless Consensus 47

where k = min{k1 , . . . , k N }. From (3.21), we see that E2 ≤ −V̇ (t)/kg, which
then implies that

∞ ∞
V̇ (t) 1 1
E(t)2 dt ≤ − dt = − (V (∞) − V (0)) ≤ V (0) < ∞
0 0 kg kg kg
(3.22)

By recalling the fact that E is uniformly continuous, which, together with (3.22),
implies that limt→∞ E(t) = limt→∞ L X (t) = 0 upon using the Barbalat lemma.
According to Lemma 3.2, we see that L X = 0 if and only if xi = x j for all i, j =
1, . . . , N . Therefore, xi = x j for all i, j = 1, . . . , N as t → ∞, which means that
the asymptotically stable leaderless consensus is achieved under the proposed control
scheme (3.11)–(3.13). In addition, from (3.21), we can deduce that V ∈ L ∞ , which
further ensures that E ∈ L ∞ , C̃ ∈ L ∞ , and θ̃ ∈ L ∞ . From the definition of C̃ and θ̃,
it is straightforward that Ĉ ∈ L ∞ and θ̂ ∈ L ∞ , which further implies that U ∈ L ∞ .
It is then concluded that the control input, the neighborhood errors, and the param-
eter estimation errors in the system remain uniformly bounded under the proposed
control. 

References

1. Song, Y.D., Huang, X.C., Wen, C.Y.: Tracking control for a class of unknown nonsquare MIMO
nonaffine systems: a deep-rooted information based robust adaptive approach. IEEE Trans.
Autom. Control 61(10), 3227–3233 (2016)
2. Chen, Z., Huang, J.: Dissipativity, stabilization, and regulation of cascade-connected systems.
IEEE Trans. Autom. Control 49(5), 635–650 (2004)
3. Hornik, K., Stinchcombe, M., White, H.: Multilayer feedforward networks are universal approx-
imators. Neural Netw. 2(5), 359–366 (1989)
4. Wang, Y.J., Song, Y.D., Lewis, F.L.: Robust adaptive fault-tolerant control of multi-agent systems
with uncertain non-identical dynamics and undetectable actuation failures. IEEE Trans. Ind.
Electron. 62(6), 3978–3988 (2015)
5. Zhang, H., Lewis, F.L.: Adaptive cooperative tracking control of higher-order nonlinear systems
with unknown dynamics. Automatica 48, 1432–1439 (2012)
6. Shen, Q., Jiang, B., Shi, P., Jun, Z.: Cooperative adaptive fuzzy tracking control for networked
unknown nonlinear multi-agent systems with time-varying actuator faults. IEEE Trans. Fuzzy
Syst. 22(3), 494–504 (2014)
7. Olfati-Saber, R., Murray, R.: Consensus problems in networks of agents with switching topology
and time-delays. IEEE Trans. Autom. Control 49, 1520–1533 (2004)
8. Zhang, H., Lewis, F.L., Qu, Z.: Lyapunov, adaptive, and optimal design techniques for coop-
erative systems on directed communication graphs. IEEE Trans. Ind. Electron. 59, 3026–3041
(2012)
9. Bernstein, D.: Matrix Mathematics: Theory, Facts, and Formulas, 2nd edn. Princeton University
Press, Princeton (2009)
Chapter 4
Cooperative Adaptive Consensus Control
for Uncertain Multi-agent Systems with
nth-Order Dynamics Under Undirected
Graph

In Chap. 3 we showed how to construct the Lyapunov function that is related to

the graph topology and parameter estimation error, and also introduced some key
concepts needed for the distributed adaptive controller design and stability analysis
on graph such as the virtual parameter estimation error technique, the core function
concept, and the distributed Lyapunov function design dependent of the graph topol-
ogy. In this chapter we address the controller design and stability analysis for the
distributed leaderless consensus control of networked multi-agent systems, including
first-order systems, second-order systems and high-order systems, all with unknown
time-varying gain and non-parametric/non-vanishing uncertainties under the undi-
rected communication topology condition. Each agent is mathematically modeled
by a continuous nonlinear system, and the communication network among agents
is modeled by a graph. Of particular interest is the development of the distributed
adaptive consensus control protocol capable of simultaneously compensating time-
varying yet unknown control gain and non-parametric/non-vanishing uncertainties
under the local communication constraints.
Section 4.1 addresses the adaptive leaderless consensus problem for first-order
continuous-time MAS with unknown time-varying gain and non-parametric/non-
vanishing uncertainties under undirected topology based on continuous control
method. The basic structure and key ideas of the distributed adaptive control pro-
tocols to confront the leaderless consensus problems emerge from this study. To
understand the relationships between the distributed adaptive controller design for
first-order systems and for general high-order systems, it was natural to study the
second-order systems and then high-order systems. Sections 4.2 and 4.3 study the
adaptive leaderless cooperative consensus for the second-order position–velocity
continuous-time uncertain systems and general high-order continuous-time uncertain

Part of Sects. 4.2 and 4.4 have been reproduced from Wang, Y. J., Song, Y. D., Lewis, F. L.: Robust
adaptive fault-tolerant control of multi-agent systems with uncertain non-identical dynamics and
undetectable actuation failures. IEEE Transactions on Industrial Electronics, vol. 62, no. 6, pp.
3978–3988, 2015  c 2015, IEEE, reprinted with permission.

© Springer Nature Switzerland AG 2019 49

Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_4
50 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

dynamical systems on graph, respectively. The control schemes proposed in this

chapter are user-friendly in that there is no need for detailed dynamic information
of the agent in controller design and implementation, resulting in structurally sim-
ple and computationally inexpensive solutions for the leaderless consensus problem
of networked uncertain multi-agent systems. Section 4.4 gives a discussion on the
cooperative adaptive consensus control of MAS with unknown time-varying gain
and non-parametric/non-vanishing uncertainties.

4.1 Cooperative Adaptive Control of First-Order

Uncertain Systems Under Undirected Topology

In this section, we address the leaderless consensus problem for first-order MAS with
unknown time-varying gain and non-parametric uncertainties on graph based on the
principles of core function, virtual parameter estimation error, and Lyapunov stability
theory introduced in Chap. 3. To help with the understanding of the fundamental idea
and technical development of the proposed method, we start with controller design
for the first-order MAS, followed by the extension to the second-order and high-order
case in later sections. This section formulates the basic structure of the distributed
control schemes to confront the leaderless consensus problems and show how to learn
the unknown non-parametric uncertainties online to ensure the leaderless consensus
by using distributed adaptive control technique on graph.

4.1.1 System Dynamics

We consider a group of N networked agents with the same dynamics as in (3.1)

ẋi (t) = gi (·)u i (t) + f i (z i (t), t) + f di (z i (t), t) (4.1)

where xi ∈ R and u i ∈ R are the state and control input of the ith (i = 1, 2, . . . , N )
agent, respectively, gi is the control gain, f i and f di
 are the system nonlinearities
and non-vanishing uncertainties, respectively; z i = j∈N i  i x j .
Assumption 4.1
a. gi (i = 1, 2, . . . , N ) is unknown, time-varying, and bounded away from zero;
that is, there exist unknown finite positive constants g i and ḡi such that 0 < g i ≤
|gi (·)| ≤ ḡi < ∞, and gi is sign-definite (w. l. o. g. here sgn(gi ) = +1).
b. Certain crude structural information on f i (z i (t), t) is available to allow an
unknown constant c f i ≥ 0 and a known scalar function ϕi (z i ) to be extracted
[2], such that | f i (z i (t), t)| ≤ c f i ϕi (z i ) for t ∈ [t0 , ∞), where ϕi (z i ) is bounded
for any xi (i = 1, 2, . . . , N ). There exists unknown finite positive constant θi
such that | f di (z i (t), t)| ≤ θi < ∞.
4.1 Cooperative Adaptive Control of First-Order Uncertain Systems … 51

The local neighborhood error for node i is


ei = ai j (xi − x j ), i = 1, . . . , N . (4.2)
j∈N i

4.1.2 Controller Design and Stability Analysis

We consider the controller design and stability analysis for the first-order uncer-
tain MAS (4.1). It should be stressed that the control scheme proposed in Sect. 3.4
for the same system model (4.1), although capable of dealing with the unknown
time-varying gain and non-parametric/non-vanishing uncertainties, might cause the
undesired chattering phenomenon when the consensus error crosses zero due to
the utilization of the sign function. In this subsection, we propose a continuous
control solution not only to compensate the unknown time-varying gain and non-
parametric/non-vanishing uncertainties but also to avoid the undesirable chatting
phenomenon.
Before moving on, we first give a useful lemma as follows.

Lemma 4.1 ([1]) For a real diagonal matrix M with appropriate dimension, if
vectors X and Y satisfy that each component of their product X T Y is nonnegative,
it then holds that
λmin (M)X T Y ≤ X T MY ≤ λmax (M)X T Y

with λmax (M) and λmin (M) denoting the maximum and minimum eigenvalues of M,
respectively.

The continuous distributed adaptive control scheme is proposed as follows,


u i = −ki ai j (xi − x j ) + u ci , i = 1, . . . , N (4.3)
j∈N i

where u ci is a compensation control term given as

ĉ f i ϕi2 ei θ̂i ei
u ci = − − , (4.4)
ϕi |ei | + ε |ei | + ε

with the updated law

γ1i ϕi2 ei2 γ2i ei2

ĉ˙ f i = −γ1i σ1i ĉ f i + , θ̂˙i = −γ2i σ2i θ̂i + , (4.5)
ϕi |ei | + ε |ei | + ε

ki , γ1i , γ2i , σ1i , σ2i > 0 are finite design constant parameters, ε is a small positive
constant, ĉ f i and θ̂i are the estimations of the virtual parameters c f i and θi (c f i and
52 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

θi are defined as in Assumption 4.1), respectively, and ϕi (·) is the readily computable
scalar function as given in Assumption 4.1.
The main result in this subsection is given as follows.

Theorem 4.1 Suppose that the graph topology G is undirected and connected. The
MAS (4.1) under Assumption 4.1 with the distributed control law proposed in (4.3)
and (4.5) achieves the CUUB leaderless consensus. In addition, the internal system
signals including the control input signal, the position error, and the parameter
estimation error remain uniformly bounded.

Proof Choose the Lyapunov function candidate as

V = V1 + V2 (4.6)

with

1  c̃ f i 1  θ̃i2
N 2 N
1 T
V1 = E ΛE, V2 = + (4.7)
2 2g γ1i 2g γ2i
i=1 i=1

where Λ = U NT Λ̃−1 U N is given the same as in Sect. 3.3.1, c̃ f i = c f i − g ĉ f i and

θ̃i = θi − g θ̂i denote the virtual parameter estimation error.
Taking the time derivative of V1 along (4.1), one gets


N
V̇1 = X T L Ẋ = E T Ẋ = ei (gi u i + f i + f di ). (4.8)
i=1

By performing the control law u i proposed in (4.3) and the compensation unit u ci
given in (4.4), the time derivative of V1 then becomes


N
V̇1 = ei [gi (−ki ei + u ci ) + f i + f di ]
i=1
 

N 
N
ĉ f i ϕi2 ei2 θ̂i ei2
≤− ki gi ei2 + −gi − gi + |ei |c f i ϕi + |ei |θi
i=1
ϕi |ei | + ε |ei | + ε
i=1


N N 
 ĉ f i ϕi2 ei2 θ̂i ei2
≤ − kg ei2 + −g −g
i=1
ϕi |ei | + ε |ei | + ε
i=1

c f i ϕi ei + c f i ϕi |ei |ε θi ei + θi |ei |ε
2 2 2
+ + (4.9)
ϕi |ei | + ε |ei | + ε

ϕi |ei | |ei |
By noting that ϕi |ei |+ε
≤ 1 and |ei |+ε
≤ 1, we then arrive at
4.1 Cooperative Adaptive Control of First-Order Uncertain Systems … 53
 

N 
N  ϕi2 ei2  ei2
V̇1 ≤ − kg ei2 + c f i − g ĉ f i + θi − g θ̂i + c f i ε + θi ε
ϕi |ei | + ε |ei | + ε
i=1 i=1
(4.10)

On the other hand, upon using the updated laws for ĉ f i and θ̂i designed in (4.5), one
gets the derivative of V2 as

 ĉ˙ f i  θ̂˙i
N N
V̇2 = (c f i − g ĉ f i )(− ) + (θi − g θ̂i )(− )
γ1i γ2i
i=1 i=1


N
ϕi2 ei2  N
ei2
= (c f i − g ĉ f i )(σ1i ĉ f i − )+ (θi − g θ̂i )(σ2i θ̂i − ).
ϕi |ei | + ε |ei | + ε
i=1 i=1

(4.11)

By combining (4.10) and (4.11), we then get the derivative of the Lyapunov function
candidate V as


N 
N 
N 
N
V̇ =V̇1 + V̇2 ≤ −kg ei2 + (c f i ε + θi ε) + σ1i c̃ f i ĉ f i + σ2i θ̃i θ̂i .
i=1 i=1 i=1 i=1

(4.12)

By using the relation that ab = 1

2
(a + b)2 − a 2 − b2 , we then have

1 1 2
c̃ f i ĉ f i = c̃ f i (c f i − c̃ f i ) = c − c̃2f i − (c f i − c̃ f i )2 ,
g 2g f i
1 1
θ̃i θ̂i = θ̃i (θi − θ̃i ) = θ 2 − θ̃i2 − (θi − θ̃i )2 . (4.13)
g 2g i

Inserting (4.13) into (4.12), we arrive at


N
V̇ ≤ − kg E T E + (c f i + θi )ε
i=1

1  1  1  1 
N N N N
− σ1i c̃2f i + σ1i c2f i − σ2i θ̃i2 + σ2i θi2 . (4.14)
2g 2g 2g 2g
i=1 i=1 i=1 i=1

By recalling that matrix Λ is positive and definite, it then holds

λmin (Λ)E T E ≤ E T ΛE ≤ λmax (Λ)E T E. (4.15)

54 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

Denote by

σ 1 = min{σ11 , . . . , σ1N }, σ 2 = min{σ21 , . . . , σ2N }, (4.16)

1  
N
d= σ1i c2f i + σ2i θi2 ∈ L ∞ .
2g
i=1

Upon using (4.14), (4.15) and (4.16) is represented as

σ  2 σ  2
N N
1
V̇ ≤ − kg E T ΛE − 1 c̃ f i − 2 θ̃i + d (4.17)
λmax (Λ) 2g 2g
i=1 i=1

Denote by

γ 1 = min{γ11 , . . . , γ1N }, γ 2 = min{γ21 , . . . , γ2N }. (4.18)

It then follows from (4.17) that

2kg 1 T σ 1γ 1 
N
c̃2f i σ 2γ 2 
N
θ̃i2
V̇ ≤ − ·E ΛE − − + d. (4.19)
λmax (Λ) 2 2g γ1i 2g γ2i
i=1 i=1

By letting
 
2kg
c = min , σ 1γ 1, σ 2γ 2 , (4.20)
λmax (Λ)

we then represent (4.19) as

V̇ ≤ − cV + d. (4.21)

From (4.21), we can conclude that the set = {(ei , c̃ f i , θ̃i )|V ≤ d/c} is globally
attractive. Once (ei , c̃ f i , θ̃i ) ∈
/ , then V̇ < 0. Therefore, there exists a finite time
t ∗ such that (ei , c̃ f i , θ̃i ) ∈ for ∀t > t ∗ . This further implies, together with (4.15),
that
 
√ 2 2d
E = ET E ≤ V1 ≤ (4.22)
λmin (Λ) λmin (Λ)c

when t ≥ t ∗ , meaning that the leaderless error E is CUUB.

In the sequel, we prove all the internal signals in the system are uniformly bounded.
By solving (4.21), it follows that
4.1 Cooperative Adaptive Control of First-Order Uncertain Systems … 55

d
V (t) ≤ exp−ct V (0) + ∈ L∞ (4.23)
c

for all t ≥ 0, from which we can deduce that E ∈ L ∞ , ĉ f i ∈ L ∞ and θ̂i ∈ L ∞ for
all i = 1, . . . , N . Therefore, U ∈ L ∞ i.e., the control input remains bounded. This
completes the proof. 

4.1.3 Simulation Example

This section gives a simulation example to illustrate the effectiveness of the dis-
tributed adaptive control scheme proposed in (4.3)–(4.5). It is shown that under the
proposed control a group of agents subject to unknown time-varying control gain,
unknown non-parametric uncertainties, and unknown bounded non-vanishing dis-
turbances achieves leaderless consensus successfully.
We use four agents to demonstrate the effectiveness of our proposed control
scheme. The communication graph among the four agents is presented in Fig. 4.1,
which satisfies the assumption that the graph topology is undirected and connected.
Each edge weight is set to be 1.
The dynamics of the four agents is of the following form,

ẋ1 =(5 + 0.5 exp−|x1 | )u 1 + 0.1 exp−x1 −x2 x1 + f d1 ,

ẋ2 =(5 + 0.5 exp−|x2 | )u 2 + 0.1 exp−x1 −x2 x22 + f d2 ,
2

ẋ3 =(5 + 0.5 exp−|x3 | )u 3 + 0.1 exp−x2 −x3 x32 + f d3 ,

2

ẋ4 =(5 + 0.5 exp−|x4 | )u 4 + 0.1 exp−x3 −x4 x4 + f d4 , (4.24)

in which the non-vanishing uncertainties are of the form: f d1 = 0.02 sin(0.05t),

f d2 = 0.01 cos(0.06t), f d3 = 0.02 cos(0.06t), and f d4 = 0.01 sin(0.05t).
The earlier theoretical analysis has declared that the nonlinear system (4.24) can
achieve CUUB leaderless consensus under the control scheme proposed in (4.3)–
(4.5). To set up the proposed controller, we only need to select the design parameters

Fig. 4.1 Communication

graph topology among the
four agents
56 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

Fig. 4.2 The position

trajectories xi of all the four
agents

Fig. 4.3 The neighborhood

errors ei of the four agents

Fig. 4.4 The control inputs

u i of the four agents

ki , γi , and σi (i = 1, 2, 3, 4) and does not need to take any time-consuming analytical

derivation for choosing design parameters. The design parameters in this simulation
are taken as: ki = 5 (i = 1, 2, 3, 4), γi = 0.01 and σi = 0.01 (i = 1, 2), ε = 0.01. In
addition, the core functions ϕi (·) (i = 1, 2, 3, 4) is easily taken as ϕ1 = |x1 |, ϕ2 = x22 ,
ϕ3 = x32 , and ϕ4 = |x4 | according to the core information of the system.
Figure 4.2 presents the position trajectories of all the four agents, and Fig. 4.3
shows the neighborhood position synchronization errors of all the four agents, both
of which verify that the position synchronization is achieved with small residual
errors in some finite time. Figure 4.4 shows the control input signals of the four
agents u i (i = 1, 2, 3, 4), from which we observe that the control input signals of all
4.1 Cooperative Adaptive Control of First-Order Uncertain Systems … 57

Fig. 4.5 The parameter

estimation ĉ f i of the four
agents

Fig. 4.6 The parameter

estimation θ̂i of the four
agents

the four agents are smooth and bounded. Figures 4.5 and 4.6 portray the parameter
estimation values of the virtual estimation parameters c f i and θi (i = 1, 2, 3, 4),
respectively, from both of which we see that all the parameter estimation values, ĉ f i
and θ̂i (i = 1, 2, 3, 4), are bounded under the proposed control algorithm.

4.2 Second-Order Systems

In this section, we address the leaderless consensus problem for the second-order
nonlinear multi-agent systems subject to non-parametric nonlinearities and unknown
bounded non-vanishing uncertainties.
Consider a group of N agents with scalar second-order dynamics

ẋi (t) = vi (t)

v̇i (t) = gi (z i , t)u i (t) + f i (z i , t) + f di (z i , t) i = 1, . . . , N (4.25)

where xi ∈ R, vi ∈ R, and u i ∈ R denote the position state, velocity state, and control
input of the ith agent, respectively, gi is the control gain, possibly time-varying and
unavailable for controller design, f i denotes the system nonlinearity, and f di denotes
58 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

the non-vanishing uncertainty

 including the external disturbance acting on the ith
agent. In addition, z i = j∈N i  i {x j , v j }.
Suppose that gi , f i , and f di satisfy Assumption 4.1. Of particular interest in this
section is to design distributed adaptive controller for the second-order nonlinear
multi-agent system (4.25) to ensure the leaderless consensus and to give a strict
stability analysis to demonstrate the stability of the system.

4.2.1 Consensus Error Dynamics

Let the local neighborhood state error still be defined as in (4.2). To solve the leader-
less consensus problem for the second-order multi-agent systems, we need to intro-
duce two filtered variables ξi and si (i = 1, . . . , N ) as

ξi = ẋi + βxi , si = ėi + βei , (4.26)

where β is a positive constant chosen by the designer. Let ξ = [ξ1 , . . . , ξ N ]T ∈ R N

and S = [s1 , . . . , s N ]T R N such that S = Lξ .
With the filtered variable ξ , the consensus error dynamics for system (4.25) can
be rewritten as a compact form,

ξ̇ = Ẍ + β Ẋ = GU + F + Fd + βv (4.27)

where X = [x1 , . . . , x N ]T ∈ R N , v = [v1 , . . . , v N ]T ∈ R N , G = diag{gi } ∈ R N ,

U = [u 1 , . . . , u N ]T ∈ R N , F = [ f 1 , . . . , f N ]T ∈ R N , and Fd = [ f d1 , . . . , f d N ]T
∈ RN .

4.2.2 Controller Design and Stability Analysis

In this subsection, we address the controller design and stability analysis for system
(4.25).
The controller for the second-order multi-agent system (4.25) is designed as

u i = −ki si + u ci , i = 1, . . . , N (4.28)

where the compensation term u ci is given as

ĉi ψi2 si θ̂i si

u ci = − − , (4.29)
ψi |si | + ε |si | + ε
4.2 Second-Order Systems 59

the updated law is designed as

γ1i ϕi2 si2 γ2i si2

ĉ˙i = −γ1i σ1i ĉi + , θ̂˙i = −γ2i σ2i θ̂i + , (4.30)
ψi |si | + ε |si | + ε

and ki , γ1i , γ2i , σ1i , and σ2i are finite positive design parameters, ĉi and θ̂i are
the estimation values of the virtual parameters ci and θi (ci = max{c f i , β} and θi
is defined by | f di | ≤ θi given as in Assumption 4.1), respectively, ψi (·) = ϕi (·) +
|vi (t)| with ϕi (·) given as in Assumption 4.1, and ε > 0 is a small finite constant.
The main result in this subsection is given as follows.

Theorem 4.2 Suppose that the graph topology G is undirected and connected, and
gi , f i , and f di satisfy Assumption 4.1. The second-order nonlinear multi-agent system
(4.25) with the distributed control scheme (4.28), the compensation control term
(4.29), and the updated control law (4.30) achieves CUUB leaderless consensus. In
addition, all the internal system signals remain bounded.

Proof With the positive definite matrix Λ = U NT Λ̃−1 U N given in Sect. 3.3.1, we
construct the distributed Lyapunov function as

1 T
V1 = S ΛS. (4.31)
2
By following the similar procedure as in the proof in (3.8), we readily obtain that

1 T 1
V1 = S ΛS = ξ T Lξ. (4.32)
2 2
Taking the derivative of V1 along the error dynamic (4.27) yields


N
V̇1 = S T ξ̇ = S T (GU + F + Fd + βv) = (gi u i + f i + f di + βvi ) (4.33)
i=1

From Assumptions 4.1, we see that | f i | ≤ c f i ϕi and | f di | ≤ θi , and we further have

that | f i + βvi | ≤ c f i ϕi + β|vi | ≤ ci ψi with ci = max{c f i , β} and ψi = ϕi + |vi |.
Upon using the control law (4.28) and the compensating unit u ci (4.29), one then
gets the time derivative of V1 as


N
V̇1 = si (gi u i + f i + f di + βvi )
i=1


N
= si [gi (−ki si + u ci ) + f i + f di + βvi ]
i=1
60 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …


N 
N
ĉi ψi2 si2 θ̂i si2
≤− ki gi si2 + [−gi − gi + |si |ci ψi + |si |θi ]
i=1
ψi |si | + ε |si | + ε
i=1


N 
N
ψi2 si2
≤ − kg si2 + [(ci − g ĉi )
i=1
ψi |si | + ε
i=1

si2
+ (θi − g θ̂i ) + ci ε + θi ε]. (4.34)
|si | + ε

Next, we construct the adaptive part of the Lyapunov function as

1  ci2 1  θi2
N N
V2 = + . (4.35)
2g i=1 γ1i 2g i=1 γ2i

By applying the updated law ĉ˙i and θ̂˙i given in (4.30), one gets the derivative of V2
as

 ĉ˙i  θ̂˙i
N N
V̇2 = (ci − g ĉi )(− ) + (θi − g θ̂i )(− )
γ1i γ2i
i=1 i=1


N
ϕi2 si2  N
si2
= (ci − g ĉi )(σ1i ĉi − )+ (θi − g θ̂i )(σ2i θ̂i − ). (4.36)
ϕi |si | + ε |si | + ε
i=1 i=1

Now, we choose the Lyapunov function candidate as

V = V1 + V2

with V1 and V2 being given in (4.32) and (4.35), respectively. By combining (4.34)
and (4.36), we thus have the derivative of V as


N 
N 
N 
N
V̇ =V̇1 + V̇2 ≤ −kg si2 + (ci ε + θi ε) + σ1i c̃i ĉi + σ2i θ̃i θ̂i . (4.37)
i=1 i=1 i=1 i=1

Upon using the inequality ab = 1

2
(a + b)2 − a 2 − b2 , we have

1 2 1
c̃i ĉi = ci − c̃i2 − (ci − c̃i )2 , θ̃i θ̂i = θ 2 − θ̃i2 − (θi − θ̃i )2 . (4.38)
2g 2g i

By inserting (4.38) into (4.37), we arrive at

4.2 Second-Order Systems 61


N
V̇ ≤ − kgS T S + (ci + θi )ε
i=1

1 
N
1 
N
1 
N
1 
N
− σ1i c̃i2 + σ1i ci2 − σ2i θ̃i2 + σ2i θi2
2g 2g 2g 2g
i=1 i=1 i=1 i=1

kg σ1 
N
σ2 
N
≤− S T ΛS − c̃i2 − θ̃i2
λmax (Λ) 2g 2g
i=1 i=1


N
1 
N
1 
N
+ (ci + θi )ε + σ1i ci2 + σ2i θi2
2g 2g
i=1 i=1 i=1

2kg 1 T σ 1γ 1 
N
c̃i2 σ 2γ 2 
N
θ̃i2
≤− ·
S ΛS − −
λmax (Λ) 2 2g γ1i 2g γ2i
i=1 i=1


N
1 
N
1 
N
+ (ci + θi )ε + σ1i ci2 + σ2i θi2 , (4.39)
2g 2g
i=1 i=1 i=1

where σ 1 = min{σ11 , . . . , σ1N }, σ 2 = min{σ21 , . . . , σ2N }, γ 1 = min{γ11 , . . . ,

2kg N
γ1N }, and γ 2 = min{γ21 , . . . , γ2N }. Let c = min{ λmax (Λ) , σ 1 γ 1 , σ 2 γ 2 }, d = i=1
1 N
(ci + θi )ε + 2g i=1
(σ1i ci2 + σ2i θi2 ). We then have from (4.39) that

V̇ ≤ − cV + d. (4.40)

Similar to the analysis given in the proof of Theorem 4.1, we readily derive that
 
√ 2 2d
S = ST S ≤ V1 ≤ (4.41)
λmin (Λ) λmin (Λ)c

after some finite time t ∗ , which further implies that E and Ė are bounded after some
finite time t ∗ . Therefore, E and Ė are CUUB.
In the sequel, we prove that all the internal system signals remain bounded. By
solving (4.40), it is deduced that

d
V (t) ≤ exp−ct V (0) + ∈ L∞ (4.42)
c

for all t ≥ 0, which then implies that S ∈ L ∞ , E ∈ L ∞ , Ė ∈ L ∞ , ĉi ∈ L ∞ , and

θ̂i ∈ L ∞ for all i = 1, . . . , N . Therefore, U ∈ L ∞ ; that is, the control input remains
bounded. 
62 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

4.2.3 Simulation Example

In this section, we demonstrate the effectiveness of the distributed adaptive control

algorithm proposed in (4.28)–(4.30) by a simulation example. For convenience, in
the simulation example, we still use four agents with the communication topology
among the agents being the same as in Sect. 4.1.3, as shown in Fig. 4.7. Each edge
weight is set to be 1.
The four agents are modeled by the following dynamics,

ẋi =vi ,
v̇i =gi u i + f i + f di , i = 1, 2, 3, 4 (4.43)

with

v̇1 =(5 + 0.5 exp−|x1 | )u 1 + l sin(v1 ) + l exp−v1 −x2 v1 + f d1 ,

v̇2 =(5 + 0.5 exp−|x2 | )u 2 + l sin(v2 ) + l exp−x1 −v2 v22 + f d2 ,
2

v̇3 =(5 + 0.5 exp−|x3 | )u 3 + l sin(v3 ) + l exp−x2 −v3 v32 + f d3 ,

2

v̇4 =(5 + 0.5 exp−|x4 | )u 4 + l sin(v4 ) + l exp−x3 −v4 v4 + f d4

where the non-vanishing uncertainties are: f d1 = 0.02 sin(0.05t), f d2 = 0.01

cos(0.06t), f d3 = 0.02 cos(0.06t), and f d4 = 0.01 sin(0.05t).
In the simulation, only the design parameters ki , γi , and σi (i = 1, 2, 3, 4) are
needed to be selected. They are chosen as: ki = 5 (i = 1, 2, 3, 4), γi = 0.1 and σi =
0.1 (i = 1, 2), ε = 0.01 and β = 1. In addition, according to the core information
of the system, the core functions ϕi (·) (i = 1, 2, 3, 4) are easily selected as ϕ1 =
1 + |v1 |, ϕ2 = 1 + v22 , ϕ3 = 1 + v32 , and ϕ4 = 1 + |v4 |.
Figures 4.8 and 4.9 show the position and velocity trajectories of all the four
agents, respectively. Figures 4.10 and 4.11 present the neighborhood position errors
and velocity synchronization errors of all the four agents, which verify that system
(4.43) achieves both the position synchronization and velocity synchronization with
small residual errors under the control algorithm proposed in (4.28). Figure 4.12

Fig. 4.7 Communication

graph topology between the
four agents
4.2 Second-Order Systems 63

Fig. 4.8 The position

trajectories xi of the four
agents

Fig. 4.9 The velocity

trajectories vi of the four
agents

Fig. 4.10 The position

neighborhood errors ei of the
four agents

Fig. 4.11 The velocity

neighborhood errors ėi of the
four agents
64 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

Fig. 4.12 The control inputs

u i of the four agents

Fig. 4.13 The parameter

estimation ĉi of the four
agents

Fig. 4.14 The parameter

estimation θ̂i of the four
agents

shows the control input signals u i (i = 1, 2, 3, 4) of the four agents, from which
we observe that all the control input signals of the four agents remain continuous
and bounded. Figures 4.13 and 4.14 represent the parameter estimation values of the
virtual parameters ci and θi (i = 1, 2, 3, 4), respectively, from both of which we see
that all the parameter estimation values ĉi and θ̂i (i = 1, 2, 3, 4) are bounded under
the proposed control scheme.
4.3 High-Order Systems 65

4.3 High-Order Systems

In this section, we address the leaderless consensus problem for high-order nonlinear
multi-agent systems subject to non-parametric uncertainties and unknown bounded
external disturbances.
We consider a group of N agents with scalar high-order dynamics

ẋq,i (t) = xq+1,i (t), q = 1, . . . , n + 1,

ẋn (t) = gi (·)u i (t) + f i (z i (t), t) + f di (z i (t), t), i = 1, . . . , N , (4.44)

where xq,i ∈ R (q = 1, . . . , n) and u i ∈ R are the qth state and control input of the ith
agent, respectively, gi is the control gain, f i denotes the system
 nonlinearity, and f di
denotes the non-vanishing uncertainty. In addition, z i = j∈N i  i {x1, j , . . . , xn, j }.
Suppose that gi , f i and f di satisfy Assumption 4.1. Of particular interest in this
section is to design distributed adaptive controller for the high-order nonlinear multi-
agent system (4.44) with unknown time-varying gain and non-parametric uncertain-
ties to make the system achieve consensus. We also give strict theoretical analysis
and numerical simulation to demonstrate and verify the effectiveness of the proposed
control algorithm.

4.3.1 Consensus Error Dynamics

Let the local neighborhood state error be defined as in (4.2). To solve the leaderless
consensus problem for high-order multi-agent systems, we introduce two filtered
variables ζi and ωi (i = 1, . . . , N ) as

ζi = β1 x1,i + β2 x2,i + · · · + βn−1 xn−1,i + xn,i , i = 1, . . . , N ,

ωi = β1 e1,i + β2 e2,i + · · · + βn−1 en−1,i + en,i , i = 1, . . . , N , (4.45)

where βq (q = 1, . . . , n − 1) are some positive constants chosen by the designer such

that the polynomial r n−1 + βn−1 r n−2 + · · · + β1 is Hurwitz. Let ζ = [ζ1 , . . . , ζ N ]T
∈ R N and ω = [ω1 , . . . , ω N ]T ∈ R N such that ω = L ζ .
With the filtered variable ζ , the consensus error dynamics for system (4.44) is
rewritten as the following compact form

ζ̇ = β1 X 2 + β2 X 3 + · · · + βn−1 X n + Ẋ n
= GU + F + Fd + β1 X 2 + β2 X 3 + · · · + βn−1 X n (4.46)

with X q = [xq,1 , . . . , xq,N ]T ∈ R N (q = 1, . . . , n), G = diag{gi } ∈ R N ×N ,

U = [u 1 , . . . , u N ]T ∈ R N , F = [ f 1 , . . . , f N ]T ∈ R N , and Fd = [ f d1 , . . . , f d N ]T ∈
RN .
66 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

4.3.2 Controller Design and Stability Analysis

In this subsection, we conduct the controller design and stability analysis for system
(4.44).
The controller for the n-order (n ≥ 2) multi-agent system (4.44) is designed as

u i = −ki ωi + u ci , i = 1, . . . , N (4.47)

where the compensation term u ci is given as

b̂i φi2 ωi θ̂i ωi

u ci = − − , (4.48)
φi |ωi | + ε |ωi | + ε

and the updated law is designed as

γ1i φi2 ωi2 γ2i ωi2

b̂˙i = −γ1i σ1i b̂i + , θ̂˙i = −γ2i σ2i θ̂i + , (4.49)
φi |ωi | + ε |ωi | + ε

where ki , γ1i , γ2i , σ1i , and σ2i are finite positive design parameters, ε > 0 is a small
finite constant, b̂i and θ̂i are the estimation values of the virtual parameters bi and θi
with

bi = max{c f i , β1 , . . . , βn−1 }, | f di | ≤ θi (4.50)

and

φi (·) = ϕi (·) + |x2,i (t)| + · · · + |xn,i (t)| (4.51)

with ϕi (·) being a scalar and readily computable function given in Assumption 4.1.
The main result in this subsection is given as follows.

Theorem 4.3 Suppose that the graph topology G is undirected and connected, and
gi , f i , and f di satisfy Assumption 4.1. The n-order (n ≥ 2) nonlinear multi-agent
system (4.44) with the distributed control scheme (4.47), the compensation control
term (4.48), and the updated control law (4.49) achieves CUUB leaderless consensus.
In addition, all the internal system signals remain bounded.

Proof With the positive definite matrix Λ = U NT Λ̃−1 U N given in Sect. 3.3.1, we
construct the distributed Lyapunov function as

1 T
V1 = ω Λω (4.52)
2
By following the similar procedure as in the proof in (3.8), we can get
4.3 High-Order Systems 67

1 T 1
V1 = ω Λω = ζ T L ζ. (4.53)
2 2
Taking the derivative of V1 along the error dynamic (4.46) yields

V̇1 = ω T ζ̇ = ω T (GU + F + Fd + β1 X 2 + β2 X 3 + · · · + βn−1 X n )


N
= ωi (gi u i + f i + f di + β1 x2,i + β2 x3,i + · · · + βn−1 xn,i ) (4.54)
i=1

Denote by bi = max{c f i , β1 , . . . , βn−1 } and φi = ϕi + |x2,i | + · · · + |xn,i |, we then

get from Assumptions 4.1 that

| f i + β1 x2,i + β2 x3,i + · · · + βn−1 xn,i |

≤c f i ϕi + β1 |x2,i | + β2 |x3,i | + · · · + βn−1 |xn,i |
≤bi φi (4.55)

Upon using the control law u i given in (4.47) and the compensating unit u ci given
in (4.48), one gets the time derivative of V1 as


N
 
V̇1 = ωi gi u i + f i + f di + β1 x2,i + β2 x3,i + · · · + βn−1 xn,i
i=1
N
= ωi gi (−ki ωi + u ci ) + f i + f di + β1 x2,i + β2 x3,i + · · · + βn−1 xn,i
i=1
 

N 
N
b̂i φi2 ωi2 θ̂i ωi2
≤− ki gi ωi2 + −gi − gi + |ωi |bi φi + |ωi |θi
φi |ωi | + ε |ωi | + ε
i=1 i=1

N 


N  φi2 ωi2  ωi2
≤ − kg ωi2 + bi − g b̂i + θi − g θ̂i + bi ε + θi ε .
φi |ωi | + ε |ωi | + ε
i=1 i=1
(4.56)

Next, we construct the adaptive part of the Lyapunov function as

1 T −1 1 T −1
V2 = B̃ Γ1 B̃ + θ̃ Γ2 θ̃ (4.57)
2g 2g

with B̃ = [b̃1 , . . . , b̃ N ]T , θ̃ = [θ̃1 , . . . , θ̃ N ]T , Γ1 = diag{γ1i }, and Γ2 = diag{γ2i }. By

applying the updated laws b̂˙i and θ̂˙i given in (4.49), one then gets the derivative of
V2 as
68 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …
   
b̂˙i θ̂˙i
N 
 N 

V̇2 = bi − g b̂i − + θi − g θ̂i −
γ1i γ2i
i=1 i=1

N 
   N 
  
φi2 ωi2 ωi2
= bi − g b̂i σ1i b̂i − + θi − g θ̂i σ2i θ̂i − .
φi |ωi | + ε |ωi | + ε
i=1 i=1

(4.58)

Now, we construct the Lyapunov function candidate as

V = V1 + V2

with V1 and V2 being given in (4.53) and (4.57), respectively. The derivative of V
thus follows from (4.56) and (4.58) that


N 
N 
N 
N
V̇ =V̇1 + V̇2 ≤ −kg ωi2 + (bi ε + θi ε) + σ1i b̃i b̂i + σ2i θ̃i θ̂i . (4.59)
i=1 i=1 i=1 i=1

Upon using the inequality, ab = 1

2
(a + b)2 − a 2 − b2 , we have

1 1
b̃i b̂i = b2 − b̃i2 − (bi − b̃i )2 , θ̃i θ̂i = θ 2 − θ̃i2 − (θi − θ̃i )2 . (4.60)
2g i 2g i

By inserting (4.60) into (4.59), we arrive at,


N
V̇ ≤ − kgω ω + (bi + θi )ε
T

i=1

1  1  1  1 
N N N N
− σ1i b̃i +
2
σ1i bi −
2
σ2i θ̃i +
2
σ2i θi2
2g 2g 2g 2g
i=1 i=1 i=1 i=1

1  1 
N N
1
≤ − kg ω Λω −
T
σ1i b̃i −
2
σ2i θ̃i2
λmax (Λ) 2g 2g
i=1 i=1


N
1 
N
1 
N
+ (bi + θi )ε + σ1i bi +
2
σ2i θi2
2g 2g
i=1 i=1 i=1

2kg 1 T σ 1γ 1 
N
b̃i2 σ 2γ 2 
N
θ̃i2
≤− · ω Λω − −
λmax (Λ) 2 2g γ1i 2g γ2i
i=1 i=1


N
1 
N
1 
N
+ (bi + θi )ε + σ1i bi +
2
σ2i θi2 , (4.61)
2g 2g
i=1 i=1 i=1
4.3 High-Order Systems 69

where σ 1 = min{σ11 , . . . , σ1N }, σ 2 = min{σ21 , . . . , σ2N }, γ 1 = min{γ11 , . . . ,

2kg
γ1N }, and γ 2 = min{γ21 , . . . , γ2N }. Let c = min{ λmax (Λ) , σ 1 γ 1 , σ 2 γ 2 },
N 1 N
d = i=1 (bi + θi )ε + 2g i=1 (σ1i bi + σ2i θi ). We then have from (4.61) that
2 2

V̇ ≤ − cV + d. (4.62)

Similar to the analysis given in the proof of Theorem 4.1, we readily derive that
 
√ 2 2d
ω = ωT ω ≤ V1 ≤ (4.63)
λmin (Λ) λmin (Λ)c

after some finite time t ∗ , which further implies that all the qth (q = 1, . . . , n) neigh-
borhood error E q are bounded after some finite time t ∗ by the definition of ω according
to Lemma 3 in [3]. Therefore, all E q (q = 1, . . . , n) are CUUB.
In the sequel, we prove all the internal system signals are bounded. By solving
(4.62), it is deduced that V (t) ≤ exp−ct V (0) + dc ∈ L ∞ for all t ≥ 0, which then
implies that ω ∈ L ∞ , E q ∈ L ∞ (q = 1, . . . , n), b̂i ∈ L ∞ and θ̂i ∈ L ∞ for all i =
1, . . . , N . Hence, U ∈ L ∞ ; i.e., the control input is bounded. This completes the
proof. 

4.4 Discussion on Cooperative Adaptive Consensus of MAS

with Non-parametric Uncertainties

1. Note that each node in systems (4.1), (4.25), and (4.44) has its own distinct
dynamics, and both the node nonlinearities f i (·) and the node non-vanishing
uncertainties (including the external disturbances) f di (·) are unknown and non-
identical. In addition, f i (·) and f di (·) associated with the ith agent are not only
related to the state of the ith agent itself but also influenced by its neighbors; this
is in contrast to most existing works that ignore the coupling effects among the
neighboring agents.
2. The impact of the parameter variation of the agent is explicitly reflected in the
models (4.1), (4.25), and (4.44). Note that such variation literally leads to unknown
and time-varying control gains in the model, thus making the control design and
stability analysis much more involved as compared to the case of known and
constant control gains [3–11], and unknown yet constant gains [12].
3. It should be stressed that in practice it would be very difficult, if not impossible,
to obtain the exact values of those bounds involved in Assumptions 4.1. In this
work, the control scheme is developed without the need for analytical estimation
of such bounds, although the fact that those bounds do exist is used for stability
analysis.
70 4 Cooperative Adaptive Consensus Control for Uncertain Multi-agent Systems …

4. The control scheme as given in (4.3) and (4.4) consists of a robust control and
an adaptive control; the later contains two crucial parameters that are adaptively
adjusted during the system operation. The updating algorithm (4.5) has a constant
decay term to prevent the estimated parameter from drifting. The role of the small
constant ε in the denominator is to ensure the continuity of the control action.
The reason for including the other term in the algorithm is to accommodate the
lumped uncertainties in the model to guarantee CUUB consensus, as seen in the
stability analysis.
5. In building the compensation control u ci , the adaptively adjusted parameters ĉ f i
and θ̂i rather than c f i and θi are used, thus avoiding the conservative and excessive
control effort. In particular, during the start-of stage, the initial control effort from
u ci can be made as small as zero by setting ĉi (0) = 0 and θ̂i (0) = 0, which can
be chosen freely by the designer.
6. In developing the control schemes, a number of virtual parameters such as c f i
and θi are defined and used in the stability analysis, but those parameters are not
involved in the control algorithms, thus analytical estimation of those parameters
(a nontrivial task) is not needed in setting up and implementing the proposed
control strategies.

References

1. Wang, Y.J., Song, Y.D., Lewis, F.L.: Robust adaptive fault-tolerant control of multi-agent
systems with uncertain non-identical dynamics and undetectable actuation failures. IEEE Trans.
Ind. Electron. 62(6), 3978–3988 (2015)
2. Song, Y.D., Huang, X.C., Wen, C.Y.: Tracking control for a class of unknown nonsquare MIMO
nonaffine systems: A deep-rooted information based robust adaptive approach. IEEE Trans.
Autom. Control 61(10), 3227–3233 (2016)
3. Zhang, H., Lewis, F.L.: Adaptive cooperative tracking control of higher-order nonlinear systems
with unknown dynamics. Automatica 48, 1432–1439 (2012)
4. Su, H., Chen, G., Wang, X., Lin, Z.: Adaptive second-order consensus of networked mobile
agents with nonlinear dynamics. Automatica 47, 368–375 (2011)
5. Yu, H., Xia, X.: Adaptive consensus of multi-agents in networks with jointly connected topolo-
gies. Automatica 48, 1783–1790 (2012)
6. Mei, J., Ren, W., Chen, J., Ma, G.: Distributed adaptive coordination for multiple Lagrangian
systems under a directed graph without using neighborsvelocity information. Automatica 49,
1723–1731 (2013)
7. Das, A., Lewis, F.L.: Distributed adaptive control for synchronization of unknown nonlinear
networked systems. Automatica 46, 2014–2021 (2010)
8. Zhang, H., Lewis, F.L., Qu, Z.: Lyapunov, adaptive, and optimal design techniques for coop-
erative systems on directed communication graphs. IEEE Trans. Indust. Elec. 59, 3026–3041
(2012)
9. Li, J., Kumar, K.D.: Decentralized fault-tolerant control for satellite attitude synchronization.
IEEE Trans. Fuzzy Syst. 20, 572–586 (2012)
10. Chen, G., Song, Y.D.: Fault-tolerant output synchronisation control of multi-vehicle systems.
IET Control Theory Appl. 8(8), 574–584 (2014)
References 71

11. Shen, Q., Jiang, B., Shi, P., Jun, Z.: Cooperative adaptive fuzzy tracking control for networked
unknown nonlinear multi-agent systems with time-varying actuator faults. IEEE Trans. Fuzzy
Syst. 22(3), 494–504 (2014)
12. Zhou, B., Wang, W., Ye, H.: Cooperative control for consensus of multi-agent systems with
actuator faults. Comput. Elect. Eng. 40(7), 2154–2166 (2014)
Chapter 5
Cooperative Adaptive Consensus
for Multi-agent Systems Under Directed
Topology

In Chap. 4, we studied the controller design and stability analysis for the cooperative
adaptive leaderless consensus control problem of networked multi-agent systems,
including first-order systems, second-order systems, and high-order systems, all with
unknown time-varying gain and non-parametric/non-vanishing uncertainties under
the undirected communication topology condition. Considering that the networked
communication among the subsystems is not always bidirectional in practice, it is
necessary to take the one-way directed communication condition into consideration.
In comparison with the undirected topology condition, the one-way directed com-
munication topology obviously complicates the underlying problem significantly.
The technical difficulties mainly arise from the asymmetry of the Laplacian matrix
associated with the directed topology graph. In this chapter, we attempt to address the
distributed leaderless consensus control problem of networked multi-agent systems
under local and one-way directed communication conditions.
The key to tackle the technical difficulty arising from the directed communica-
tion constraints depends on how to construct the Lyapunov function on graph. In
Sect. 5.1, we derive some results related to the Laplacian matrix, which is crucial
to construct the Lyapunov function suitable for the leaderless consensus stability
analysis, and show how to construct the special Lyapunov function on graph. In
Sect. 5.2, we address the leaderless cooperative consensus control problem for first-
order continuous-time multi-agent systems subject to uncertainties under directed
topology. The basic results and key ideas of cooperative control under directed topol-
ogy emerge from this study. Section 5.3 investigates the leaderless consensus control
problem of nonlinear multi-agent systems with second-order dynamics in the pres-
ence of unknown time-varying gain and non-parametric/non-vanishing uncertain-
ties. The focus of this section is the controller design and stability analysis for the
cooperative consensus of second-order uncertain multi-agent systems, in which both
unknown time-varying gain and non-parametric/non-vanishing modeling uncertain-
ties should be compensated and the directed topology condition should be addressed
© Springer Nature Switzerland AG 2019 73
Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_5
74 5 Cooperative Adaptive Consensus for Multi-agent Systems …

simultaneously. Section 5.4 gives a simulation example to illustrate the effectiveness

of the distributed control scheme proposed in Sect. 5.3 for the second-order uncertain
multi-agent systems under directed topology.

5.1 Graph Laplacian Potential Under Directed Topology

Graph Laplacian potential was introduced in [1] for undirected graph with 0–1 adja-
cency elements. Later, it was extended to weighted undirected graphs [2]. In this
section, we extend the concept of Laplacian potential to directed graphs and show
how the Lyapunov function is related to the directed graph Laplacian and the neigh-
borhood error. In the next section, we show how to apply such Laplacian potential
in the Lyapunov analysis of the leaderless consensus problems for nonlinear MAS
under directed topology.
Before moving on, we first present several useful lemmas. First, it is important
to note that for directed communication topology the Laplacian matrix L is no
longer symmetric in contrast to the case associated with undirected graph where
the Laplacian matrix is symmetric, and semi-positive definite with one simple zero
eigenvalue. This imposes technical challenge in constructing Lyapunov function. To
circumvent the technical difficulty in constructing Lyapunov function arising from
asymmetric property of the original Laplacian matrix L under directed graph, a
new matrix Q is introduced that exhibits the property of symmetry and semi-positive
definiteness with one simple zero eigenvalue, as stated in the following Lemma.

Lemma 5.1 Suppose that the digragh G is strongly connected. Let p = [ p1 , p2 ,

. . . , pn ]T be the left eigenvector of L associated with the zero eigenvalue. Define

Q = (diag( p)L + L T diag( p)), (5.1)

where diag( p) = diag{ p1 , . . . , p N } ∈ R N ×N > 0, then Q is a Laplacian matrix cor-

responding to a connected undirected graph topology. More specifically, Q is sym-
metric, positive semi-definite and has a simple zero eigenvalue with all the other
eigenvalues positive and real.

Proof It is worth noting that in the work by [3] it has been shown that Q is sym-
metric and positive semi-definite. Here we further reveal that such Q is a Laplacian
corresponding to a connected undirected graph and thus has only one simple zero
eigenvalue.  
First, note that p T L = 01×N , which implies that Nj=1 ai j pi = Nj=1 a ji p j for
all i, j ∈ {1, . . . , N }, and then it is derived that
5.1 Graph Laplacian Potential Under Directed Topology 75

Q = diag( p)L + L T diag( p)

⎡ ⎤

N
⎢ (a1 j p1 + a j1 p j ) · · · −a1N p1 − a N 1 p N ⎥
⎢ j=1 ⎥
⎢ ⎥
⎢ −a21 p2 − a12 p1 · · · −a2N p2 − a N 2 p N ⎥
=⎢⎢ .. .. .. ⎥
⎥ (5.2)
⎢ . . . ⎥
⎢ ⎥
⎣ N ⎦
−a N 1 p N − a1N p1 · · · (a N j p N + a j N p j )
j=1

Let z i j = ai j pi + a ji p j (i, j ∈ {1, . . . , N }) and Z = [z i j ]. It is interesting to note

that since pi > 0 ([4]), we have z i j > 0 ⇔ ai j > 0 or a ji > 0, and z ii = 0 ⇔
aii = 0. Note that A = [ai j ] is the weighted adjacency matrix of the strongly con-
nected digraph G , and moreover, z i j = z ji . It is thus seen that Z so defined is sym-
metric and is a weighted adjacency matrix associated with a connected undirected
graph according to the definition of the adjacency matrix in graph theory. Define
bi = Nj=1 z i j and B = diag(b1 , . . . , b N ) ∈ R N ×N . It is straightforward from (5.2)
that Q = B − Z , and it thus holds that Q is a Laplacian of a connected undirected
graph topology by the Laplacian definition, that is, Q is symmetric, positive semi-
definite and has a simple zero eigenvalue with all the other eigenvalues positive and
real. 

The significance of the feature on Q (defined in (5.1)) as revealed in Lemma

5.1 is that the original directed graph problem (where the Laplacian matrix L is
non-symmetric, thus imposing significant challenge for consensus design because
one cannot use such L to construct Lyapunov function directly) can be treated as
an undirected graph problem (where the newly defined matrix Q is equivalent to the
Laplacian with the feature as stated in Lemma 5.1). Furthermore, such feature allows
for the development of the following two lemmas that are crucial for the construction
of the distributed part of Lyapunov function subject to directed graph.

Lemma 5.2 For a strongly connected directed network with Laplacian L , let Q be
defined as in Lemma 5.1, then ∀x = 0 N ,

x T Qx = 0 (5.3)

if and only if

x = c1 N (5.4)

x T Qx
where c = 0 is a constant. Moreover, min xT x
exists and
x=c1 N

N
x T Qx
0 < min ≤ λi (Q). (5.5)
x=c1 N xT x i=2
76 5 Cooperative Adaptive Consensus for Multi-agent Systems …

Proof According to Lemma 5.1, Q is a Laplacian matrix of a connected undi-

rected graph. Let Ξ be the diagonal matrix associated with Q, that is, there exists
an orthogonal matrix R = (r1 , r2 , . . . , r N ) such that Q = RΞ R T . Let y = R T x =
[y1 , . . . , y N ]T . Then

N
x T Qx = x T RΞ R T x = y T Ξ y = λi (Q)yi2
i=1
=0· y12 + λ2 (Q)y22 + · · · + λ N (Q)y N2 , (5.6)

where λi (Q) (i = 1, . . . , N ) denotes the eigenvalue of Q. From (5.6), it can be

concluded that for ∀x = 0 N (i.e., ∀y = 0 N ),

x T Qx = y T Λy = 0 (5.7)

if and only if

y1 = 0,
(5.8)
yi = 0, i = 2, . . . , N

Note that r1 , r2 , . . . , r N are the N eigenvectors of Q associated with the N different

eigenvalues λ1 (Q), λ2 (Q), · · · , λ N (Q), respectively. Thus, R N = span{r1 , r2 , . . . ,
r N }, and moreover, ri ⊥ r j (i = j, i, j = 1, . . . , N ). There exist some constants ai
(i = 1, . . . , N ) such that

x = a1 r 1 + a2 r 2 + · · · + a N r N (5.9)

and then

yi = riT x = riT (a1r1 + a2 r2 + · · · + a N r N ) = ai riT ri . (5.10)

Thus, the condition in (5.8) is equal to

a1r1T r1 = 0, a1 = 0,
⇐⇒
ai riT ri = 0, i = 2, . . . , N ai = 0, i = 2, . . . , N

which implies that

x = a1 r 1 . (5.11)

Note that 1 N is the eigenvector of Q associated with the simple zero eigenvalue, i.e.,
λ1 (Q), therefore r1 = k1 N , with k being a nonzero constant, and then

x = a1 k1 N = c1 N (5.12)
5.1 Graph Laplacian Potential Under Directed Topology 77

where c = a1 k. Thus, ∀x = 0 N ,

x T Qx
min ≤ min x T Qx = min yT Ξ y
x=c1 N xT x x=c1 N ,x T x=1 x=c1 N ,y T y=1
N
≤ min λi (Q)yi2 ≤ λ2 (Q). (5.13)
x=c1 N ,y T y=1,y3 =···=y N =0
i=1

On the other hand, ∀x = 0 N , since x = c1 N , then x T Qx = 0, which, together with

the fact that Q is a positive semi-definite, implies x T Qx > 0, and therefore (5.5)
holds. 
Upon using Lemma 5.2, the following result can be established.
Lemma 5.3 ∀E = 0 N , there exists a constant km > 0 such that

ET QE
≥ km . (5.14)
ET E
N
Proof Note that p T E = p T L X = 0, i.e., i=1 pi ei = 0, and pi > 0 according
to [4], from which we know that for E = 0 and i = 1, . . . , N , it is impossible that
sgn(ei ) = 1 (or sgn(ei ) = −1). Therefore, E = c1 N with c being a nonzero constant.
T
According to Lemma 5.2, there exists a constant km = min EE TQEE > 0 such that
E=0 N
ET QE
ET E
≥ km > 0. 
According to the above analysis, we next construct the first part of Lyapunov
function candidate as
V1 = E T diag( p)E, (5.15)

from which we see that if V1 → 0 (V1 ∈ L ∞ ), then E → 0 (E ∈ L ∞ ), implying that

the leaderless consensus (CUUB leaderless consensus) is achieved.
The next section shows how the graph Laplacian matrix works in the Lyapunov
stability analysis of leaderless consensus of nonlinear MAS under the local directed
topology.

5.2 Leaderless Consensus of First-Order Systems

Under Directed Topology

5.2.1 Problem Formulation

The focus of this section is to show how to tackle the directed topology in the
leaderless consensus control problem, so here we consider the following normal-
form continuous-time system with first-order dynamics,
78 5 Cooperative Adaptive Consensus for Multi-agent Systems …

ẋi (t) = gi (z i (t), t)u i (t) + f i (z i (t), t) + f di (z i (t), t) (5.16)

where xi ∈ R and u i ∈ R are the state and control input of the ith (i = 1, 2, . . . , N )
agent, respectively, gi is the control gain, f i denotes the system nonlinearities, and
f di denotes the non-vanishing uncertainties; z i = j∈N i i x j .

Assumption 5.1 The control gain gi is known and nonzero, f i (z i , t) and f di (z i , t)

are known, either continuously differentiable or Lipschitz.

5.2.2 Controller Design and Stability Analysis

for First-Order Systems

The control objective of this section is to design a control strategy under the directed
topology condition such that the leaderless consensus errors uniformly converge to
zero asymptotically, and meanwhile, to give the strict stability analysis.
To this end, the control input for each ith (i = 1, . . . , N ) agent is designed as

1
ui = − [kei + f i + f di ] , (5.17)
gi

where k > 0 is design parameter chosen by the designer.

Theorem 5.1 Distributed Leaderless Consensus Control under Directed Topology

Consider the nonlinear MAS as described by (5.16) under Assumption 5.1. Suppose
that the communication topology G is directed and strongly connected. If the dis-
tributed control law (5.17) is applied, then asymptotically leaderless consensus is
achieved in that the leaderless consensus errors ei (i = 1, . . . , N ) converge to zero
as t → ∞.

Proof By substituting (5.17) into (5.16), one gets

ẋi = −kei . (5.18)

We then choose the Lyapunov function candidate as

1 T
V = E diag( p)E, (5.19)
2
whose derivative along (5.18) is

V̇ (t) = E T diag( p) Ė = E T (diag( p)L ) Ẋ = −k E T (diag( p)L )E. (5.20)

5.2 Leaderless Consensus of First-Order Systems Under Directed Topology 79

Upon using Lemma 5.3, we have

N
kkm T kkm
V̇ (t) = − k E (diag( p)L ) E ≤ −
T
E E =− ei2 . (5.21)
2 2 i=1

Note that
N
1 p̄
V (t) = E T diag( p)E ≤ ei2 , (5.22)
2 2 i=1

where p̄ = max{ p1 , . . . , p N }, and p = [ p1 , . . . , p N ]T is the left eigenvector of L

associated with its zero eigenvalue. It thus follows from (5.21) and (5.22) that

kkm
V̇ (t) ≤ − V (t). (5.23)
p̄

By solving the differential inequality (5.23), we get

V (t) ≤ exp− (t−t0 )

kkm
p̄ V (t0 ). (5.24)

From (5.24) we conclude that, for any given initial state X (t0 ) ∈ R N ,

V (t) → 0 as t → +∞, (5.25)

which further implies

E → 0 as t → +∞, (5.26)

meaning that the leaderless consensus is achieved asymptotically under the proposed
control (5.17). 

5.3 Leaderless Consensus of Second-Order Uncertain

Systems Under Directed Topology

The focus of this section is to address the leaderless consensus problem of second-
order dynamic multi-agent systems with unknown time-varying gain and non-
parametric uncertainties under directed topology.
80 5 Cooperative Adaptive Consensus for Multi-agent Systems …

5.3.1 Problem Formulation—Second-Order Uncertain

Systems Under Directed Topology

Consider a group of subsystems with second-order uncertain dynamics described by

ẋi (t) =vi (t),

v̇i (t) =gi (z i , t)u i (t) + f i (z i , t) + f di (z i , t), (5.27)

where xi ∈ R, vi ∈ R, and u i ∈ R are the position state, velocity state, and control
input of the ith (i = 1, 2, . . . , N ) agent, respectively, gi denotes the control gain,
possibly time-varying and unavailable for controller design, f i denotes the system
nonlinearities, and f di denotes the non-vanishing uncertainties ( f i and f di are either
continuously differentiable or Lipschitz). In addition, z i = j∈N i i {x j , v j }.
The following standard assumptions are in order.

Assumption 5.2
a. The unknown and time-varying gain gi is sign-definite (w. l. o. g. here sgn(gi ) =
+1) and 0 < g i ≤ |gi (·)| ≤ ḡi < ∞ with g i , ḡi being unknown finite constants,
which ensures the global controllability of the system.
b. It is assumed that f i (z i , t) is unknown and non-parameterized. Thus, they are not
available in the finite-time control design in this section. Certain crude structural
information is available to allow an unknown finite positive constant c f i and a
known scalar function ϕi (z i ) such that | f i (z i , t)| ≤ c f i ϕi (z i ) for all t ≥ t0 , where
ϕi (z i ) is bounded for all z i (i = 1, . . . , N ).
c. The unknown and non-vanishing f di (z i , t) is bounded by some finite positive
constant θi , that is, | f di (z i , t)| ≤ θi < ∞.

In addition, it is assumed for ease of notation that the agents’ states are scalars,
i.e., xi ∈ R, vi ∈ R. If the states are vectors, i.e., xi ∈ Rm and vi ∈ Rm , the results of
this chapter can be extended by using the standard method involving the Kronecker
products.
The local neighborhood state error for node i is

ei = ai j (xi − x j ) (5.28)
j∈N i

Let E = [e1 , . . . , e N ]T , X = [x1 , . . . , x N ]T such that E = L X .

The distributed leaderless consensus control design problem confronted herein
is as follows: (1) Design control scheme for all agents i in G under the directed
communication topology condition to make an agreement. The control scheme must
be distributed in the sense that they can only depend on local information about
the agent itself and its neighbors in graph. (2) Overcome the technical difficulty in
5.3 Leaderless Consensus of Second-Order Uncertain Systems … 81

constructing the Lyapunov function arising from asymmetric property of the original
Laplacian matrix under the directed communication topology condition. (3) Tackle
unknown and time-varying control gains and accommodate unknown non-parametric
uncertainties and unknown non-vanishing nonlinearities.

5.3.2 Leaderless Consensus Control Design and Stability

Analysis

The control objective of this section is to design a distributed adaptive controller under
the directed topology condition such that the impacts arising from the asymmetric
property of the original Laplacian matrix under directed topology can be removed,
and meanwhile the unknown time-varying control gain, unmodeling nonlinearities,
and non-vanishing uncertainties can be compensated, and finally, the leaderless con-
sensus of the networked multi-agent systems with second-order uncertain dynamics
is ensured.
The control input for each ith (i = 1, . . . , N ) agent is designed to consist of two
parts: (1) the negative feedback control term u 0i and (2) the compensation control
term u ci , which is of the following form,

u i = u 0i + u ci , (5.29)

in which the negative feedback control term u 0i is generated by

u 0i = −c1 δi , (5.30)

and (2) the compensation control term u ci is generated by

u ci = −ĉ f i ϕi tanh(δi ϕi /τi ) − θ̂i tanh(δi /τi ) (5.31)

with the updated laws

ĉ˙ f i = −γ1i σ1i ĉ f i + γ1i ϕi tanh(δi ϕi /τi )δi ,

θ̂˙ = −γ σ θ̂ + γ tanh(δ /τ )δ ,
i 2i 2i i 2i i i i (5.32)

where δi = vi − vi∗ with vi∗ = −c2 ei , c1 , c2 > 0 are design parameters, ĉ f i and θ̂i
are, respectively, the estimations of c f i and θi (the parameters to be estimated), ϕi (·)
is the scalar and readily computable function as given in Assumption 5.2, and τi , γ1i ,
γ2i , σ1i , and σ2i are positive design parameters chosen by the designer.

Theorem 5.2 Distributed Adaptive Control for Leaderless Consensus under

Directed Topology Consider the second-order nonlinear MAS as described by (5.27)
under Assumption 5.2. Let the communication topology be directed and strongly
82 5 Cooperative Adaptive Consensus for Multi-agent Systems …

connected. If the distributed control laws (5.29)–(5.32) are applied, then leaderless
consensus is achieved in that
(1) the leaderless consensus positive errors ei (i = 1, . . . , N ) and velocity errors
converge to a small residual set Ω1 defined by
⎛ ⎞
2dkv 2c22 dkv 2dkv ⎠
Ω1 = |ei | ≤ , |vi − v j | ≤ 2 ⎝ + c2 ,
pkd kd pkd

∀i, j ∈ {1, . . . , N } , (5.33)

where kd and kv are given, respectively, in (5.58) and (5.61), which are explicitly
computable;
(2) the generalized parameter estimate errors c̃ f i and θ̃i converge to a small set Ω1 p
defined by
 
2 2 2 2
Ω1 p = c g γ̄1 dkv , |θ̃i | ≤ c g γ̄2 dkv , ∀i ∈ {1, . . . , N } ; (5.34)
kd 2 kd 2

(3) all signals in the closed-loop system remain uniformly bounded.

Proof The proof of the result can be done by the following six steps.
Step 1. Construct the distributed part of the Lyapunov function candidate as

1 T
V1 (t) = E diag( p)E. (5.35)
2
Taking the time derivative of V1 (t) along (5.27) yields that

V̇1 (t) = E T diag( p) Ė = E T (diag( p)L )v (5.36)

where v = Ẋ . Let v∗ = −c2 E be the virtual control of v, with c2 being a design

constant. Upon using Lemma 5.3, we have

V̇1 (t) = − c2 E T (diag( p)L )E + E T (diag( p)L )(v − v∗ )

c2 km T
≤− E E + E T (diag( p)L )(v − v∗ )
2
c2 km
N N
  N
=− ei2 + vi − vi∗ ji e j , (5.37)
2 i=1 i=1 j=1

where ji is the ( j, i)th element of diag( p)L. By recalling that δi = vi − vi∗ , we get
that
5.3 Leaderless Consensus of Second-Order Uncertain Systems … 83

N
  N N N
vi − vi∗ ji e j ≤ |δi | | ji ||e j |
i=1 j=1 i=1 j=1
⎡ 2 ⎛ ⎞2 ⎤
N N N N
1⎣
≤ max |δi | |ei | ≤ max |δi | +⎝ |e j |⎠ ⎦
i=1 j=1
2 i=1 j=1
 
1
N N
N max
N
 2 
≤ max N |δi | + N
2
|ei | 2
= δi + ei2 , (5.38)
2 i=1 i=1
2 i=1

 2

N 
N
where max = max j,i∈{1,...,N } | ji |, and the fact that x ≤N x 2 has been
i=1 i=1
used. By substituting (5.38) into (5.37), it thus follows that

c2 km
N
N max
N
 
V̇1 (t) ≤ − ei2 + δi2 + ei2 . (5.39)
2 i=1
2 i=1

Step 2. Define the second part of the Lyapunov function candidate as

N
1
V2 (t) = (vi − vi∗ )2 . (5.40)
2c22 i=1

Taking the derivative of V2 (t) yields

N
1
V̇2 (t) = (vi − vi∗ )(v̇i − v̇i∗ )
c22 i=1
N N
1 1
= 2 δi (v̇i − v̇i∗ ) = 2 δi (v̇i − c2 ėi )
c2 i=1
c2 i=1
⎡ ⎤
N
1 ⎣δi v̇i − c2 δi
= 2 ai j (vi − v j )⎦ . (5.41)
c2 i=1 j∈N i


Let ā = max∀i∈{1,...,N } { ai j } and b̄ = max∀i, j{1,...,N } {ai j }, we then have from the
j∈N i
second term of the right side of (5.41) that
84 5 Cooperative Adaptive Consensus for Multi-agent Systems …

N
1
− c2 δi ai j (vi − v j )
c22 i=1 j∈N i
 
 
1
N
 
≤ 
|δi |  ai j (vi − v j )
c2 i=1  j∈N i 
 
 
1
N
q
 
= |δi |  ai j (vi − v j )
c2 i=1  j∈N i 
⎛ ⎞
N
1
≤ |δi | ⎝ā|vi | + b̄ |v j |⎠ . (5.42)
c2 i=1
j∈N i

Note that
   
|δi ||v j | ≤|δi | v j − v∗j  + |δi | v∗j  ≤ |δi ||δ j | + c2 |δi ||e j |
1  c2  2 
≤ |δi |2 + |δ j |2 + |δi | + |e j |2 , (5.43)
2 2
we then have from (5.42) that
⎛ ⎞
N
1
|δi | ⎝ā|vi | + b̄ |v j |⎠
c2 i=1 j∈N i

āc2  2 
N
1
≤ ā|δi |2 + |δi | + |ei |2
c2 i=1 2
1  2  b̄c2  2 
+ b̄ |δi | + |δ j |2 + |δi | + |e j |2
2 2
j∈N i j∈N i

āc2  2  b̄c2 N̄  2 
N
1
≤ ā|δi |2 + |δi | + |ei |2 + b̄ N̄ |δi |2 + |δi | + |ei |2
c2 i=1 2 2

ā + b̄ N̄
N !" c2 # 2 c2 2 $
= 1+ |δi | + |ei | , (5.44)
c2 i=1
2 2

where N̄ denoting the maximum number of the in-degree and out-degree of each ith
agent for all i ∈ {1, . . . , N }. By combining (5.42) and (5.44), the second term of the
right side of (5.41) can be further written as
5.3 Leaderless Consensus of Second-Order Uncertain Systems … 85

N
1
− c2 δi ai j (vi − v j )
c22 i=1 j∈N i

ā + b̄ N̄
N !" c2 # 2 c2 2 $
≤ 1+ |δi | + |ei | . (5.45)
c2 i=1
2 2

By applying the control laws given in (5.29)–(5.31) to the first term of the right
hand of (5.41), we have

N N
δi v̇i = δi (gi u i + f i + f di )
i=1 i=1
N 
≤ − c1 gi δi2 − δi gi ĉ f i ϕi tanh(δi ϕi /τi ) − δi gi θ̂i tanh(δi /τi )
i=1

+ |δi |c f i ϕi + |δi |θi

N 
≤ − c1 gδi2 − g ĉ f i δi ϕi tanh(δi ϕi /τi ) − g θ̂i δi tanh(δi /τi )
i=1

+ c f i δi ϕi tanh(δi ϕi /τi ) + 0.2785τi c f i + θi δi tanh(δi /τi ) + 0.2785τi θi

N 
= − c1 gδi2 + 0.2785τi (c f i + θi ) + (c f i − g ĉ f i )δi ϕi tanh(δi ϕi /τi )
i=1

+ (θi − g θ̂i )δi tanh(δi /τi ) (5.46)

where we have used the fact that 0 ≤ |s| − s · tanh(s/k) ≤ 0.2785k [5].
By substituting (5.45) and (5.46) into (5.41), we arrive at

N N N
1  
V̇2 (t) ≤k1 δi2 + k2 ei2 + 2
0.2785τi c f i + θi
i=1 i=1 i=1
c2

+ (c f i − g ĉ f i )δi ϕi tanh(δi ϕi /τi ) + (θi − g θ̂i )δi tanh(δi /τi ) (5.47)

where

c1 g ā + b̄ N̄ " c2 # ā + b̄ N̄
k1 = − + 1+ , k2 = . (5.48)
c22 c2 2 2

Step 3. Note that in (5.47) the parameter estimation error of the form •˜ = • −
g •ˆ is involved, which motivates us to introduce the generalized weight parameter
86 5 Cooperative Adaptive Consensus for Multi-agent Systems …

estimation errors c̃ f i (i ∈ {1, . . . , N }) and θ̃i as follows:

c̃ f i = c f i − g ĉ f i , θ̃i = θi − g θ̂i , (5.49)

with which we introduce the third part of the Lyapunov function candidate as

N N
c̃2f i θ̃i2
V3 (t) = + . (5.50)
i=1
2c22 gγ1i i=1
2c22 gγ2i

By applying the adaptive laws for ĉ f i and θ̂i given in (5.32), we get the derivative
of V3 as

ĉ˙ f i θ̂˙i
N N
c̃ f i θ̃i
V̇3 (t) = (− ) + (− )
i=1
c22 γ1i i=1
c22 γ2i

c̃ f i  
N
= 2
σ1i ĉ f i − δi ϕi tanh(δi ϕi /τi )
i=1
c2

θ̃i ! $
N
+ σ2i θ̂i − δi tanh(δi /τi )
i=1
c22
N N
σ1i σ2i
= c̃ f i ĉ f i + θ̃i θ̂i
i=1
c22 i=1
c22
N
1
− (c f i − g ĉ f i )δi ϕi tanh(δi ϕi /τi )
i=1
c22
N
1
− (θi − g θ̂i )δi tanh(δi /τi ). (5.51)
i=1
c22

Step 4. Define the Lyapunov function candidate as

V (t) = V1 (t) + V2 (t) + V3 (t) (5.52)

where V1 (t), V2 (t), and V3 (t) are given in (5.35), (5.40), and (5.50), respectively.
By combining (5.39), (5.47), and (5.51), we then arrive at

N N N
σ1i
V̇ (t) ≤ − k3 δi2 − k4 ei2 + c̃ f i ĉ f i
i=1 i=1 i=1
c22
N N
σ2i 0.2785τi (c f i + θi )
+ θ̃i θ̂i + (5.53)
i=1
c22 i=1
c22
5.3 Leaderless Consensus of Second-Order Uncertain Systems … 87

where

N max c1 g ā + b̄ N̄ " c2 #
k3 = − + − 1+ ,
2 c22 c2 2
c2 km N max ā + b̄ N̄
k4 =− − . (5.54)
2 2 2
!  $
Thus, c1 and c2 can be chosen as c1 > c22 g −1 N 2max + ā+cb̄2 N̄ 1 + c2
2
and c2 >
 
1
km
N max + ā + b̄ N̄ such that k3 and k4 > 0.
Upon using the inequality, ab = 21 [(a + b)2 − a 2 − b2 ], we have

1  2  1 ! 2 $
c̃ f i ĉ f i = c f i − c̃2f i − (c f i − c̃ f i )2 , θ̃i θ̂i = θi − θ̃i2 − (θi − θ̃i )2 .
2g 2g
(5.55)

By substituting (5.55) into (5.53), one gets

N N N N
σ1 σ2
V̇ (t) ≤ −k3 δi2 − k4 ei2 − c̃2f i − θ̃i2 + d (5.56)
i=1 i=1
2gc22 i=1
2gc22 i=1

where σ 1 = min{σ11 , . . . , σ1N }, σ 2 = min{σ21 , . . . , σ2N }, and

 
1
N
σ1i c2f i + σ2i θi2  
d= 2 + 0.2785τi c f i + θi < ∞. (5.57)
c2 i=1
2g

By introducing kd as
 
σ σ
kd = min k3 , k4 , 1 2 , 2 2 , (5.58)
2gc2 2gc2

we further represent (5.56) as

N " #
V̇ (t) ≤ −kd δi2 + ei2 + c̃2f i + θ̃i2 + d. (5.59)
i=1

Step 5. We prove that there exists some finite time t ∗ > 0 such that the leaderless
error and the parameter estimation error will enter into a compact after t ∗ .
88 5 Cooperative Adaptive Consensus for Multi-agent Systems …

Note that
N
1 T p̄
V1 (t) = E diag( p)E ≤ ei2 ,
2 2 i=1
N N
1 1
V2 (t) = (vi − vi∗ )2 = δi2 ,
2c22 i=1
2c22 i=1
N N
1 1
V3 (t) ≤ 2
c̃2f i + 2
θ̃i2 (5.60)
2c2 gγ 1 i=1
2c2 gγ 2 i=1

where p̄ = max{ p1 , . . . , p N } ( p = [ p1 , . . . , p N ]T is the left eigenvector of L

associated with its zero eigenvalue), γ 1 = min{γ11 , . . . , γ1N } and γ 2 = min{γ21 ,
. . . , γ2N }. Let
 
p̄ 1 1 1
kv = max , , , , (5.61)
2 2c22 2c22 gγ 2c22 gγ
1 2

we then have from (5.60) that

N " #
V (t) ≤ kv δi2 + ei2 + c̃2f i + θ̃i2 . (5.62)
i=1

By combining (5.59) and (5.62), we then arrive at

kd
V̇ (t) ≤ − V (t) + d, (5.63)
kv

from which we can conclude that the set Θ = {(xi , vi ) : V (t) < dk kd
v
} is globally
/ Θ, then V̇ < 0. Therefore, there exists a finite time t ∗
attractive. Once (xi , vi ) ∈
such that (xi , vi ) ∈ Θ for ∀t > t ∗ . This further implies that

dkv
V (t) < (5.64)
kd

when t > t ∗ .
Step 6. Derive the estimation for steady-state errors of all agents.
Note that for all i = 1, . . . , N , we have
&
% '
' N
2 2 2dkv
|ei | = ei ≤ (
2
ei2 ≤ V1 (t) ≤ V (t) ≤ , (5.65)
i=1
p p pkd
5.3 Leaderless Consensus of Second-Order Uncertain Systems … 89

with p = min{ p1 , . . . , p N } ( p = [ p1 , . . . , p N ]T is the left eigenvector of L associ-

ated with its zero eigenvalue). Note that
&
' % %
' N
2c22 dkv
|vi − vi∗ | ≤( δi2 ≤ 2c22 V2 (t) ≤ 2c22 V (t) ≤ . (5.66)
i=1
kd

On the other hand,

2dkv
|vi∗ | = | − c2 ei | ≤ c2 . (5.67)
pkd

It thus follows from (5.66) and (5.67) that

2c22 dkv 2dkv

|vi | ≤ |vi − vi∗ | + |vi∗ | ≤ + c2 , (5.68)
kd pkd

which then implies, for ∀i, j ∈ {1, . . . , N }, that

⎛ ⎞
2c22 dkv 2dkv ⎠
|vi − v j | ≤ |vi | + |v j | ≤ 2 ⎝ + c2 . (5.69)
kd pkd

In addition, for all ∀i ∈ {1, . . . , N },

&
' %
' N
2 2
|c̃ f i | ≤( c̃2f i ≤ 2c22 g γ̄1 V3 (t) ≤ c g γ̄1 dkv , (5.70)
i=1
kd 2

and

2 2
|θ̃i | ≤ c g γ̄2 dkv , (5.71)
kd 2

with γ̄1 = max{γ11 , . . . , γ1N }, and γ̄2 = max{γ21 , . . . , γ2N }.

From the above analysis, we conclude that under the proposed control scheme
(5.29)–(5.32), the position and velocity errors between neighbor agents will converge
to a small region Ω1 , defined by
⎧ ⎛ ⎞ ⎫
⎨ 2dkv 2c22 dkv 2dkv ⎠ ⎬
Ω1 = |ei | ≤ , |vi − v j | ≤ 2 ⎝ + c2 , ∀i, j ∈ {1, . . . , N } ,
⎩ pkd kd pkd ⎭
(5.72)
90 5 Cooperative Adaptive Consensus for Multi-agent Systems …

and the generalized parameter estimation converges to the region Ω1 p given as

 
2 2 2 2
Ω1 p = |c̃ f i | ≤ c g γ̄1 dkv , |θ̃i | ≤ c g γ̄2 dkv , ∀i ∈ {1, . . . , N } (5.73)
kd 2 kd 2

when t > t ∗ . 

5.4 Simulation Examples

This section gives a simulation example on a group of networked autonomous surface

vessels (ASVs) [6] with second-order uncertain dynamics to illustrate the effective-
ness of the distributed adaptive control algorithm proposed in (5.29)–(5.32) under
the local and directed communication topology condition. It is shown that under
the proposed control scheme, the group of agents subject to unknown time-varying
control gain, unknown non-parametric nonlinearities, and unknown bounded non-
vanishing uncertainties achieves consensus successfully under the directed topology
condition.
We use four ASVs to demonstrate the effectiveness of our proposed control
scheme. The communication topology among the four ASVs is presented in Fig. 5.1,
which satisfies the assumption that the graph topology is directed and strongly
connected. Each edge weight is taken as 0.1. The left eigenvector of L associated
with eigenvalue 0 is [2, 1, 1, 1]T .
The dynamics of the kth (k = 1, 2, 3, 4) ASV is modeled by
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
m x,k 0 0 v̇x,k f x,k u x,k u x,k
⎣ 0 m y,k 0 ⎦ · ⎣v̇ y,k ⎦ = ⎣ f y,k ⎦ + ⎣u y,k ⎦ + ⎣d y,k ⎦ (5.74)
0 0 m z,k v̇z,k f z,k dx,k dz,k

in which Mk = diag{m x,k , m y,k , m z,k } denotes the mass matrix; rk = [xk , yk , z k ]T ,
vk = ṙk = [vx,k , v y,k , vz,k ]T , and u k = [u x,k , u y,k , u z,k ]T denote the position, velocity,

Fig. 5.1 Directed

communication topology
among the four ASVs
5.4 Simulation Examples 91

and control input vector, respectively; Dk = [dx,k , d y,k , dz,k ]T is environment distur-
bance, and f k = [ f x,k , f y,k , f z,k ]T represents coriolis, centripetal, and hydrodynamic
damping forces and torques acting on the body, where
⎡ ⎤ ⎡ ⎤
A x,k + A|x,k| |vx,k | −m y,k vz,k 0 vx,k
fk = ⎣ m x,k vz,k B y,k + B|y,k| |v y,k | 0 ⎦ · ⎣v y,k ⎦ (5.75)
0 0 C z,k + C|z,k| |vz,k | vz,k

Fig. 5.2 Position error

convergence of the four
ASVs under the proposed
finite-time control scheme
92 5 Cooperative Adaptive Consensus for Multi-agent Systems …

In the simulation, the physical parameters are taken as: Mk = diag{600 + 6(−1)k
+ 6Δm(t), 1000 + 10(−1)k + 10Δm(t), 800 + 8(−1)k + 8Δm(t)} with Δm(t) =
sin(π t/50 − π ), A x,k = −1 + 0.1(−1)k , A|x,k| = −25 + 2.5(−1)k , B y,k = −10 +
(−1)k , B|y,k| = −200 + 20(−1)k , C z,k = −0.5 + 0.05(−1)k , and C|z,k| = −1500 +
150(−1)k for k = 1, 2, 3, 4. The external disturbance is taken as Dk = [3 + 3(−1)k
sin(t/50) + 2 sin(t/10), −1 + 3(−1)k sin(t/20 − π/6) + 2 sin(t), −5(−1)k
sin(0.1t) − sin(t + π/3)]T .

Fig. 5.3 Velocity error

convergence of the four
ASVs under the proposed
finite-time control scheme
5.4 Simulation Examples 93

The simulation objective is that the four ASVs are required to achieve consen-
sus by using the proposed control law given in (5.29)–(5.32) under directed com-
munication topology condition. The initial conditions of the vessels are r1 (0) =
(−1.5m, 0m, π/3rad), r2 (0) = (−1m, 2m, −π/4rad), r3 (0) = (1.5m, 2m, −π/
9rad), r4 (0) = (2m, 1m, π/4rad), vk (0) = (0, 0, 0) (k = 1, 2, 3, 4), respectively. The
control parameters are taken as: s = 2, c1 = 2000, and c2 = 3. In addition, the ini-
tial values of the estimates are chosen as ĉ f x,k = ĉ f y,k = ĉ f z,k = 0 and θ̂x,k = θ̂ y,k =

Fig. 5.4 Control inputs of

the four ASVs under the
proposed finite-time control
scheme
94 5 Cooperative Adaptive Consensus for Multi-agent Systems …

θ̂z,k = 0 for k = 1, 2, 3, 4. The simulation runs for 15 s. The position error conver-
gence results in x-direction, y-direction, and z-direction under the proposed dis-
tributed adaptive control scheme are represented in Fig. 5.2. In addition, the velocity
error convergence results in x-direction, y-direction, and z-direction under the pro-
posed finite-time control scheme are given in Fig. 5.3. From both Figs. 5.2 and 5.3,
we see that the consensus is achieved under the proposed control scheme. Figure 5.4
shows the control input signals of the four ASVs under the proposed finite-time
control scheme, from which we see the control input signal is smooth and bounded
under the local and directed communication condition.

References

1. Olfati-Saber, R., Murray, R.: Consensus protocols for networks of dynamic agents. In: Proceed-
ings of America Control Conference, Denver, CO, pp. 951–956 (2003)
2. Olfati-Saber, R., Murray, R.: Consensus problems in networks of agents with switching topology
and time-delays. IEEE Trans. Autom. Control 49(9), 1520–1533 (2004)
3. Zhang, H., Lewis, F.L., Qu, Z.: Lyapunov, adaptive, and optimal design techniques for coop-
erative systems on directed communication graphs. IEEE Trans. Ind. Electron. 59, 3026–3041
(2012)
4. Qu, Z.: Cooperative Control of Dynamical Systems: Applications to Autonomous Vehicles.
Springer, London (2009)
5. Polycarpout, M.M., Ioannout, P.A.: A robust adaptive nonlinear control design. In: American
Control Conference, pp. 1365–1369 (1993)
6. Almeida, J., Silvestre, C., Pascoal, A.M.: Cooperative control of multiple surface vessels with
discrete-time periodic communications. Int. J. Robust Nonlinear Control 22(4), 398–419 (2012)
 Part III
Finite-Time Cooperative Adaptive
Control for Multiple Networked
Nonlinear Systems
Chapter 6
Finite-Time Leaderless Consensus
Control for Systems with First-Order
Uncertain Dynamics

This chapter investigates the problem of finite-time leaderless consensus of networked

multi-agent systems with first-order uncertain dynamics under local communication
topology condition. Finite-time convergence behavior is of special importance in
cooperative control of MAS, but the vast majority research on finite-time control of
MAS has been focused on linear systems or nonlinear systems with nonlinearities that
can be linearly parameterized, that is, the nonlinearities in that systems are assumed
to exhibit the linear parametric property. The control results on finite-time distributed
control of nonlinear MAS with unknown non-parametric and non-vanishing uncer-
tainties are scarce. Extending the existing finite-time control methods for linear sys-
tems or nonlinear systems with linearly parameterized nonlinearities to MAS subject
to non-parametric and non-vanishing uncertainties encounters significant technical
challenge. The main hindrance stems from the fact that, in the presence of the non-
parametric uncertainties, the commonly used adaptive control law cannot be used to
derive the finite-time convergence because it can not ensure an important relation
that is crucial to derive the finite-time convergence result.
This chapter presents a solution to this problem by focusing on achieving leader-
less consensus of MAS with first-order non-parametric uncertain dynamics in finite-
time with sufficient accuracy. To circumvent the aforementioned technical obstacles,
we introduce the locally defined neighborhood error and fraction power-based adap-
tive updated law for the parameter estimation to embed into the control scheme,
which renders the crucial relation V̇ ≤ cV α + d (0 < α < 1) needed for the finite-
time convergence to be satisfied.

Part of Sect. 6.3 has been reproduced from Wang, Y. J., Song, Y. D., Krstic, M., and Wen, C.
Y.: Fault-tolerant finite-time consensus for multiple uncertain nonlinear mechanical systems under
single-way directed communication interactions and actuation failures. Automatica, vol. 63, pp.
374–383, 2016 © 2016 Elsevier Ltd., reprinted with permission.

© Springer Nature Switzerland AG 2019 97

Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_6
98 6 Finite-Time Leaderless Consensus Control for Systems …

Section 6.1 represents some useful lemmas that is related to the finite-time stability
analysis for the multi-agent systems. Section 6.2 addresses the leaderless consensus
control problem for the first-order multi-agent systems with unknown time-varying
gain and non-parametric uncertainties under undirected topology. A distributed finite-
time leaderless consensus control scheme is derived such that the consensus error
uniformly converges to a small residual set in finite-time, and all the internal sig-
nals are ensured to be uniformly bounded. In addition, the finite convergence time for
each agent to reach the required consensus configuration is explicitly established and
recipes for control parameter selection to make the residual errors as small as desired
are provided. Section 6.3 addresses the leaderless consensus control problem for the
first-order multi-agent systems with known control gain and bounded non-parametric
uncertainties under the directed topology condition, in which the technical difficulty
arising from the non-symmetric property of the Laplacian matrix is circumvented in
the finite-time stability analysis. Section 6.4 gives a simulation example to confirm
the effectiveness and benefits of the proposed finite-time leaderless consensus con-
trol scheme for the multi-agent systems with first-order non-parametric uncertain
dynamics.

6.1 Preliminaries for the Finite-Time Control

In this section, we represent some useful lemmas for the finite-time stability analysis.
We first introduce the definition related to finite-time stability.

Lemma 6.1 ([1]) Consider the system

ẋ = f (x, t), f (0, t) = 0, x ∈ U ⊂ R N (6.1)

where f : U × R+ → R N is continuous on an open neighborhood U of the ori-

gin x = 0. Suppose there exists a continuously differentiable function V (x, t) :
U × R+ → R, a real number c > 0 and 0 < α < 1, and a neighborhood U0 ⊂ U
of the origin such that V (·) is positive definite and V̇ (x, t) + cV α (x, t) ≤ 0 on U0 ,
where V̇ (x) = ∂∂Vx (x) f (x). Then the origin is a finite-time-stable equilibrium of (6.1).
Moreover, there exists a finite settling time T ∗ satisfying

V (x(0))1−α
T∗ ≤ (6.2)
c(1 − α)

such that for any given initial state x(t0 ) ∈ U0 /{0}, limt→T ∗ V (x, t) = 0 and
V (x, t) = 0 for t ≥ T ∗ .

Lemma 6.2 ([2]) If h = h 2 / h 1 ≥ 1, where h 1 , h 2 > 0 are odd integers, then |x −

y|h ≤ 2h−1 |x h − y h |.
6.1 Preliminaries for the Finite-Time Control 99

Lemma 6.3 ([2]) If 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are odd integers, then
|x h − y h | ≤ 21−h |x − y|h .

Lemma 6.4 ([2]) For x, y ∈ R, if c, d > 0, then |x|c |y|d ≤ c/(c + d)|x|c+d
+ d/(c + d)|y|c+d .
N
Lemma 6.5 ([3]) For xi ∈ R, i = 1, 2, . . . , N , 0 < h ≤ 1, then ( i=1 |xi |)h ≤
N  N
i=1 |x i | ≤ N ( i=1 |xi |) .
h 1−h h

Lemma 6.6 ([4]) For x ∈ R, y ∈ R, 0 < h = h 1 / h 2 ≤1, where h 1 , h 2 > 0 are pos-
itive odd integers, then x h (y − x) ≤ 1+h
1
y 1+h − x 1+h .
 1+h 
Proof Let g(x) = x h (c − x) − 1+h 1
c − x 1+h . Taking the derivation of g(x)
with respect to x yields

d
g(x) = hx h−1 (c − x) − x h + x h = hx h−1 (c − x) (6.3)
dx

It is straightforward to derive that g(x) arrives its maximum value at x = c, which

implies

g(x) ≤ g(c) = 0, ∀x ∈ R. (6.4)

Thus, x h (y − x) ≤ 1
1+h
(y 1+h − x 1+h ) holds. 

6.2 Finite-Time Leaderless Consensus Control Under

Undirected Topology

In this section, we address the leaderless consensus control problem for networked
multi-agent systems with first-order uncertain dynamics. The agent dynamics are
subject to unknown time-varying gain and unknown non-parametric uncertainties,
meaning that the parameters are unknown and time-varying and the dynamics are not
allowed to appear in the control protocols. This section formulates the basic structure
of the distributed finite-time control protocols to confront these problems, allowing
the leaderless consensus to be achieved in finite-time with sufficient accuracy in
spite of the unknown time-varying gain and unknown non-parametric uncertainties.
We introduce the locally defined neighborhood error of fractional-order and fraction
power-based adaptive updated law for the parameter estimation to embed into the
control scheme and show how to achieve the finite-time convergence by the negative
feedback of the fractional-order local error and how to compensate the unknown non-
parametric uncertainties by using the adaptive control technique based on fraction
power-based adaptive updated law.
100 6 Finite-Time Leaderless Consensus Control for Systems …

6.2.1 Problem Formulation of Finite-Time Leaderless

Consensus for First-Order Uncertain Systems

Consider a group of N agents with scalar first-order dynamics

ẋi (t) = gi (xi , t)u i (t) + f i (z i , t) + f di (z i , t) (6.5)

where xi ∈ R and u i ∈ R are the state and control input of the ith (i = 1, 2, . . . , N )
agent, respectively, gi denotes the control gain, f i denotes the system nonlinearities
and f di denotes the non-vanishinguncertainties ( f i and f di are either continuously
differentiable or Lipschitz); z i = j∈N i  i x j .
Standard assumptions for the existence of unique solutions and global controlla-
bility of system are made.

Assumption 6.1

a. The control effectiveness gain gi (unknown and time-varying) is sign-definite

(w. l. o. g. here sgn(gi ) = +1) and 0 < g i ≤ |gi (·)| ≤ ḡi < ∞ with g i , ḡi being
unknown finite constants, which ensures the global controllability of the system.
b. It is assumed that f i (z i , t) is unknown and non-parameterized. Thus, they are
not available in the finite-time control design. Moreover, certain crude structural
information is available to allow that | f i (z i , t)| ≤ c f i ϕi (z i ) for all t ≥ t0 , where
c f i ≥ 0 is an unknown finite constant and ϕi (z i ) is a known scalar function, and
ϕi (z i ) is bounded for any z i (xi ) (i = 1, 2, . . . , N ).
c. It is assumed that f di (z i , t) is unknown and non-vanishing, and f di (z i , t) is
bounded by some unknown finite constant θi , i.e., | f di (z i , t)| ≤ θi < ∞.

For ease of notation, the agents’ states are scalars, i.e., xi ∈ R. If the states are
vectors, i.e., xi ∈ Rn , the results of this chapter can be extended by using the standard
method involving the Kronecker products. In addition, it should be stressed that
although the bounds mentioned in Assumption 6.1 are assumed to exist, they are not
used in the controller design and do not have to be known. They appear in the error
bounds in the control results derived in Theorem 6.1 as can be seen later.
The local neighborhood state error for node i is denote by

ei = ai j (xi − x j ) (6.6)
j∈N i

Let E = [e1 , . . . , e N ]T , X = [x1 , . . . , x N ]T such that E = L X .

The distributed finite-time leaderless consensus control design problem con-
fronted herein is as follows: (1) Design control schemes for all agents i in G to
make an agreement in finite time. The control scheme must be distributed in the
sense that they can only depend on local information about the agent itself and its
neighbors in graph, and meanwhile, the control scheme should be endowed with the
finite-time quality; (2) Tackle the unknown and time-varying control effectiveness
6.2 Finite-Time Leaderless Consensus Control Under Undirected Topology 101

gain explicitly; (3) Accommodate the unknown non-parametric nonlinearities and

unknown non-vanishing uncertainties upon using the fraction power-based adaptive
updated law.

6.2.2 Finite-Time Leaderless Consensus Control Design

The control objective of this section is to design a control strategy such that the
leaderless consensus error uniformly converges to a small residual set in finite time,
in spite of the unknown time-varying gain and unknown non-parametric uncertainties.
In addition, the finite convergence time for each agent to reach the required consensus
configuration is explicitly established and recipes for control parameter selection to
make the residual errors as small as desired are provided.
To this end, the control input for each ith (i = 1, . . . , N ) agent is designed to con-
sist of two parts: (1) the negative feedback control term u 0i , and (2) the compensation
control term u ci , which is of the following form,

u i = u 0i + u ci (6.7)

in which the negative feedback control term based on the fractional-order local error
is generated by
q
u 0i = −ki ei (6.8)

and (2) the compensation control term is generated by

u ci = −ĉ f i ϕi tanh(ei ϕi /τi ) − θ̂i tanh(ei /τi ) (6.9)

with the fraction power-based updated law

ĉ˙ f i = −γ1i σ1i ĉ f i + γ1i ϕi tanh(ei ϕi /τi )ei ,

q

θ̂˙i = −γ2i σ2i θ̂i + γ2i tanh(ei /τi )ei ,

q
(6.10)

where 0 < q = q1 /q2 < 1 (q1 , q2 are positive odd integers), ĉ f i and θ̂i are the esti-
mations of c f i and θi (the parameters to be estimated), respectively, ϕi (·) is the scalar
and readily computable function as given in Assumption 6.1, ki , τi , γ1i , γ2i , σ1i , and
σ2i are positive finite design parameters chosen arbitrarily by the designer.

Remark 6.1 It is seen from (6.8) to (6.10) that the proposed control structure is of
“sublinear” form in that the control scheme is built upon fraction state feedback
control and the adaptation rule (i.e., the leakage term) is also of fraction form, rather
than just simply linear state feedback. It is such sublinear control with fraction leakage
adaptation that ensures the multi-agent systems to have the attractive performance
as stated in the next theorem.
102 6 Finite-Time Leaderless Consensus Control for Systems …

6.2.3 Finite-Time Leaderless Consensus Stability Analysis

We need to prove a relation firstly, which is crucial to derive the main results.

Lemma 6.7 For the generalized weight parameter estimation error •˜ = • − g •ˆ and
the constant 0 < q = q1 /q2 < 1 (q1 and q2 are positive odd integrators), it holds
that

1  (q−1)(1−q) 
•˜ •ˆ q ≤ g −q 2 − 2q−1 •˜ 1+q
1+q

q 1
2−(q−1) (1+q) •1+q .
2
+ 1 − 2q−1 + + (6.11)
1+q 1+q

Proof Firstly, by noting that •˜ = • − g •ˆ , we get

q
•˜ •ˆ q = •˜ g −1 (• − •˜ ) = g −q •˜ (• − •˜ )q . (6.12)

According to Lemma 6.6, we let x = • − •˜ , y = •, and then y − x = •˜ , thus we

have
1
•˜ •ˆ q ≤ g −q •1+q − (• − •˜ )1+q . (6.13)
1+q
q1 +q2
By noting that 1 + q = 1 + q1
q2
= q2
, and q1 + q2 is an even number, we then get
from (6.13) that

1 
•˜ •ˆ q = g −q •1+q − |• − •˜ | |• − •˜ |q . (6.14)
1+q

Upon using Lemma 6.3, we arrive at

1  
•˜ •ˆ q ≤ g −q •1+q − 2q−1 |• − •˜ | •q − •˜ q  . (6.15)
1+q

Note that both the function g(s) = s (s ∈ R)  q f (s)q = s (s ∈ R) are non-

 and function
q

|• |  q
decreasing, we then have − •˜ • − •˜ = (• − •˜ ) • − •˜ , and (6.15) becomes
q

1
•˜ •ˆ q ≤ g −q •1+q − 2q−1 •1+q +2q−1 • •˜ q + 2q−1 •˜ •q −2q−1 •˜ 1+q . (6.16)
1+q

Upon using Lemma 6.4, we get that

  1  q−1 1+q q
2q−1 •˜ •q = 2q−1 •˜ •q ≤ 2 •˜ + •1+q , (6.17)
1+q 1+q
6.2 Finite-Time Leaderless Consensus Control Under Undirected Topology 103

and
 q  q
2q−1 • •˜ q = 2q−1 • 2q−1 •˜ 2−(q−1)q = 2(q−1)(1−q) • 2q−1 •˜
q  q−1 1+q 1  (q−1)(1−q) 1+q
≤ 2 •˜ + 2 • . (6.18)
1+q 1+q

By inserting (6.17) and (6.18) into (6.16), we then arrive at


1 q  q−1 1+q 1  (q−1)(1−q) 1+q
•˜ •ˆ q ≤ g −q •1+q −2q−1 •1+q 2 •˜ + 2 •
1+q 1+q 1+q
1  q−1 1+q q
+ 2 •˜ + •1+q −2q−1 •˜ 1+q
1+q 1+q

−q 1  (q−1)(1−q) 
=g 2 − 2q−1 •˜ 1+q
1+q

q 1
2−(q−1) (1+q) •1+q ,
2
+ 1 − 2q−1 + + (6.19)
1+q 1+q

which then implies that (6.11) holds. 

Theorem 6.1 (Distributed Adaptive Finite-Time Control for Leaderless Consensus)

Suppose that the communication topology is undirected and connected, and Assump-
tion 6.1 is satisfied. The nonlinear MAS subject to unknown time-varying gain and
non-parametric/non-vanishing uncertainties as described by (6.5) with the proposed
control scheme (6.7)–(6.10) achieves the finite-time leaderless consensus in that

(1) the leaderless consensus errors ei (i = 1, . . . , N ) converge to a small residual

set Ω1 defined by
⎧ ⎛ 1+q ⎞ 1+q ⎫
⎪  1
⎪
⎨ 2 2
kv d ⎠ ⎬
Ω1 = |ei | ≤ ⎝ , ∀i = 1, . . . , N (6.20)
⎪
⎩ λmin (Λ) η2 η1 kd ⎪
⎭

in a finite time T ∗ satisfying

1+q
1−q
∗ V (t0 ) 2 kv 2
T ≤ (6.21)
(1 − η2 )η1 kd c̃( 1−q
2
)

where 0 < η1 ≤ 1, 0 < η2 < 1, V (t0 ) is known, kd and kv are given in (6.37)
and (6.41), respectively, which are explicitly computable;
104 6 Finite-Time Leaderless Consensus Control for Systems …

(2) the generalized parameter estimate errors c̃ f i and θ̃i converge to a small set Ω1 p
defined by
⎛ 1+q ⎞ 1+q 1
  2
k v d
Ω1 p = |c̃ f i | ≤ 2g γ̄1 ⎝ ⎠ ,
η2 η1 k d
⎛ ⎞ 1+q
1

1+q
2

⎝ k v d
⎠
|θ̃i | ≤ 2g γ̄2 , ∀i = 1, . . . , N (6.22)
η2 η1 k d

in the finite time T ∗ ;

(3) all signals in the closed-loop system remain uniformly bounded.

Proof The proof of the result can be done by the following five steps.
Step 1. Construct the distributed part of the Lyapunov function candidate as

1 T 1
V1 (t) = E ΛE = X T L X (6.23)
2 2

where Λ = U NT Λ̃−1 U N , Λ̃ = diag{a, λ2 , . . . , λ N } (a is an arbitrarily finite positive

constant), U N is an orthogonal matrix such that L = U NT diag{0, λ2 , . . . , λ N }U N =
U NT Λ0 U N , with Λ0 = diag{0, λ2 , . . . , λ N } provided by Lemma 3.1. Such matrix Λ
is symmetric and positive definite, implying that the distributed part of Lyapunov
function V1 (t) is well defined. Taking the time derivative of V1 (t) along (6.5) yields
that


N 
N
V̇1 (t) = E T Ẋ = ei ẋi = ei (gi u i + f i + f di ). (6.24)
i=1 i=1

Inserting the control input u i given in (6.7) and u 0i given in (6.8) into (6.24) yields

N 
 1+q
V̇1 (t) = −ki gi ei + gi ei u ci + ei f i + ei f di
i=1
N 
 1+q
≤ −ki gi ei + gi ei u ci + |ei |c f i ϕi + |ei |θi (6.25)
i=1

By using the fact that 0 ≤ |s| − s · tanh(s/k) ≤ 0.2785k [5], and by inserting the
compensation control term u ci given in (6.9) into (6.25), we arrive at
6.2 Finite-Time Leaderless Consensus Control Under Undirected Topology 105


N
1+q
V̇1 (t) ≤ − ki gi ei − gi ei ϕi ĉ f i tanh(ei ϕi /τi ) − gi ei θ̂i tanh(ei /τi )
i=1

+ c f i ei ϕi tanh(ei ϕi /τi ) + θi ei tanh(ei /τi ) + 0.2785τi (c f i + θi )

N
1+q
≤ − ki gi ei + (c f i − g ĉ f i )ei ϕi tanh(ei ϕi /τi )
i=1

+ (θi − g θ̂i )ei tanh(ei /τi ) + 0.2785τi (c f i + θi ) . (6.26)

Step 2. Note that in (6.26) the parameter estimation error of the form •˜ = • − g •ˆ is
involved, which motivates us to introduce the generalized weight parameter estima-
tion errors c̃ f and θ̃ as follows,

c̃ f = c f − g ĉ f , θ̃ = θ − g θ̂ (6.27)

where c̃ f = [c̃ f 1 , . . . , c̃ f N ]T ∈ R N , and θ̃ = [θ̃1 , . . . , θ̃ N ]T ∈ R N .

Upon using the generalized weight parameter estimation error, we define the
distributed part of the Lyapunov function candidate as

1 T −1 1 T −1
V2 (t) = c̃ f Γ1 c̃ f + θ̃ Γ2 θ̃ (6.28)
2g 2g

where Γ1 = diag{γ11 , . . . , γ1N } ∈ R N ×N and Γ2 = diag{γ21 , . . . , γ2N } ∈ R N ×N .

Taking the derivative of V2 yields
! " ⎛ ⎞ ! " ! "
˙
 g ĉ˙ f i  1 ⎝ g θ̂i ⎠  ĉ˙ f i  θ̂˙i
N N N N
1
V̇2 (t) = c̃ f i − + θ̃i − = c̃ f i − + θ̃i − .
g γ1i g γ2i γ1i γ2i
i=1 i=1 i=1 i=1
(6.29)

Now applying the adaptive laws for ĉ f i and θ̂i given in (6.10)–(6.29), we get


N
q q 
V̇2 (t) = σ1i c̃ f i ĉ f i − c̃ f i ϕi ei tanh(ei ϕi /τi ) + σ2i θ̃i θ̂i − θ̃i ei tanh(ei /τi ) .
i=1
(6.30)

Step 3. Define the Lyapunov function candidate as

V (t) = V1 (t) + V2 (t) (6.31)

where V1 (t) and V2 (t) are given in (6.23) and (6.28), respectively.
By combining the derivative of V1 (t) given in (6.26) and the derivative of V2 (t)
given in (6.30), and recalling that c̃ f = c f − g ĉ f and θ̃ = θ − g θ̂ , we then arrive at
106 6 Finite-Time Leaderless Consensus Control for Systems …

V̇ (t) = V̇1 (t) + V̇2 (t)


N
1+q q q
≤ − ki gi ei + 0.2785τi (c f i + θi ) + σ1i c̃ f i ĉ f i + σ2i θ̃i θ̂i . (6.32)
i=1

Upon using Lemma 6.7, it is straightforward that


q 1  (q−1)(1+q)  1+q
c̃ f i ĉ f i ≤ 2 − 2q−1 c̃ f i
g (1 + q)
q

q 1
2−(q−1) (1+q) c f i ,
2 1+q
+ 1 − 2q−1 + + (6.33)
1+q 1+q

q 1  (q−1)(1+q)  1+q
θ̃i θ̂i ≤ q 2 − 2q−1 θ̃i
g (1 + q)

q 1
2−(q−1) (1+q) θi
2 1+q
+ 1 − 2q−1 + + . (6.34)
1+q 1+q

By substituting (6.33) and (6.34) into (6.32), one gets

N 
 1+q 1+q 1+q
V̇ (t) ≤ − kd1 ei + kd2 c̃ f i + kd3 θ̃i +d (6.35)
i=1

where

kd1 = kg
σ 
kd2 = q 1 2q−1 − 2(q−1)(1+q)
g (1 + q)
σ 
kd3 = q 2 2q−1 − 2(q−1)(1+q)
g (1 + q)
N  
1 q 1
2−(q−1) (1+q)
2
d= 0.2785τi (c f i + θi ) + q 1 − 2q−1 + +
g (1 + q) 1+q 1+q
i=1

1+q 1+q
× σ1i c f i + σ2i θi < ∞, (6.36)

with k = min{k1 , . . . , k N }, σ 1 = min{σ11 , . . . , σ1N }, and σ 2 = min{σ21 , . . . , σ2N }.

Note that 2q−1 − 2(q−1)(1+q) > 0, which ensures that kd2 > 0 and kd3 > 0. By intro-
ducing kd as

kd = min{kd1 , kd2 , kd3 }, (6.37)

6.2 Finite-Time Leaderless Consensus Control Under Undirected Topology 107

we can further represent (6.35) as

N 
 1+q 1+q 1+q
V̇ (t) ≤ −kd ei + c̃ f i + θ̃i + d. (6.38)
i=1

Step 4. We prove that there exists a finite time T ∗ > 0 that can be explicitly expressed
and a bounded constant 0 < ζ < ∞ such that V (t) < ζ when t ≥ T ∗ in the sequel.
Note that

λmax (Λ)  2
N
1
V1 (t) = E T ΛE ≤ ei , (6.39)
2 2 i=1

1  2 1  2
N N
V2 (t) ≤ c̃ f i + θ̃ , (6.40)
2γ 1 g i=1 2γ 2 g i=1 i

where γ 1 = min{γ11 , . . . , γ1N } and γ 2 = min{γ21 , . . . , γ2N }. By introducing kv as

λmax (Λ) 1 1
kv = max{ , , }, (6.41)
2 2γ 1 g 2γ 2 g

we then get from (6.39) and (6.40) that

N 

V (t) ≤ kv ei2 + c̃2f i + θ̃i2 . (6.42)
i=1

According to Lemma 6.5, we get from (6.42) that

# N $ 1+q
1+q 1+q  2
1+q  
N
1+q 1+q 1+q
V (t) 2 ≤ kv 2
ei + c̃ f i + θ̃i
2 2 2
≤ kv 2 ei + c̃ f i + θ̃i .
i=1 i=1
(6.43)

η1 k d
Let c̃ = 1+q (kd and kv are given in (6.37) and (6.41), respectively, and 0 < η1 ≤ 1).
2
kv
By combining (6.38) and (6.43), it thus follows that
1+q
V̇ (t) ≤ −c̃V (t) 2 + d. (6.44)
2
Let Θ = {X : V (t) < ( ηd2 c̃ ) 1+q , 0 < η2 < 1}. According to Theorem 5.2 in [6], for
2 1+q
any X ∈ / Θ for all t ∈ [0, tx ], it holds that V (t) ≥ (d/η2 c̃) 1+q i.e., d ≤ η2 c̃V (t) 2 ,
for all t ∈ [0, tx ]. This fact, together with (6.44), implies that
1+q 1+q 1+q
V̇ (t) ≤ −c̃V (t) 2 + η2 c̃V (t) 2 = −(1 − η2 )c̃V (t) 2 (6.45)
108 6 Finite-Time Leaderless Consensus Control for Systems …

2
for all t ∈ [0, tx ]. Note that V (t) ≥ (d/η2 c̃) 1+q > 0 for t ∈ [0, tx ], it thus follows
1+q
V (t0 )1− 2
from (6.45) that tx < (1−η2 )c̃(1− 1+q )
according to Lemma 6.1. Therefore for ∀t ≥ T ∗ ,
2
∗
with T satisfying
1+q
V (t0 )1− 2
T∗ ≤ , (6.46)
(1 − η2 )c̃(1 − 1+q
2
)

we have
d 1+q 2
V (t) < ( ) = ζ. (6.47)
η2 c̃

Step 5. Derive the estimation for steady-state errors of all agents.

Note that for all i = 1, 2, . . . , N , we have
% 
 & N
& 2V1 (t)
|ei | = ei ≤2 ' ei ≤
2

i=1
λmin (Λ)
   1
2V (t) 2 d 1+q
≤ ≤ . (6.48)
λmin (Λ) λmin (Λ) η2 c̃

In addition, for all i = 1, 2, . . . , N ,

%
 & N 
&
|c̃ f i | = c̃ f i ≤ '
2
c̃2f i ≤ 2g γ̄1 V2 (t)
i=1
   1
d 1+q
≤ 2g γ̄1 V (t) ≤ 2g γ̄1 (6.49)
η2 c̃

and

  1
d 1+q
|θ̃i | ≤ 2g γ̄2 (6.50)
η2 c̃

with γ̄1 = max{γ11 , . . . , γ1N } and γ̄2 = max{γ21 , . . . , γ2N }.

Thus under the proposed finite-time control scheme (6.7)–(6.10), the leaderless
consensus errors between neighbor agents will converge to a small region Ω1 , defined
by
(   1 )
2 d 1+q
Ω1 = |ei | ≤ , ∀i = 1, . . . , N , (6.51)
λmin (Λ) η2 c̃
6.2 Finite-Time Leaderless Consensus Control Under Undirected Topology 109

and the generalized parameter estimation errors will converge to the region Ω1 p ,
   1
  1 
d 1+q d 1+q
Ω1 p = |c̃ f i | ≤ 2g γ̄1 , |θ̃i | ≤ 2g γ̄2 , ∀i = 1, . . . , N
η2 c̃ η2 c̃
(6.52)

in the finite time T ∗ . 

6.2.4 Comparison with Regular State Feedback-Based

Infinite-Time Method

Note that if we take the value of the fraction power q as 1, the control scheme
(6.7)–(6.10) reduces to the following regular state feedback-based infinite-time
control

u i = −ki ei − ĉ f i ϕi tanh(ei ϕi /τi ) − θ̂i tanh(ei /τi ), (6.53)

with the updated laws

ĉ˙ f i = −γ1i σ1i ĉ f i + γ1i ϕi tanh(ei ϕi /τi )ei ,

θ̂˙i = −γ2i σ2i θ̂i + γ2i tanh(ei /τi )ei . (6.54)

With such regular state feedback-based control, the following result is derived.
Theorem 6.2 Consider system (6.5) under Assumption 6.1 and undirected con-
nected topology condition. If controlled by the adaptive control law given in (6.53)–
(6.54), the cooperative uniformly ultimately bounded (UUB) leaderless consensus is
achieved in that the leaderless consensus errors converge to a set Ω2 defined by
(  )
2kv d
Ω2 = |ei | ≤ , ∀i = 1, . . . , N , (6.55)
λmin (Λ)η2 η1 kd

and the generalized parameter estimate errors c̃ f i and θ̃i converge to a set Ω2 p given
as
⎧   ⎫
⎨  2g γ̄1 kv d   2g γ̄2 kv d ⎬
Ω2 p = c̃ f i  ≤ , θ̃i  ≤ , ∀i = 1, . . . , N , (6.56)
⎩ η2 η1 k d η2 η1 k d ⎭

respectively, where 0 < η1 ≤ 1, 0 < η2 < 1, V (t0 ) is known, kd and kv are given,
respectively, in (6.37) and (6.41) with q = 1, which are explicitly computable.
110 6 Finite-Time Leaderless Consensus Control for Systems …

Proof Following the same line as in the proof of Theorem 6.1 and by setting q = 1,
it can be derived that
kd
V̇ (t) ≤ − V (t) + d (6.57)
kv

where kd , kv , and d are defined the same as in Theorem 6.1 associated with q = 1.
Define
 
kv d
Θ̄ = X : V (t) < , 0 < η2 < 1 (6.58)
η2 k d

η2 kd V (t)
Since d ≤ kv
when X ∈
/ Θ̄, it follows from (6.57) that

kd η2 k d
V̇ (t) ≤ − V (t) + V (t) < 0. (6.59)
kv kv

Hence, V̇ (t) < 0 once X ∈ / Θ̄. Note that Θ̄ is a global attractive region due to the
fact that Θ̄ is a level set of the Lyapunov function, that is, the states would not escape
from Θ̄ after entering it.
In addition, it can be readily derived that under the control laws (6.53)–(6.54)
the leaderless consensus errors converge to the set Ω2 specified by (6.55) and the
generalized estimation errors converge to Ω2 p given in (6.56), respectively. 

6.2.5 Notes

It is worth noting that several parameters in the proposed control scheme

(6.7)–(6.10) can be adjusted, which are roughly identified into three groups: one
is in the definition of kd , another one is in the definition of kv , and the third one is
in the definition of d. The finite-time T ∗ defined in (6.21) can be made smaller by
1+q
letting kv 2 /kd smaller, i.e., making kv smaller and kd larger; the error convergence
1+q
bounds in region Ω1 can be made smaller by letting dkv 2 /kd smaller, i.e., making
kv and d smaller and kd larger. Upon examining the definition of kv , kd , and d, it is
seen that increasing ki lead to an increasing kd ; increasing ki , γ1i and γ2i results in
a decreasing kv ; increasing ki and decreasing σ1i , σ2i and τi lead to a smaller d. In
addition, by increasing the parameters σ1i , σ2i , γ1i , and γ2i , the adaptation rate for ĉ f i
and θ̂i becomes faster. However, certain tradeoff between transient performance and
control precision need to be made in practice. This recipe for parameter selection
makes it clear and straightforward to choose the suitable parameters to affect the
settling time T ∗ as defined in (6.21) and the control precision as reflected in Ω1 and
Ω2 , enabling the proposed control scheme with the abilities to make both the residual
errors and the convergence time as small as desired.
6.2 Finite-Time Leaderless Consensus Control Under Undirected Topology 111

It is seen that by choosing the control gains ki large enough, all convergence
regions, i.e., Ω1 and Ω1 p in the proposed finite-time control scheme, Ω2 and Ω2 p in
the regular state feedback-based control scheme, can be reduced as small as desired.
Nevertheless, due to control saturation constraint, ki are not allowed to be set too
large. While in the proposed method one can choose the fractional power q properly
to enhance the disturbance rejection performance and increase the control precision
without the need for excessively large ki .
The main contributions of this section are summarized as follows. First, the dis-
tributed adaptive finite-time control scheme is established based upon a fractional-
order local state negative feedback and fractional power-based adaptive updated law,
allowing the leaderless consensus to be achieved in finite time with sufficient accu-
racy. Second, by introducing the generalized parameter estimation error, the unknown
and time-varying control gain inherent in the systems is dealt with gracefully. Third,
with the norm-bounding skill the lumped uncertainties of each agent are converted
into a simplified relation that only involves a virtual parameter to be estimated and a
computable scalar function with core information, thus facilitating the online com-
putations and control design, and the unknown and non-parametric uncertainties are
explicitly accommodated.

6.3 Finite-Time Leaderless Consensus Control Under

Directed Topology

In this section, we address the finite-time leaderless consensus problem for networked
multi-agent systems with first-order dynamics under directed graph topology. We
first introduce the Lyapunov function design for the stability analysis under directed
topology, and then we start with controller design and stability analysis to explain
the fundamental idea and the technical development of the proposed method.

6.3.1 Problem Formulation

The focus of this section is to show how to tackle the technical difficulty arising
from the asymmetric Laplacian matrix under the directed communication topology
condition in the finite-time leaderless consensus problem, so here we consider the
first-order normal-form system (6.5) under the directed communication topology
condition in the case that the control gain gi is known and nonzero, and f i (z i , t),
f di (z i , t) are known.
Standard assumptions for the existence of unique solutions and global controlla-
bility of system are made, e.g., f i and f di are either continuously differentiable or
Lipschitz.
112 6 Finite-Time Leaderless Consensus Control for Systems …

The finite-time leaderless consensus control design problem confronted herein is

as follows: (1) Design control schemes for all agents i in G to make an agreement
in finite-time under directed topology. The control schemes must be distributed in
the sense that they can only depend on local information about the agent itself and
its neighbors in graph, and meanwhile, the control schemes should be endowed
with the finite-time quality; (2) Overcome the technical difficulty in constructing
the Lyapunov function arising from asymmetric property of the original Laplacian
matrix under the directed communication topology condition.

6.3.2 Lyapunov Function Design Under Directed Topology

In this subsection, we introduce the Lyapunov function design for the finite-time sta-
bility analysis under the strongly connected directed topology. In the next subsection,
we will show how to apply such Lyapunov function in the Lyapunov analysis of the
finite-time leaderless consensus problem for multi-agent systems under the directed
topology condition.
Before moving on, we first present several useful lemmas. It is important to note
that the Laplacian matrix L is no longer symmetric under the one-way directed
topology, which imposes significant challenge for the finite-time leaderless consen-
sus control design because one cannot use such L to construct Lyapunov function
directly. To circumvent the technical difficulty in constructing Lyapunov function
arising from asymmetric property of the original Laplacian matrix L under directed
graph, we introduce a new matrix Q [7],

Q = (diag( p)L + L T diag( p)), (6.60)

where diag( p) = diag{ p1 , . . . , p N } ∈ R N ×N > 0 with p = [ p1 , . . . , p N ]T being the

left eigenvector of L associated with its zero eigenvalue. Such defined matrix Q
is actually a Laplacian corresponding to a connected undirected graph [4], which
exhibits the property of symmetry and semi-positive definiteness with one simple
zero eigenvalue.
The significance of the feature on Q is that the original directed graph problem
can be treated as an undirected graph problem. Furthermore, such feature allows for
the development of the following two lemmas that are crucial for the construction of
the Lyapunov function design subject to directed graph.
Lemma 6.8 For a strongly connected digraph G with Laplacian L , let Q be defined
as in (6.60), then ∀x = 0 N , x T Qx = 0 if and only if x = c1 N , where c = 0 is a
T
constant. Moreover, min xx TQxx
exists and
x =c1 N

x T Qx n
0 < min T
≤ λi (Q). (6.61)
x =c1n x x i=2
6.3 Finite-Time Leaderless Consensus Control Under Directed Topology 113

Upon using Lemma 6.8, the following result can be established.

q q T
Lemma 6.9 ∀E q = 0 N (E q = e1 , . . . , e N ), there exists a constant km > 0 such
that

(E q )T Q E q
≥ km . (6.62)
(E q )T E q
N
Proof Note that for p T E = L X = 0, i.e., i=1 pi ei = 0, and pi > 0 according
to [8], from which we know that for E = 0 and i = 1, . . . , N , it is impossible that
sgn(ei ) = 1 (or sgn(ei ) = −1). Therefore, E = c1 N with c being a nonzero constant.
q
Note that sgn(ei ) = sgn(ei ) (i = 1, . . . , N ), it is straightforward that E q = c1 N .
(E q )T Q E q
According to Lemma 6.8, there exists a constant km = min (E q )T E q
> 0 such that
q E =0 N
(E q )T Q E q
(E q )T E q
≥ km > 0. 
According to the above analysis, we construct the Lyapunov function candidate as

1  1+q T 1+q
V = E 2 diag( p)E 2 , (6.63)
1+q
 1+q
1+q
1+q T
T
where E 2 = e1 2 , . . . , e N2 . Hereafter, we let E h = e1h , . . . , ehN with
h ∈ R.

6.3.3 Finite-Time Leaderless Consensus Control Design

and Stability Analysis Under Directed Topology

The control objective of this section is to design a finite-time control strategy under
the directed topology condition such that the leaderless consensus errors uniformly
converge to zero in finite-time, and meanwhile, to give the strict stability analysis.
To this end, the control input for each ith (i = 1, . . . , N ) agent is designed as

1 q 
ui = − kei + f i + f di , (6.64)
gi

where 0 < q = q1 /q2 < 1, q1 , q2 are positive odd integers, and k > 0 is a design
parameter chosen by the designer.
Theorem 6.3 (Distributed Finite-Time Control for Leaderless Consensus under
Directed Topology) Consider the nonlinear MAS as described by (6.5) under the
assumptions that gi is known and nonzero and f i (z i , t), f di (z i , t) are known. Let
the communication topology be directed and strongly connected. If the distributed
control law (6.64) is applied, then finite-time leaderless consensus is achieved in that
114 6 Finite-Time Leaderless Consensus Control for Systems …

(1) the leaderless consensus errors ei (i = 1, . . . , N ) converge to zero in a finite

time T ∗ , which satisfies
2q 1−q
2 p̄ 1+q [(1 + q) V (t0 )] 1+q
T∗ ≤ , (6.65)
η1 kkm (1 − q)

where p̄ = max{ p1 , . . . , p N } ( p = [ p1 , . . . , p N ]T is the left eigenvector of L

associated with it zero eigenvalue), 0 < η1 ≤ 1, V (t0 ) is known, which is explic-
itly computable.

Proof By applying the control scheme (6.64) into (6.5), we get

q
ẋi = −kei . (6.66)
1+q 1+q
We then choose the Lyapunov function candidate V = 1
1+q
(E 2 )T diag( p)E 2 ,
whose derivative along (6.66) is
 T  T
V̇ (t) = E q diag( p) Ė = E q (diag( p)L ) Ẋ
 T
= −k E q (diag( p)L )E q , (6.67)

q q T
with E q = e1 , . . . , e N . Upon using Lemma 6.9, we have

V̇ (t) = − k(E q )T (diag( p)L ) E q

kkm  q T q kkm 
N
≤− E E =− (ei )2q . (6.68)
2 2 i=1

Note that

1  1+q p̄  1+q
N
T 1+q
V (t) = E 2 diag( p)E 2 ≤ e . (6.69)
1+q 1 + q i=1 i

According to Lemma 6.5, we then get from (6.69) that,

 ! N " 1+q
2q

 
2q 2q N
2q p̄ 1+q
1+q p̄ 1+q
2q
V (t) 1+q ≤ ei ≤ ei (6.70)
1+q i=1
1+q i=1

2q
Let c̃ = 21 η1 kkm ( 1+q
p̄
) 1+q (0 < η1 ≤ 1). It thus follows from (6.68) and (6.70) that

2q
V̇ (t) ≤ −c̃V (t) 1+q . (6.71)
6.3 Finite-Time Leaderless Consensus Control Under Directed Topology 115

Then according to Lemma 6.1, we get from (6.71) that, there exists a finite-time
constant T ∗ , which satisfies,
2q
V (t0 )1− 1+q
T∗ ≤  , (6.72)
c̃ 1 − 1+q 2q

such that for any given initial state X (t0 ) ∈ R N /{0}, limt→T ∗ V (X, t) = 0 and
V (X, t) = 0 for t ≥ T ∗ , which further implies that

E → 0 as t → T ∗− ,
E = 0 when t ≥ T ∗ . (6.73)

This implies that the leaderless consensus is achieved in the finite time T ∗ under the
proposed control (6.64). In addition, by substituting c̃ into (6.72), we further get that,
2q 1−q

∗ 2 p̄ 1+q [(1 + q) V (t0 )] 1+q

T ≤ , (6.74)
η1 kkm (1 − q)

which then yields (6.65). 

6.4 Simulation Example

This section gives a simulation example on a first-order nonlinear multi-agent system

with unknown time-varying gain and non-parametric uncertainties to illustrate the
feasibility of the distributed adaptive finite-time control algorithm proposed in (6.7)–
(6.10).
We use four agents to demonstrate the effectiveness of our proposed control
scheme. The communication graph among the four agents is presented in Fig. 6.1,
which satisfies the assumption that the graph topology is undirected and connected.
Each edge weight is set to be 1.
The dynamics of the four agents is of the following form,

ẋ1 =(5 + exp−x1 )u 1 + 0.1 exp−x1 −x2 x1 + f d1 ,

2 2 2

ẋ2 =(5 + exp−x2 )u 2 + 0.1 exp−x2 −x3 x22 + f d2 ,

2 2 2

ẋ3 =(5 + exp−x3 )u 3 + 0.1 exp−x2 −x4 x32 + f d3 ,

2 2 2

ẋ4 =(5 + exp−x4 )u 4 + 0.1 exp−x1 −x4 x4 + f d4 ,

2 2 2
(6.75)

in which the non-vanishing uncertainties are of the form: f d1 = 0.02 sin(0.05t),

f d2 = 0.01 cos(0.06t), f d3 = 0.02 cos(0.05t), and f d4 = 0.01 sin(0.06t).
116 6 Finite-Time Leaderless Consensus Control for Systems …

Fig. 6.1 Communication graph topology among the four agents

The earlier theoretical analysis has declared that the nonlinear system (6.75) can
achieve finite-time leaderless consensus under the control scheme proposed in (6.7)–
(6.10). To set up the proposed controller, we only need to select the design parameters
ki , γi , and σi (i = 1, 2, 3, 4) and does not need to take any time-consuming analytical
derivation for choosing design parameters. The simulation runs for 0.6s. The initial
condition are X (0) = [2, 1, −1, −2]T . In addition, the initial values of the param-
eter estimates are chosen as ĉ f i (0) = θ̂i (0) = 0 for i = 1, 2, 3, 4. The simulation
is conducted by applying the control laws given in (6.7)–(6.10), where the design
parameters are taken as: ki = 1 (i = 1, 2, 3, 4), γi = 0.01, and σi = 0.01 (i = 1, 2),
ε = 0.01. In addition, the core functions ϕi (·) (i = 1, 2, 3, 4) are easily taken as
ϕ1 = |x1 |, ϕ2 = x22 , ϕ3 = x32 , and ϕ4 = |x4 | according to the core information of the
system.
Figure 6.2 presents the position trajectories of all the four agents, and Fig. 6.3
shows the neighborhood position synchronization errors of all the four agents, both

2 x1 x2 x3 x4

0
X

−1

−2
0 0.1 0.2 0.3 0.4 0.5
Time (s)

Fig. 6.2 Position trajectories xi (i = 1, 2, 3, 4) of all the four agents

6.4 Simulation Example 117

5 e1 e2 e3 e4

0
E

−5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (s)

Fig. 6.3 Neighborhood errors ei (i = 1, 2, 3, 4) of the four agents under the proposed finite-time
control scheme (6.7)

3 u1 u2 u3 u4

0
U

−1

−2

−3
0 0.1 0.2 0.3 0.4 0.5 0.6
Time (s)

Fig. 6.4 Control inputs u i (i = 1, 2, 3, 4) of the four agents

of which verify that the position synchronization is achieved in finite time. Figure 6.4
shows the control input signals of the four agents u i (i = 1, 2, 3, 4), from which
we observe that the control input signals of all the four agents are continuous and
bounded.
To show that better performance is achieved with the proposed finite-time con-
trol scheme, we also tested the convergence property between the proposed finite-
time control scheme (6.7)–(6.10) and the typical infinite-time-based adaptive scheme
(6.53)–(6.54). Both the two control laws are applied to system (6.75) by using the
same design parameters. It is observed from Figs. 6.3, 6.5, and 6.6 that the conver-
gence rate is faster and the error precision is better with the finite-time controller
compared with the infinite-time controller.
118 6 Finite-Time Leaderless Consensus Control for Systems …

6 e1 e2 e3 e4

0
E

−2

−4

−6
0 0.1 0.2 0.3 0.4 0.5 0.6
Time (s)

Fig. 6.5 Neighborhood errors ei (i = 1, 2, 3, 4) of the four agents under the infinite-time control
scheme (6.53)

5
e1 under finite-time controller
e1 under infinite-time controller
4
e3 under infinite-time controller
e3 under infinite-time controller
3

2
E

−1
0 0.1 0.2 0.3 0.4 0.5 0.6
Time (s)

Fig. 6.6 Neighborhood error convergence comparison of agent 1 and 3 under the two different
control schemes

References

1. Bhat, S.P., Bernstein, D.S.: Continuous finite-time stabilization of the translational and rotational
double integrators. IEEE Trans Autom. Control 43(5), 678–682 (1998)
2. Qian, C., Lin, W.: Non-Lipschitz continuous stabilizer for nonlinear systems with uncontrollable
unstable linearization. Syst. Control Lett. 42(3), 185–200 (2001)
3. Hardy, G., Littlewood, J., Polya, G.: Inequalities. Cambridge University Press, Cambridge (1952)
4. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Fault-tolerant finite time consensus for multiple
uncertain nonlinear mechanical systems under single-way directed communication interactions
and actuation failures. Automatica 63, 374–383 (2016)
5. Polycarpout, M.M., Ioannout, P.A.: A robust adaptive nonlinear control design. In: American
Control Conference, pp. 1365–1369 (1993)
References 119

6. Bhat, S.P., Bernstein, D.S.: Finite-time stability of continuous autonomous systems. SIAM J.
Control Optim. 38(3), 751–766 (2000)
7. Zhang, H., Lewis, F.L., Qu, Z.: Lyapunov, adaptive, and optimal design techniques for coop-
erative systems on directed communication graphs. IEEE Trans. Ind. Electron. 59, 3026–3041
(2012)
8. Qu, Z.: Cooperative Control of Dynamical Systems: Applications to Autonomous Vehicles.
Springer, London (2009)
Chapter 7
Finite-Time Consensus for Systems
with Second-Order Uncertain Dynamics
Under Undirected Topology

In Chap. 6, we have addressed the finite-time leaderless consensus control prob-

lem for networked multi-agent systems with first-order uncertain dynamics. In this
chapter, we investigate the distributed adaptive finite-time consensus control for
cooperative multi-agent systems with second-order dynamics where the unknown
time-varying effectiveness gain and non-parametric uncertainties are involved. By
considering the requirement in most practical applications that both position and
velocity must be controlled such as in formation control and coordination among
multiple unmanned aerial vehicles, and also considering the fact that during system
operation, mass property of the system might change due to payload releasing, reload-
ing, and/or fuel burning, thus leading to unknown and time-varying parameters in the
system, a MAS model with second-order dynamics that explicitly accounts for non-
parametric uncertainties, non-vanishing external disturbances as well as unknown
time-varying effectiveness gain is necessary to be established and studied. In par-
ticular, to study the finite-time control of such networked multi-agent systems with
second-order uncertain dynamics is highly requirable.
However, extending the finite-time control methods for first-order systems to
second-order case encounters significant technical challenge. The main hindrance
stems from the fact that the filtering technique commonly used for system with
second-order dynamics cannot be used here to derive the finite-time result upon
utilizing Babarlat lemma or UUB theory because in the finite-time stability analysis,
the convergence of the filtered error to zero or to a small value cannot ensure the finite
time convergence of the original error. To circumvent this technical difficulty, we
introduce a virtual controller and resort to adding a power integrator technique, which
allows the networked multi-agent systems with second-order uncertain dynamics to
achieve the leaderless consensus in finite time with sufficient accuracy.
Section 7.1 formulates the leaderless consensus control problem for multi-agent
systems with second-order dynamics in the presence of unknown time-varying gain
and non-parametric uncertainties under undirected topology. Section 7.2 represents
some useful lemmas that is related to the finite-time stability analysis for the multi-
agent systems. Section 7.3 addresses the leaderless consensus controller design and

© Springer Nature Switzerland AG 2019 121

Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_7
122 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

stability analysis for the second-order uncertain multi-agent systems. A distributed

finite-time leaderless consensus control scheme is derived such that the consensus
error uniformly converges to a small residual set in finite time and all the internal
signals are ensured to be uniformly bounded. In addition, the finite convergence time
for each agent to reach the required consensus configuration is explicitly established
and recipes for control parameter selection to make the residual errors as small
as desired are provided. Section 7.4 conducts numerical simulation to confirm the
effectiveness of the theoretical results obtained in the previous sections. Section 7.5
gives some notes on the finite-time control of the second-order uncertain multi-agent
systems.

7.1 Formulation of Finite-Time Consensus

for Second-Order Uncertain Systems

In this section, we formulate the leaderless consensus control problem for networked
multi-agent systems with second-order dynamics. The agent dynamics are subject
to unknown time-varying gain and unknown non-parametric uncertainties, meaning
that the parameters are unknown and time-varying and the dynamics are not allowed
to appear in the control protocols.
We consider a group of N agents with scalar second-order dynamics

ẋi (t) = vi (t),

v̇i (t) = gi (z i , t)u i (t) + f i (z i , t) + f di (z i , t), (7.1)

where xi ∈ R, vi ∈ R, and u i ∈ R are the position state, velocity state, and control
input of the ith (i = 1, 2, . . . , N ) agent, respectively, gi denotes the control effective-
ness gain, possibly time-varying and unavailable for controller design, f i denotes the
system nonlinearities, and f di denotes the non-vanishing uncertainties  ( f i and f di are
either continuously differentiable or Lipschitz). In addition, z i = j∈N i  i {x j , v j }.
Standard assumptions for existence of unique solutions and global controllability
of system are made.

Assumption 7.1 a. The control gain gi is unknown and time-varying and is sign-
definite (w. l. o. g. here sgn(gi ) = +1). In addition, 0 < g i ≤ |gi (·)| ≤ ḡi < ∞
with g i , ḡi being unknown finite constants, which ensures the global controlla-
bility of the system.
b. Certain crude structural information is available to allow the unknown and
non-parametric uncertainty f i (z i , t) satisfy | f i (z i , t)| ≤ c f i ϕi (z i ) for all t ≥ t0 ,
where c f i ≥ 0 is an unknown finite constant and ϕi (z i ) is a known scalar func-
tion, and ϕi (z i ) is bounded for all z i (i = 1, . . . , N ).
c. The unknown and non-vanishing f di (z i , t) is bounded by some unknown finite
constant θi , that is, | f di (z i , t)| ≤ θi < ∞.
7.1 Formulation of Finite-Time Consensus for Second-Order Uncertain Systems 123

It is assumed for ease of notation that the agents’ states are scalars, i.e., xi ∈ R,
vi ∈ R. If the states are vectors, i.e., xi ∈ Rn and vi ∈ Rn , the results of this chapter
can be extended by using the standard method involving the Kronecker products.
The local neighborhood state error for node i is

ei = ai j (xi − x j ) (7.2)
j∈N i

Let E = [e1 , . . . , e N ]T , X = [x1 , . . . , x N ]T such that E = L X .

The distributed finite-time leaderless consensus control design problem con-
fronted herein is as follows: (1) Design control schemes for all agents i with second-
order uncertain dynamics in G to make an agreement in finite time. The control
schemes must be distributed in the sense that they can only depend on local infor-
mation about the agent itself and its neighbors in graph, and meanwhile, the control
schemes should be endowed with the finite-time quality; (2) Tackle unknown and
time-varying control gains explicitly; (3) Accommodate the unknown non-parametric
uncertainties and unknown non-vanishing nonlinearities by using the control method
based on fractional power-based adaptive laws.

7.2 Useful Lemmas

In this section, we represent some useful lemmas for the finite-time stability analysis.

Lemma 7.1 ([1]) Suppose there exists a continuously differentiable function

V (x, t) : U × R+ → R (U ⊂ R N is an open neighborhood of the origin), a real
number c > 0 and 0 < α < 1, such that V (x, t) is positive definite and V̇ (x, t) +
cV (x, t)α ≤ 0 on U0 (U0 ⊂ U), then V (x, t) is locally in finite-time convergent with
a finite settling time satisfying T ∗ ≤ V (x(0))
1−α

c(1−α)
, such that for any given initial state
x(t0 ) ∈ U0 /{0}, limt→T ∗ V (x, t) = 0 and V (x, t) = 0 for t ≥ T ∗ .

Lemma 7.2 ([2]) If h = h 2 / h 1 ≥ 1, where h 1 , h 2 > 0 are odd integers, then |x −

y|h ≤ 2h−1 |x h − y h |.

Lemma 7.3 ([2]) If 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are odd integers, then
|x h − y h | ≤ 21−h |x − y|h .

Lemma 7.4 ([2]) For x, y ∈ R, if c, d > 0, then |x|c |y|d ≤ c/(c + d)|x|c+d + d/
(c + d)|y|c+d .
N
Lemma 7.5 ([3]) For xi ∈ R, i = 1, 2, . . . , N , 0 < h ≤ 1, then ( i=1 |xi |)h ≤
N  N
i=1 |x i | ≤ N ( i=1 |xi |)h .
h 1−h

Lemma 7.6 ([4]) For x ∈ R, y ∈ R, 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are pos-

itive odd integers, then x h (y − x) ≤ 1+h
1
(y 1+h − x 1+h ).
124 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

Lemma 7.7 ([5]) For a connected undirected graph G , the Laplacian matrix L has
the following properties: (1) L is semi-definite. (2) 0 is a simple eigenvalue of L
and 1 is the associated eigenvector. (3) Assuming the eigenvalue of L is denoted as
0, λ2 (L), . . . , λ N (L) satisfying 0 ≤ λ2 (L) ≤ · · · ≤ λ N (L), then the second smallest
eigenvalue λ2 (L) > 0. Furthermore, if 1TN X = 0, then X T L X ≥ λ2 (L)X T X .

Lemma 7.8 For a connected undirected graph G , it holds that

E T E ≥ λ2 (L)X T L X (7.3)

where E = [e1 , . . . , e N ]T is the leaderless consensus error vector defined by (7.2)

and X = [x1 , . . . , x N ]T is the position state vector.

Proof Note that E = L X . Thus, E T E = X T L T L X = X T L 2 X . Since L is a diag-

onalizable symmetric semi-positive definite matrix, then it is easy to prove that
L 1/2 is also a symmetric semi-positive definite matrix, and L 2 = L 1/2 L L 1/2 . Let
 T  1/2 
W = L 1/2 1 N , we get W T W = L 1/2 1 N L 1 N = 1TN L1 N = 0. Thus, W = 0 N ,
which yields W X = 0, i.e., 1 N L X = 0. Then according to Lemma 7.7, it holds
T T 1/2

that
 T  
E T E = L 1/2 X L L 1/2 X ≥ λ2 (L)X T L X. (7.4)

Lemma 7.9 For the generalized weight parameter estimation error •˜ = • − g •ˆ and
the constant 0 < q = q1 /q2 < 1 (q1 and q2 are positive odd integrators), it holds
that

q −q 1  (q−1)(1−q) 
•˜ •ˆ ≤ g 2 − 2q−1 •˜ 1+q
1+q

q 1
2−(q−1) (1+q) •1+q .
2
+ 1 − 2q−1 + + (7.5)
1+q 1+q

Proof The proof can be seen in Lemma 6.7, so it is omitted here.

7.3 Distributed Adaptive Finite-Time Control

for Second-Order Uncertain Systems

This section represents the basic structure of the distributed adaptive finite-time
control protocols for the second-order uncertain multi-agent systems, allowing the
leaderless consensus to be achieved in finite time in spite of the unknown time-
varying gain and unknown non-parametric uncertainties. We show how to realize
7.3 Distributed Adaptive Finite-Time Control for Second-Order Uncertain Systems 125

the finite-time convergence by fractional-order state feedback and how to compen-

sate the unknown non-parametric uncertainties by using adaptive control techniques
based on the fractional-order adaptive updated law to guarantee finite-time leaderless
consensus of second-order uncertain multi-agent systems.

7.3.1 Finite-Time Consensus Control Design

The control objective of this section is to design distributed adaptive finite-time con-
troller such that the impacts arisen from the time-varying control gain, unmodeling
nonlinearities, and non-vanishing uncertainties can be compensated and meanwhile
the leaderless consensus of the networked multi-agent system is ensured in finite
time.
Before design the controller, we first introduce the virtual control of vi as

vi∗ = −c2 ei ,
q
(7.6)

where 0 < q = q1 /q2 < 1, q1 , q2 are positive odd integers, and c2 > 0 is an arbitrar-
ily finite constant. We also introduce the virtual error as
1
1
δi = viq − (vi∗ ) q . (7.7)

The control input for each ith (i = 1, . . . , N ) agent is designed to consist of two
parts: (1) the negative feedback control term u 0i and (2) the compensation control
term u ci , which is of the following form

u i = u 0i + u ci , (7.8)

in which the negative feedback control term u 0i is generated by

2q−1
u 0i = −c1 δi , (7.9)

and (2) the compensation control term u ci is generated by

2−q 2−q
u ci = −ĉ f i ϕi tanh(δi ϕi /τi ) − θ̂i tanh(δi /τi ) (7.10)

with the updated laws

ĉ˙ f i = −γ1i σ1i ĉ f i + γ1i ϕi tanh(δi ϕi /τi )δi ,

q 2−q 2−q

θ̂˙i = −γ2i σ2i θ̂i + γ2i tanh(δi /τi )δi ,

q 2−q 2−q
(7.11)
126 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

where δi is defined by (7.7), 0 < q = q1 /q2 < 1, q1 , q2 are positive odd integers,
c1 , c2 > 0 are design parameters, ĉ f i and θ̂i are, respectively, the estimations of c f i
and θi (the parameters to be estimated), ϕi (·) is the scalar and readily computable
function as given in Assumption 7.1, and ki , τi , γ1i , γ2i , σ1i , and σ2i are positive finite
design parameters chosen by the designer.

7.3.2 Finite-Time Consensus Stability Analysis

Theorem 7.1 (Distributed Adaptive Finite-Time Consensus of Second-Order

Uncertain Systems) Suppose that Assumption 7.1 holds and the communication topol-
ogy is undirected and connected. The uncertain MAS (7.1) with the distributed adap-
tive control law (7.8)–(7.11) achieves finite-time leaderless consensus in that
(1) the leaderless consensus positive errors ei (i = 1, . . . , N ) and velocity errors
converge to a small residual set Ω1 defined by

 1  12 (q+1)
2kv d 1+q c
Ω1 = |ei | ≤ , |vi − v j | ≤ 2 2q q −1
λmin (Λ) η2 η1 k d q 2 22
⎛ ⎞ q
 q 1+q
2
1+q 
2 2
⎝ dkv ⎠ , ∀i, j ∈ {1, . . . , N } ,
+ c2 (7.12)
λmin (Λ) η2 η1 k d

in a finite time T ∗ satisfying

1+q
1−q
∗ 2V (t0 ) 2 kv 2
T = , (7.13)
(1 − η2 )(1 − q)η1 kd

where 0 < η1 ≤ 1, 0 < η2 < 1, V (t0 ) is known, kd and kv are given respectively
in (7.42) and (7.46), which are explicitly computable;
(2) the generalized parameter estimate errors c̃ f i and θ̃i converge to a small set Ω1 p
defined by

  1
d 1+q
Ω1 p = |c̃ f i | ≤ 2kγ g γ̄1 ,
η2 c̃
  1 
d 1+q
|θ̃i | ≤ 2kγ g γ̄2 , ∀i ∈ {1, . . . , N } (7.14)
η2 c̃

in the finite time T ∗ ;

(3) all signals in the closed-loop system remain uniformly bounded.
7.3 Distributed Adaptive Finite-Time Control for Second-Order Uncertain Systems 127

Proof The proof of the result can be done by the following six steps.
Step 1. Construct the distributed part of the Lyapunov function candidate as

1 T 1
V1 (t) = E ΛE = X T L X (7.15)
2 2

where Λ = U NT Λ̃−1 U N , Λ̃ = diag{a, λ2 , . . . , λ N } (a is an arbitrarily positive con-

stant), U N is an orthogonal matrix such that L = U NT diag{0, λ2 , . . . , λ N }U N =
U NT Λ0 U N , with Λ0 = diag{0, λ2 , . . . , λ N } provided by Lemma 3.1. Such matrix
Λ is symmetric and positive definite, implying that the distributed part of Lyapunov
function V1 (t) is well defined. Taking the time derivative of V1 (t) along (7.1) yields
that


N
V̇1 (t) = E T Ẋ = ei vi . (7.16)
i=1

Let vi∗ = −c2 ei be the virtual control of vi , then we have

q


N 
N 
N 
N
ei vi∗ vi∗ ) ei (vi − vi∗ ).
1+q
V̇1 (t) = + ei (vi − = −c2 ei + (7.17)
i=1 i=1 i=1 i=1

1
1
By recalling that δi = viq − (vi∗ ) q , we get from Lemma 7.3 that


N 
N
ei (vi − vi∗ ) ≤ |ei ||vi − vi∗ |
i=1 i=1
 1  q 

N
 q 
N
= |ei |  viq ∗ q1
− (vi )  ≤ 21−q
 |ei ||δi |q . (7.18)
i=1 i=1

Upon using Lemma 7.4, it follows from (7.18) that

 21−q   1+q  21−q   1+q 

N N N
ei (vi − vi∗ ) ≤
1+q
|ei | + q|δi |1+q = ei + qδi .
i=1
1 + q i=1 1 + q i=1
(7.19)

By substituting (7.19) into (7.17), we arrive at

 21−q   1+q 
N N
1+q 1+q
V̇1 (t) ≤ − c2 ei + ei + qδi . (7.20)
i=1
1 + q i=1
128 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

Step 2. Define the second part of the Lyapunov function candidate by adding a
power integrator [2]

N 
 vi  1 
1 1 2−q
V2 (t) = 1+1/q
ς q − (vi∗ ) q dς (7.21)
21−q c2 i=1 vi∗

which is positive semi-definite and C 1 [2]. Taking the derivative of V2 (t) yields

N  1
 2−q
1 1
V̇2 (t) = 1+1/q
viq − (vi∗ ) q v̇i
21−q c2 i=1
 1  1
vi 1 1−q d(−(vi∗ ) q )
+ (2 − q) ς q − (vi∗ ) q dς ·
vi∗ dt
  
 2−q
N vi  1 
1 1 1−q
1

= 1+1/q
δi v̇i + (2 − q) ς q − (vi∗ ) q dς · c2 e˙i
q

21−q c2 i=1 vi
∗

 N   vi  
1 1 1 1−q
ς q − (vi∗ ) q
2−q
= 1+1/q
δi v̇i + (2 − q) dς
21−q c2 i=1 vi∗
1 

× c2q ai j (vi − v j ) . (7.22)

j∈N i

Since
   
 vi  1    1 1−q     
 1 1−q   
 vi − vi∗  = |δi |1−q vi − vi∗ 
1
 ς q − (vi∗ ) q dς  ≤  viq − (vi∗ ) q
 vi∗   
≤ |δi |1−q 21−q |δi |q = 21−q |δi | (7.23)

then the second term of the right side of (7.22) becomes

N 
  1  1 
2−q vi 1 1−q

1+1/q
ς q − (vi∗ ) q dς · c2q ai j (vi − v j )
21−q c2 i=1 vi∗ j∈N i
 
 
2−q   
N
≤ 
|δi |  ai j (vi − v j ) . (7.24)
c2 i=1  j∈N i 


Let ā = max∀i∈{1,...,N } { ai j }, b̄ = max∀i, j{1,...,N } {ai j }, and N̄ denoting the
j∈N i
maximum number of the out-degree of all the ith (i ∈ {1, . . . , N }) agents in the
graph G . We then see that
7.3 Distributed Adaptive Finite-Time Control for Second-Order Uncertain Systems 129
  ⎛ ⎞
 
2−q    2−q  
N N
|δi |  ai j (vi − v j ) ≤ |δi | ⎝ā|vi | + b̄ |v j |⎠
c2 i=1  j∈N i  c2 i=1
j∈N i

2−q   
N
≤ |δi | ā|vi | + b̄ N̄ |vi | . (7.25)
c2 i=1

Upon using Lemmas 7.3 and 7.4, we have

   
|δi ||vi | ≤ |δi | vi − vi∗  + |δi | vi∗  ≤ 21−q |δi ||δi |q + c2 |δi ||ei |q
c2  1+q 
≤ 21−q |δi |1+q + |δi | + q|ei |1+q . (7.26)
1+q

Upon using (7.25) and (7.26), (7.24) can be further written as

N 
  1  1 
2−q vi 1 1−q

1+1/q
ς q − (vi∗ ) q dς · c2q ai j (vi − v j )
21−q c2 i=1 vi∗ j∈N i
N 

2−q c2 1+q c2 1+q
≤ (ā + b̄ N̄ ) 21−q + δi + e . (7.27)
c2 i=1
1+q 1+q i

By applying the control laws given in (7.8)–(7.10) to the first term of the right
hand of (7.22), we have


N
2−q 
N
2−q
δi v̇i = δi (gi u i + f i + f di )
i=1 i=1
N 
    
1+q 2−q 2−q 2−q 2−q
≤ − c1 gi δi − δi gi ĉ f i ϕi tanh δi ϕi /τi − δi gi θ̂i tanh δi /τi
i=1
2−q 2−q
+ |δi |c f i ϕi + |δi |θi

N 
    
1+q 2−q 2−q 2−q 2−q
≤ − c1 gδi − g ĉ f i δi ϕi tanh δi ϕi /τi − g θ̂i δi tanh δi /τi
i=1
   
2−q 2−q 2−q 2−q
+ c f i δi ϕi tanh δi ϕi /τi + 0.2785τi c f i + θi δi tanh δi /τi + 0.2785τi θi

N 
    
1+q 2−q 2−q
= − c1 gδi + 0.2785τi (c f i + θi ) + c f i − g ĉ f i δi ϕi tanh δi ϕi /τi
i=1
   
2−q 2−q
+ θi − g θ̂i δi tanh δi /τi (7.28)

where we have used the fact that 0 ≤ |s| − s · tanh(s/k) ≤ 0.2785k [6].
130 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

By substituting (7.27)–(7.28) into (7.22), we arrive at

   
N
1+q
N
1+q
N
1  
V̇2 (t) ≤ k1 δi + k2 ei + 1+1/q
0.2785τi c f i + θi
i=1 i=1 21−q c2 i=1
       
2−q 2−q 2−q 2−q
+ cfi − g ĉ f i δi ϕi tanh δi ϕi /τi + θi − g θ̂i δi tanh δi /τi
(7.29)

where

c1 g 2−q c2
k1 = − + (ā + b̄ N̄ ) 21−q + ,
21−q c2
1+1/q c2 1+q
2−q
k2 = (ā + b̄ N̄ ). (7.30)
1+q

Step 3. Note that in (7.29) the parameter estimation error of the form •˜ = • −
g •ˆ is involved, which motivates us to introduce the generalized weight parameter
estimation errors c̃ f i (i ∈ {1, . . . , N }) and θ̃i as follows:

c̃ f i = c f i − g ĉ f i , θ̃i = θi − g θ̂i , (7.31)

with which we introduce the third part of the Lyapunov function candidate as


N
c̃2f i 
N
θ̃i2
V3 (t) = + . (7.32)
i=1
2kγ gγ1i i=1
2kγ gγ2i

1+1/q
where kγ = 21−q c2 .
By applying the adaptive laws for ĉ f i and θ̂i given in (7.11), we get the derivative
of V3 as
   
 ĉ˙ f i  θ̂˙i
N N
c̃ f i θ̃i
V̇3 (t) = − + −
i=1
kγ γ1i k
i=1 γ
γ2i
 c̃ f i  
N
q 2−q 2−q
= σ1i ĉ f i − δi ϕi tanh(δi ϕi /τi )
i=1
kγ
 θ̃i  
N
q 2−q 2−q
+ σ2i θ̂i − δi tanh(δi /τi )
k
i=1 γ
7.3 Distributed Adaptive Finite-Time Control for Second-Order Uncertain Systems 131


N
σ1i q

N
σ2i q
= c̃ f i ĉ f i + θ̃i θ̂i
i=1
kγ i=1
kγ
 1    
N
2−q 2−q
− c f i − g ĉ f i δi ϕi tanh δi ϕi /τi
i=1
kγ
 1    
N
2−q 2−q
− θi − g θ̂i δi tanh δi /τi . (7.33)
k
i=1 γ

Step 4. Define the Lyapunov function candidate as

V (t) = V1 (t) + V2 (t) + V3 (t) (7.34)

where V1 (t), V2 (t), and V3 (t) are given in (7.15), (7.21), and (7.32), respectively.
By combining (7.20), (7.29), and (7.33), we then arrive at


N
1+q

N
1+q

N
σ1i q
V̇ (t) ≤ − k3 δi − k4 ei + c̃ f i ĉ f i
i=1 i=1 i=1
kγ

N
σ2i q

N
0.2785τi (c f i + θi )
+ θ̃i θ̂i + (7.35)
i=1
kγ i=1
kγ

where

21−q q c1 g 2−q c2
k3 = − + − (ā + b̄ N̄ ) 21−q + ,
1+q 1−q
2 c2
1+1/q c2 1 + q
21−q 2−q
k4 = c2 − − (ā + b̄ N̄ ). (7.36)
1+q 1+q

1+1/q −1 21−q q
Thus, c1 and c2 can be chosen as c1 > 21−q c2 g + 2−q (ā + b̄ N̄ )
   1−q 
1+q c2

2 1−q
+ 1+q and c2 > 1+q 2
c2 1
+ (2 − q)(ā + b̄ N̄ ) such that k3 and k4 > 0.
Upon using Lemma 7.9, it is straightforward that

q 1  (q−1)(1−q)  1+q
c̃ f i ĉ f i ≤ 2 − 2q−1 c̃ f i
g (1 + q)
q

q 1
2−(q−1) (1+q) c f i ,
2 1+q
+ 1 − 2q−1 + + (7.37)
1+q 1+q

q 1  (q−1)(1−q)  1+q
θ̃i θ̂i ≤ q 2 − 2q−1 θ̃i
g (1 + q)

q 1
2−(q−1) (1+q) θi
2 1+q
+ 1 − 2q−1 + + . (7.38)
1+q 1+q
132 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

By substituting (7.37) and (7.38) into (7.35), one gets


N
1+q

N
1+q

N
1+q

N
1+q
V̇ (t) ≤ −k3 δi − k4 ei − k5 c̃ f i − k6 θ̃i +d (7.39)
i=1 i=1 i=1 i=1

where
 
σ 1 2(q−1)(1−q) − 2q−1
k5 = 1+1/q
,
21−q c2 g q (1 + q)
 (q−1)(1−q) 
σ2 2 − 2q−1
k6 = 1+1/q q
,
21−q c2 g (1 + q)
N  1+q 1+q
q 1 (σ1i c f i + σ2i θi )
2−(q−1) (1+q) ·
2
d= 1 − 2q−1 + +
i=1
1+q 1+q 21−q c2
1+1/q q
g (1 + q)
 
0.2785τi c f i + θi
+ 1+1/q
< ∞, (7.40)
21−q c2

(q−1)(1−q)
with σ 1 = min{σ
 1−q11 , . . ., σ1N } and σ 2 = min{σ21 , . . . , σ2N }. Note that 2 −
2 q−1
=2 q−1
2 − 1 > 0, and thus, k5 > 0 and k6 > 0. We further represent
(7.39) as

N 
 
1+q 1+q 1+q 1+q
V̇ (t) ≤ −kd δi + ei + c̃ f i + θ̃i +d (7.41)
i=1

where

kd = min{k3 , k4 , k5 , k6 }. (7.42)

Step 5. We prove that there exists a finite time T ∗ > 0 and a bounded constant
0 < ζ < ∞ such that V (t) < ζ when t ≥ T ∗ in the sequel.
According to Lemma 7.8, we get

1 T 1 1  N
V1 (t) = X LX ≤ ET E = e2 . (7.43)
2 2λ2 (L ) 2λ2 (L ) i=1 i

Upon using Lemma 7.3, we have

7.3 Distributed Adaptive Finite-Time Control for Second-Order Uncertain Systems 133

N 
 vi  2−q
1 1 1
V2 (t) = 1+1/q
ς q − (vi∗ ) q dς
21−q c2 i=1 vi∗

N  1

 2−q   
1  q 
 vi − vi∗ 
1
≤  vi − (vi∗ ) q
21−q c2
1+1/q  
i=1

1 
N
1 
N
≤ 1+1/q
|δi |2−q · 21−q |δi |q = 1+1/q
δi2 . (7.44)
21−q c2 i=1 c2 i=1

From the definition of V3 (t) given in (7.32), it is not hard to see that

1 
N
1 
N
V3 (t) ≤ 1+1/q
c̃2f i + 1+1/q
θ̃i2 , (7.45)
22−q c2 gγ1m i=1 22−q c2 gγ2m i=1

with γ1m = min{γ11 , . . . , γ1N } and γ2m = min{γ21 , . . . , γ2N }. Let


1 1 1 1
kv = max , 1+1/q , , , (7.46)
2λ2 (L ) c2 2−q
2 c2
1+1/q 2−q
gγ1m 2 c2
1+1/q
gγ2m

we then have from (7.43)–(7.44) that

N 
 
V (t) ≤ kv δi2 + ei2 + c̃2f i + θ̃i2 . (7.47)
i=1

Upon using Lemma 7.5, we then arrive at

1+q
N 
 
1+q 1+q 1+q 1+q 1+q
V (t) 2 ≤ kv 2 ei + δi + c̃ f i + θ̃i . (7.48)
i=1

η1 k d
Let c̃ = 1+q (kd and kv are given in (7.42) and (7.46), respectively, and 0 < η1 ≤ 1).
2
kv
It thus follows from (7.41) and (7.48) that
1+q
V̇ (t) ≤ −c̃V (t) 2 + d. (7.49)

  1+q
2 
Let Θ = (xi , vi ) : V (t) < d
η2 c̃
, 0 < η2 < 1 . According to Theorem 5.2
2
in [7], for any (xi , vi ) ∈
/ Θ for all t ∈ [0, tx ], it holds that V (t) ≥ (d/η2 c̃) 1+q , i.e.,
1+q
d ≤ η2 c̃V (t) 2 , for all t ∈ [0, tx ]. This fact, together with (7.49), implies that
1+q 1+q 1+q
V̇ (t) ≤ −c̃V (t) 2 + η2 c̃V (t) 2 = −(1 − η2 )c̃V (t) 2 (7.50)
134 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

2
for all t ∈ [0, tx ]. Note that V (t) ≥ (d/η2 c̃) 1+q > 0 for t ∈ [0, tx ], it thus follows
1+q
V (t0 )1− 2
from (7.50) that tx <   according to Lemma 7.1. Therefore for ∀t ≥ T ∗ ,
(1−η2 )c̃ 1− 1+q
2
with T ∗ satisfying
1+q
V (t0 )1− 2
T∗ =  , (7.51)
(1 − η2 )c̃ 1 − 1+q
2

we have
 2
d 1+q
V (t) < = ζ. (7.52)
η2 c̃

Step 6. Derive the estimation for steady-state errors of all agents.

Note that for all i ∈ {1, . . . , N }, we have
!
 " N
" 2V1 (t)
|ei | = ei ≤ #
2
ei2 ≤
i=1
λmin (Λ)
 1
2V (t) 2 d 1+q
≤ ≤ . (7.53)
λmin (Λ) λmin (Λ) η2 c̃

According to Lemma 7.2, we have |ς 1/q − (vi∗ )1/q | ≥ 21−1/q |ς − vi∗ |1/q , and thus,
if vi ≥ vi∗ , it holds that

N  
1  vi  1/q 2−q
V2 (t) ≥ 1+1/q
21−1/q ς − vi∗ dς
21−q c2 i=1 vi∗

N 

1 vi   2 −1
= 1+1/q
2(1−1/q)(2−q) ς − vi∗ q dς
21−q c2 i=1 vi∗

2(1−1/q)(2−q)    2 −1+1 vi

N
1 ∗
= ς − vi∗ q
2
q
− 1 + 1 21−q c2
1+1/q
i=1
vi

q21− q  
2 N
2
= 1+1/q
vi − vi∗ q . (7.54)
c2 i=1

If vi < vi∗ , the proof of (7.54) is similar. Then we have

7.3 Distributed Adaptive Finite-Time Control for Second-Order Uncertain Systems 135

 N  q2
  2 2
q
 2
|vi − vi∗ | = vi − vi∗ q ≤ (vi − vi∗ ) q
i=1
 1+1/q
 q2 1
(q+1) 1
(q+1)  q
c2 c22 q c22 d 1+q
≤ V2 (t) = V (t) ≤
−1 2
2 . (7.55)
η2 c̃
q q q q
q21−2/q q 2 22 q 2 2 2 −1

On the other hand, from (7.53), we see that

 q  q
2 2 d 1+q
|vi∗ | = | − c2 ei | ≤ c2
q
. (7.56)
λmin (Λ) η2 c̃

It thus follows from (7.55) and (7.56) that

⎡ ⎤
2 (q+1)
1  q  q
c2 2 2 d 1+q
|vi | ≤ |vi − vi∗ | + |vi∗ | ≤ ⎣ + c2 ⎦ (7.57)
λmin (Λ) η2 c̃
q q
q 2 2 2 −1

which then implies, for ∀i, j ∈ {1, . . . , N }, that

⎡ ⎤
2
1
(q+1)  q  q

⎣ c2 2 ⎦ d
2 1+q
|vi − v j | ≤ |vi | + |v j | ≤ 2 + c2 . (7.58)
λmin (Λ) η2 c̃
q q
q 2 2 2 −1

In addition, for all ∀i ∈ {1, . . . , N },

!
 " N 
"
|c̃ f i | = c̃ f i ≤ #
2
c̃2f i ≤ 2kγ g γ̄1 V3 (t)
i=1
   1
d 1+q
≤ 2kγ g γ̄1 V (t) ≤ 2kγ g γ̄1 . (7.59)
η2 c̃

and

  1
d 1+q
|θ̃i | ≤ 2kγ g γ̄2 , (7.60)
η2 c̃

with γ̄1 = max{γ11 , . . . , γ1N }, and γ̄2 = max{γ21 , . . . , γ2N }.

From the above analysis, we conclude that under the proposed finite-time control
scheme (7.8)–(7.11), the position and velocity errors between neighbor agents will
converge to a small region Ω1 , defined by
136 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

 1  12 (q+1)
2kv d 1+q c
Ω1 = |ei | ≤ , |vi − v j | ≤ 2 2q q −1
λmin (Λ) η2 η1 k d q 2 22
⎛ ⎞ q
 q 1+q
2
1+q 
2 2
⎝ dk v
⎠ , ∀i, j ∈ {1, . . . , N } ,
+ c2 (7.61)
λmin (Λ) η2 η1 k d

and the generalized parameter estimation converges to the region Ω1 p given as

  1
d 1+q
Ω1 p = |c̃ f i | ≤ 2kγ g γ̄1 ,
η2 c̃
  1 
d 1+q
|θ̃i | ≤ 2kγ g γ̄2 , ∀i ∈ {1, . . . , N } (7.62)
η2 c̃

in the finite time T ∗ . 

7.4 Numerical Simulations

To verify the effectiveness of the proposed finite-time control algorithms, we perform

the numerical simulation on a group of four networked autonomous surface vessels
(ASVs) [8] with nonlinear dynamics.
The dynamics of the kth (k = 1, 2, 3, 4) ASV is modeled by

Mk v̇k = f k (rk , vk ) + u k + Dk (7.63)

where Mk = diag{m x,k , m y,k , m z,k } denotes the mass matrix; rk = [xk , yk , z k ]T ,
vk = [vx,k , v y,k , vz,k ]T , and u k = [u x,k , u y,k , u z,k ]T denote the position, velocity, and
control input vector, respectively; Dk is environment disturbance, and f k (rk , vk ) rep-
resents coriolis, centripetal, and hydrodynamic damping forces and torques acting
on the body, where
⎡ ⎤ ⎡ ⎤
A x,k + A|x,k| |vx,k | −m y,k vz,k 0 vx,k
f k (rk , vk ) = ⎣ m x,k vz,k B y,k + B|y,k| |v y,k | 0 ⎦ · ⎣v y,k ⎦
0 0 C z,k + C|z,k| |vz,k | vz,k
(7.64)

In the simulation, the physical parameters are taken as: Mk = diag{600 + 6(−1)k
+ 6Δm(t), 1000 + 10(−1)k + 10Δm(t), 800 + 8(−1)k + 8Δm(t)} (Δm(t) = sin
(π t/50 − π )), A x,k = −1 + 0.1(−1)k , A|x,k| = −25 + 2.5(−1)k , B y,k = −10
+ (−1)k , B|y,k| = −200 + 20(−1)k , C z,k = −0.5 + 0.05(−1)k , and C|z,k| = −1500
+ 150(−1)k for k = 1, 2, 3, 4. The external disturbance is taken as Dk = [3 +
7.4 Numerical Simulations 137

3(−1)k sin(t/50) + 2 sin(t/10), −1 + 3(−1)k sin(t/20 − π/6)+ 2 sin(t), −5(−1)k

sin(0.1t) − sin(t + π/3)]T . The networked communication topology is undirected
and connected as shown in Fig. 7.1. Each edge weight is taken as 0.1.
The simulation objective is that the four ASVs are required to achieve con-
sensus in a finite time by using the proposed control law given in (7.8)–(7.11)
under the undirected and connected topology. The initial conditions of the ves-
sels are r1 (0) = (−1m, 0m, π/3rad), r2 (0) = (−0.5m, 1.5m, −π/4rad), r3 (0) =
(1m, 1.5m, −π/9rad), r4 (0) = (1.5m, 0.5m, π/4rad), vk (0) = (0, 0, 0) (k = 1, 2,
3, 4), respectively. The control parameters are chosen as: s = 4, c1 = 2000, and
c2 = 2. In addition, the initial values of the estimates are chosen as θ̂x,k = θ̂ y,k =
θ̂z,k = 0 and ζ̂x,k = ζ̂ y,k = ζ̂z,k = 0 for k = 1, 2, 3, 4. The simulation runs for 15s.
The trajectories of the four ASVs are represented in Fig. 7.2, which shows the trajec-
tory of each ASV from the initial position to the finial position. The position error
convergence results in x-direction, y-direction, and z-direction under the finite-time
control scheme proposed in (7.8)–(7.11) are represented in Fig. 7.3. In addition, the

Fig. 7.1 Undirected connected topology among the four ASVs

1.5

0.5

−0.5
−1.5 −1 −0.5 0 0.5 1 1.5 2

Fig. 7.2 Trajectory of each ASV from the initial position to the final position
138 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

ex,1 ex,2 ex,3 ex,4

0.3

0.2

0.1
Ex

−0.1

−0.2

0 2 4 6 8 10
Time (s)
(a) E in x-direction.
0.3 ey,1 ey,2 ey,3 ey,4
0.2

0.1

0
Ey

−0.1

−0.2

−0.3

−0.4
0 2 4 6 8 10
Time (s)
(b) E in y-direction.
ez,1 ez,2 ez,3 ez,4
0.3

0.2

0.1
Ez

−0.1

−0.2
0 2 4 6 8 10
Time (s)
(c) E in z-direction.
Fig. 7.3 Position error convergence of the four ASVs under the proposed finite-time control scheme
7.4 Numerical Simulations 139

0.2
ėx,1 ėx,2 ėx,3 ėx,4

0.1

0
Ėx

−0.1

−0.2

0 2 4 6 8 10
Time (s)
(a) Ė in x-direction.

ėy,1 ėy,2 ėy,3 ėy,4

0.2

0.1
Ėy

−0.1

−0.2

0 2 4 6 8 10
Time (s)
(b) Ė in y-direction.
0.2
ėz,1 ėz,2 ėz,3 ėz,4

0.1

0
Ėz

−0.1

−0.2

0 2 4 6 8 10
Time (s)
(c) Ė in z-direction.
Fig. 7.4 Velocity error convergence of the four ASVs under the proposed finite-time control scheme
140 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

2000 ux,1 ux,2 ux,3 ux,4

1000
Ux

−1000

−2000
0 2 4 6 8 10
Time (s)
(a) U in x-direction.
2000 uy,1 uy,2 uy,3 uy,4

1000
Uy

−1000

−2000
0 2 4 6 8 10
Time (s)
(b) U in y-direction.
2000 uz,1 uz,2 uz,3 uz,4
1500

1000

500

0
Uz

−500

−1000

−1500

−2000
0 2 4 6 8 10
Time (s)
(c) U in z-direction.
Fig. 7.5 Control inputs of the four ASVs under the proposed finite-time control scheme
7.4 Numerical Simulations 141

0.2

Ex 0

−0.2 ex,1 with c1 = 2000, c2 =2

ex,4 with c1 = 2000, c2 =2
ex,1 with c1 = 5000, c2 =4
ex,4 with c1 = 5000, c2 =4
−0.4
0 2 4 6 8 10
Time (s)
(a) E x-direction .

ėx,1 with c1 = 2000, c2 =2

0.4
ėx,4 with c1 = 2000, c2 =2
ėx,1 with c1 = 5000, c2 =4
0.3
ėx,4 with c1 = 5000, c2 =4
0.2
Ėx

0.1

−0.1
0 2 4 6 8 10
Time (s)
(b) Ė in x-direction.
Fig. 7.6 Error convergence comparison with x-direction under the finite-time control scheme with
different control parameters

velocity error convergence results in x-direction, y-direction, and z-direction under

the proposed finite-time control scheme are given in Fig. 7.4. It is observed from
both Figs. 7.3 and 7.4 that the proposed finite-time controller (7.8)–(7.11) leads to
fast convergence rate and high control accuracy. Figure 7.5 shows the control input
signals of the four ASVs in x-direction, y-direction and z-direction, respectively,
under the proposed finite-time control scheme, from which we see that the control
input signals of all the four ASVs in the three directions are continuous and bounded.
To see the effects of the control parameters on the error convergence time, we
consider two group of control parameters: s = 4, c1 = 2000 and c2 = 2; s = 4,
c1 = 5000 and c2 = 4, respectively. The position and velocity error convergence
comparison results are depicted in Fig. 7.6. It is observed from the comparison results
that the finite time T ∗ specified in (7.13) can be adjusted smaller by choosing larger
controller parameters c1 and c2 , but it will result in larger initial control effort.
142 7 Finite-Time Consensus for Systems with Second-Order Uncertain Dynamics …

Remark 7.1 It should be mentioned that the control parameter c1 is set relatively
large as compared with c2 in the simulation. This is because max{m x,k , m y,k , m z,k } =
1010, leading to the equivalent control gain 1/1010, thus in order to have enough
power to steer the system, the designed control parameter c1 has to be chosen large
enough to counteract the small control gain 1/1010.

7.5 Notes

The majority of the research on leaderless consensus control of nonlinear MAS has
been focused on the systems with nonlinearities that can be linearly parameterized,
and the convergence time is not finite. The work in this chapter explicitly addressed
the finite-time leaderless consensus control of nonlinear MAS with second-order
dynamics in the presence of unknown time-varying gain and non-parametric uncer-
tainties. It is shown that under the proposed control scheme, not only the finite-
time convergence is realized but also the unknown time-varying gain and the non-
parametric uncertainties are compensated successfully.
Not considered in this chapter is the local directed communication topology con-
dition, which obviously further complicates the underlying problem significantly. For
agents with second-order dynamics in the presence of non-parametric uncertainties
under local and one-way directed communication topology, the finite-time leaderless
consensus control is still a challenging problem, thus worthy of further studying. We
give a solution to this problem in the next chapter.

References

1. Bhat, S.P., Bernstein, D.S.: Continuous finite-time stabilization of the translational and rotational
double integrators. IEEE Trans. Autom. Control 43(5), 678–682 (1998)
2. Qian, C., Lin, W.: Non-Lipschitz continuous stabilizer for nonlinear systems with uncontrollable
unstable linearization. Syst. Control Lett. 42(3), 185–200 (2001)
3. Hardy, G., Littlewood, J., Polya, G.: Inequalities. Cambridge University Press, Cambridge (1952)
4. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Fault-tolerant finite time consensus for multiple
uncertain nonlinear mechanical systems under single-way directed communication interactions
and actuation failures. Automatica 63, 374–383 (2016)
5. Olfati-Saber, R., Murray, R.: Consensus problems in networks of agents with switching topology
and time-delays. IEEE Trans. Autom. Control 49(9), 1520–1533 (2004)
6. Polycarpout, M.M., Ioannout, P.A.: A robust adaptive nonllinear control design. In: American
Control Conference, pp. 1365–1369 (1993)
7. Bhat, S.P., Bernstein, D.S.: Finite-time stability of continuous autonomous systems. SIAM J.
Control Optim. 38(3), 751–766 (2000)
8. Almeida, J., Silvestre, C., Pascoal, A.M.: Cooperative control of multiple surface vessels with
discrete-time periodic communications. Int. J. Robust Nonlinear Control 22(4), 398–419 (2012)
Chapter 8
Finite-Time Consensus for Systems
with Second-Order Uncertain Dynamics
Under Directed Topology

In Chap. 7, we have explicitly addressed the finite-time leaderless consensus con-

trol problem of nonlinear MAS with second-order dynamics in the presence of
unknown time-varying gain and non-parametric uncertainties under the local and
undirected communication condition. By considering that in most practical applica-
tions, the networked communication among the subsystems is not only local but also
not bidirectional, and it is highly desirable to consider the local and one-way directed
communication condition. For agents with second-order dynamics in the presence
of non-parametric uncertainties under local and one-way directed communication
topology, the finite-time leaderless consensus control problem is interesting and also
challenging, worthy of further studying. Compared with the control results derived
in Chap. 7, the finite-time leaderless consensus of second-order nonlinear MAS with
non-parametric uncertainties under one-way directed communication topology is
much more challenging, where the one-way directed communication topology obvi-
ously complicates the underlying problem significantly. This is mainly because the
symmetric property does not hold for the directed Laplacian, which imposes a sig-
nificant technique challenge to extend the existing finite-time adaptive consensus
control methods derived from undirected graph to directed graph. In this chapter, we
attempt to provide a solution to this problem under directed graph topology. Such a
solution is made possible by deriving some crucial algebraic connectivity properties
established associated with the newly defined Laplacian matrix.
Section 8.1 formulates the leaderless consensus control problem for multi-agent
systems with second-order dynamics in the presence of unknown time-varying gain
and non-parametric uncertainties under directed topology. Section 8.2 represents
some useful results that are helpful to tackle the technical difficulties arising from the
directed communication topology in the finite-time stability analysis for distributed
control. Section 8.3 addresses the leaderless consensus controller design and stability

Part of Sect. 8.2 has been reproduced from Wang, Y. J., Song, Y. D., Krstic, M., and Wen, C. Y.:
Fault-tolerant finite time consensus for multiple uncertain nonlinear mechanical systems under
single-way directed communication interactions and actuation failures. Automatica, vol. 63, pp.
374–383, 2016 © 2016 Elsevier Ltd., reprinted with permission.

© Springer Nature Switzerland AG 2019 143

Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_8
144 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

analysis for the second-order uncertain multi-agent systems under directed topology,
in which the technical difficulties arising from the non-symmetric property of the
Laplacian matrix are circumvented in the finite-time stability analysis. Section 8.4
conducts numerical simulation to confirm the effectiveness of the theoretical results
obtained in the previous sections. Section 8.5 gives some notes on the finite-time
control of the second-order uncertain multi-agent systems.

8.1 Formulation of Finite-Time Consensus

for Second-Order Uncertain Systems Under
Directed Topology

In this section, we formulate the leaderless consensus control problem for networked
multi-agent systems with second-order uncertain dynamics under directed topology,
where the agent dynamics are subjected to unknown time-varying gain and unknown
non-parametric uncertainties. In the next section, we will introduce the Lyapunov
function design for the finite-time stability analysis under directed topology, and then
we start with controller design and stability analysis to explain the fundamental idea
and the technical development of the proposed method.

8.1.1 Control Problem Formulation Under Directed Topology

The focus of this section is to address the finite-time leaderless consensus problem
of second-order dynamic multi-agent systems with unknown time-varying gain and
non-parametric uncertainties under the directed topology.
We consider the system described by

ẋi (t) = vi (t),

v̇i (t) = gi (z i , t)u i (t) + f i (z i , t) + f di (z i , t), (8.1)

where xi ∈ R, vi ∈ R, and u i ∈ R are the position state, velocity state, and control
input of the ith (i = 1, 2, . . . , N ) agent, respectively, gi denotes the control gain,
possibly time-varying and unavailable for controller design, f i denotes the system
nonlinearities,
 and f di denotes the non-vanishing uncertainties. In addition, z i =
j∈N i i {x j , v j }.


The following standard assumptions are in order.

Assumption 8.1 a. Standard assumptions for the existence of unique solutions and
global controllability of system are made; e.g., f i and f di are either continuously
differentiable or Lipschitz, and gi (unknown and time-varying) is sign-definite
(w. l. o. g. here sgn(gi ) = +1) and 0 < g i ≤ |gi (·)| ≤ ḡi < ∞ with g i , ḡi being
unknown finite constants, which ensures the global controllability of the system.
8.1 Formulation of Finite-Time Consensus for Second-Order Uncertain … 145

b. It is assumed that f i (z i , t) is unknown and non-parameterized. Thus, they are

not available in the finite-time control design in this section. Moreover, certain
crude structural information is available to allow that | f i (z i , t)| ≤ c f i ϕi (z i ) for
all t ≥ t0 , where c f i ≥ 0 is an unknown finite constant and ϕi (z i ) is a known
scalar function, and ϕi (z i ) is bounded if z i is bounded.
c. It is assumed that f di (z i , t) is unknown and non-vanishing, and f di (z i , t) is
bounded by | f di (z i , t)| ≤ θi < ∞ with θi being some unknown finite constant.
In addition, it is assumed for ease of notation that the agents’ states are scalars,
i.e., xi ∈ R, vi ∈ R. If the states are vectors, i.e., xi ∈ Rn and vi ∈ Rn , the results of
this chapter can be extended by using the standard method involving the Kronecker
products.
The local neighborhood state error for node i is

ei = ai j (xi − x j ) (8.2)
j∈N i

Let E = [e1 , . . . , e N ]T , X = [x1 , . . . , x N ]T such that E = L X .

The distributed finite-time leaderless consensus control design problem con-
fronted herein is as follows: (1) design control schemes for all agents i in G under
the directed communication topology condition to make an agreement in finite time.
The control schemes must be distributed in the sense that they can only depend on
local information about the agent itself and its neighbors in graph, and meanwhile,
the control schemes should be endowed with the finite-time quality; (2) overcome
the technical difficulty in constructing the Lyapunov function arising from asym-
metric property of the original Laplacian matrix under the directed communication
topology condition; and (3) tackle unknown and time-varying control gains, and
accommodate unknown non-parametric uncertainties and unknown non-vanishing
nonlinearities by using the control method based on fractional-order adaptive laws.

8.2 Preliminaries

8.2.1 Useful Lemmas Related to Finite-Time Control

The following lemmas are useful for establishing the finite-time stability results of
the MAS.
Lemma 8.1 ([1]) Suppose there exists a continuously differentiable function
V (x, t) : U × R+ → R (U ⊂ R N is an open neighborhood of the origin), a real
number c > 0 and 0 < α < 1, such that V (x, t) is positive definite and V̇ (x, t) +
cV (x, t)α ≤ 0 on U0 (U0 ⊂ U), then V (x, t) is locally in finite-time convergent with
a finite settling time satisfying T ∗ ≤ V (x(0))
1−α

c(1−α)
, such that for any given initial state
x(t0 ) ∈ U0 /{0}, limt→T ∗ V (x, t) = 0 and V (x, t) = 0 for t ≥ T ∗ .
146 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

Lemma 8.2 ([2]) If h = h 2 / h 1 ≥ 1, where h 1 , h 2 > 0 are odd integers, then |x −

y|h ≤ 2h−1 |x h − y h |.
Lemma 8.3 ([2]) If 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are odd integers, then
|x h − y h | ≤ 21−h |x − y|h .
Lemma 8.4 ([2]) For x, y ∈ R, if c, d > 0, then |x|c |y|d ≤ c/(c + d)|x|c+d + d/
(c + d)|y|c+d .
N
Lemma 8.5 ([3]) For xi ∈ R, i = 1, 2, . . . , N , 0 < h ≤ 1, then ( i=1 |xi |)h ≤
N  N
i=1 |x i | ≤ N ( i=1 |xi |) .
h 1−h h

Lemma 8.6 ([4]) For x ∈ R, y ∈ R, 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are pos-

itive odd integers, then x h (y − x) ≤ 1+h
1
(y 1+h − x 1+h ).
Lemma 8.7 For the generalized weight parameter estimation error •˜ = • − g •ˆ and
the constant q = 4h−1
4h+1
(h ∈ Z+ ), it holds that
  4q
3q−1 1 + q 1−3q 8q(q−1) 2(q−1)
•˜ •ˆ 1+q ≤ g 1+q 2 (1+q)2 − 2 1+q •˜ 1+q
4q
⎛ −16q(q−1)2
⎞
3q − 1 (1 + q)2 (1+q)3
+ ⎝1 − 2 1+q + ⎠ • 1+q .
2q−2 4q
+ (8.3)
4q 4q

Proof Firstly, by noting that •˜ = • − g •ˆ , we get

3q−1
 3q−1 1−3q 3q−1
•˜ •ˆ 1+q = •˜ g −1 (• − •˜ )
1+q
= g 1+q •˜ (• − •˜ ) 1+q . (8.4)

According to Lemma 8.6, we let x = • − •˜ , y = •, and then y − x = •˜ , and we

then have
3q−1 1 + q 1+q
4q 4q

•˜ (• − •˜ ) 1+q ≤ • − (• − •˜ ) 1+q . (8.5)
4q

4q 4h+1 −1
3× 4h−1
Note that 1+q
=1+ 3q−1
1+q
and 3q−1
1+q
= 1+ 4h−1
= 2h−1
2h
(h ∈ Z+ ), we thus see
4h+1
3q−1
that both the function g(s) = s (s ∈ R) and function f (s) = s 1+q (s ∈ R) are non-
decreasing, and then we have
4q 3q−1 3q−1
(• − •˜ ) 1+q = (• − •˜ ) (• − •˜ ) 1+q = |• − •˜ | |• − •˜ | 1+q , (8.6)

from which, (8.5) becomes

3q−1 1 + q 1+q
4q 3q−1

•˜ (• − •˜ ) 1+q ≤ • − |• − •˜ | |• − •˜ | 1+q . (8.7)
4q
8.2 Preliminaries 147

Upon using Lemma 8.3, we arrive at

1 + q 1+q  3q−1 
3q−1 4q 2q−2  3q−1 
•˜ (• − •˜ ) 1+q ≤ • − 2 1+q |• − •˜ | • 1+q − •˜ 1+q  . (8.8)
4q
3q−1
Similarly, by noting that the function g(s) = s (s ∈ R) and function f (s) = s 1+q
 3q−1 3q−1 
(s ∈ R) are non-decreasing, we then have |• − •˜ | • 1+q − •˜ 1+q 
 3q−1 3q−1

= (• − •˜ ) • 1+q − •˜ 1+q , and (8.8) becomes

3q−1 1+q 4q 2q−2 4q 2q−2 3q−1
•˜ (• − •˜ ) 1+q ≤ • 1+q −2 1+q • 1+q +2 1+q • 1+q •˜
4q
2q−2 3q−1 2q−2 4q
+ 2 1+q • •˜ 1+q − 2 1+q •˜ 1+q . (8.9)

Upon using Lemma 8.4, we get that

2q−2 3q−1
 2q−2  3q−1 1 + q  2q−2  1+q
4q
3q − 1 1+q
4q
2 1+q •˜ • 1+q = 2 1+q •˜ • 1+q ≤ 2 1+q •˜ + • , (8.10)
4q 4q

and

2q−2 3q−1 2q−2

 2q−2  3q−1 2q−2 1−3q
+ 2 1+q · 1+q
1+q
2 1+q • •˜ 1+q =2 1+q • 2 1+q •˜
2q−2 2−2q
 2q−2  3q−1
=2 1+q · 1+q • 2 1+q •˜
1+q

  2  1+q
3q − 1  2q−2  1+q
4q
1+q
4q
−4 q−1
≤ 2 1+q • + 2 1+q •˜ . (8.11)
4q 4q

By inserting (8.10)–(8.11) into (8.9), we then arrive at


1+q 1 + q  2q−2  1+q 3q − 1 1+q
4q
3q−1 4q 2q−2 4q 4q
•˜ (• − •˜ ) 1+q ≤ • 1+q −2 1+q • 1+q + 2 1+q •˜ + •
4q 4q 4q
  2  1+q
3q − 1  2q−2  1+q
4q
1+q
4q
−4 q−1 2q−2 4q
+ 2 1+q • + 2 1+q •˜ − 2 1+q •˜ 1+q
4q 4q

1 + q  8q(q−1) 2(q−1)
 4q
≤ 2 (1+q)2 − 2 1+q •˜ 1+q
4q
⎛ −16q(q−1)2
⎞
3q − 1 (1 + q)2 (1+q)3
+ ⎝1 − 2 1+q + ⎠ • 1+q , (8.12)
2q−2 4q
+
4q 4q

which then implies (8.3). 

148 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

8.2.2 Lyapunov Function Design Under Directed Topology

In this subsection, we introduce the Lyapunov function design for the finite-time
stability analysis under strongly connected directed topology. In the next subsection,
we will show how to apply such Lyapunov function in the Lyapunov analysis of
the finite-time leaderless consensus problem for multi-agent systems under directed
topology.
Before moving on, we first present several useful lemmas. It is important to note
that the Laplacian matrix L is no longer symmetric under the one-way directed
topology, which imposes a significant challenge for the finite-time leaderless consen-
sus control design because one cannot use such L to construct Lyapunov function
directly. To circumvent the technical difficulty in constructing Lyapunov function
arising from asymmetric property of the original Laplacian matrix L under directed
graph, we introduce a new matrix Q [5],

Q = (diag( p)L + L T diag( p)), (8.13)

where diag( p) = diag{ p1 , . . . , p N } ∈ R N ×N > 0 with p = [ p1 , . . . , p N ]T being the

left eigenvector of L associated with its zero eigenvalue. Such defined matrix Q
is actually a Laplacian corresponding to a connected undirected graph [4], which
exhibits the property of symmetry and semi-positive definiteness with one simple
zero eigenvalue.
The significance of the feature on Q is that the original directed graph problem
can be treated as an undirected graph problem. Furthermore, such feature allows for
the development of the following two lemmas that are crucial for the construction of
the Lyapunov function design subject to directed graph.

Lemma 8.8 ([4]) For a strongly connected digraph G with Laplacian L , let Q be
defined as in (8.13), then ∀x = 0 N , x T Qx = 0 if and only if x = c1 N , where c = 0
T
is a constant. Moreover, min xx TQx
x
exists and
x =c1 N

x T Qx  N
0 < min ≤ λi (Q). (8.14)
x =c1 N x T x
i=2

Upon using Lemma 8.8, the following result can be established.

 q q T
Lemma 8.9 ([4]) ∀E q = 0 N (E q = e1 , . . . , e N , q > 0), there exists a finite con-
stant km > 0 such that

(E q )T Q E q
≥ km . (8.15)
(E q )T E q
8.2 Preliminaries 149

According to the above analysis, we construct the Lyapunov function related to

the directed graph as

1  1+q T 1+q
V = E 2 diag( p)E 2 , (8.16)
1+q
 1+q
1+q
1+q T
 T
where E 2 = e1 2 , . . . , e N2 . Hereafter, we let E h = e1h , . . . , ehN with
h ∈ R+ .

8.3 Distributed Adaptive Finite-Time Control Design

and Stability Analysis Under Directed Topology

The control objective in this section is to design a distributed adaptive finite-time

controller under the directed topology condition such that the impacts arisen from
the asymmetric property of the original Laplacian matrix under directed topology
can be removed, meanwhile the unknown time-varying control gain, unmodeling
nonlinearities, and non-vanishing uncertainties can be compensated, and finally the
leaderless consensus of the networked uncertain multi-agent systems with second-
order uncertain dynamics is ensured in finite time.
The control input for each ith (i = 1, . . . , N ) agent is designed to consist of two
parts: (1) the negative feedback control term u 0i and (2) the compensation control
term u ci , which is of the following form

u i = u 0i + u ci , (8.17)

in which the negative feedback control term u 0i is generated by

2q−1
u 0i = −c1 δi , (8.18)

and the compensation control term u ci is generated by

u ci = −ĉ f i ϕi tanh(δi ϕi /τi ) − θ̂i tanh(δi /τi ) (8.19)

with the updated laws

3q−1

ĉ˙ f i = −γ1i σ1i ĉ f1+q

i + γ1i ϕi tanh(δi ϕi /τi )δi ,

θ̂˙i = −γ2i σ2i θ̂i 1+q + γ2i tanh(δi /τi )δi ,

3q−1

(8.20)
150 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

1
1
where δi = viq − (vi∗ ) q with vi∗ = −c2 ei and q = 4h−1
q
4h+1
(h ∈ Z+ ), c1 , c2 > 0 are
design parameters, ĉ f i and θ̂i are, respectively, the estimations of c f i and θi (the
parameters to be estimated), ϕi (·) is the scalar and readily computable function as
given in Assumption 8.1, and τi , γ1i , γ2i , σ1i , and σ2i are positive design parameters
chosen by the designer.

Theorem 8.1 (Distributed Adaptive Finite-Time Control for Leaderless Consensus

under Directed Topology) Consider the nonlinear MAS as described by (8.1) under
Assumption 8.1. Let the communication topology be directed and strongly connected.
If the distributed control laws (8.17)–(8.20) are applied, then finite-time leaderless
consensus is achieved in that

(1) The leaderless consensus positive errors ei (i = 1, . . . , N ) and velocity errors

converge to a small residual set Ω1 defined by

   1+q
1
  2q1   1+1/q  1+q
q

1+q d c2 (1 + q)
Ω1 = |ei | ≤ kv , |vi − v j | ≤ 2
p η2 η1 k d 2q−1/q q

  1+q  

q 2q

1+q  dkv1+q
+ c2 , ∀i, j ∈ {1, . . . , N } , (8.21)
p η2 η1 k d

in a finite time T ∗ satisfying

2q
1−q

∗ V (t0 ) 1+q kv1+q (1 + q)

T = , (8.22)
(1 − η2 )η1 kd (1 − q)

where 0 < η1 ≤ 1, 0 < η2 < 1, V (t0 ) is known, and kd and kv are given, respec-
tively, in (8.49) and (8.54), which are explicitly computable.
(2) The generalized parameter estimate errors c̃ f i and θ̃i converge to a small set
Ω1 p defined by

    1+q
1+1/q d 4q
Ω1 p = |c̃ f i | ≤ 2 c2
2−q g γ̄1 kv ,
η2 η1 k d
   1+q 
1+1/q d 4q
|θ̃i | ≤ 2 c2
2−q g γ̄2 kv , ∀i ∈ {1, . . . , N } (8.23)
η2 η1 k d

in the finite time T ∗ .

(3) All signals in the closed-loop system remain uniformly bounded.
8.3 Distributed Adaptive Finite-Time Control Design and Stability Analysis … 151

Proof The proof of the result can be done by the following six steps.
Step 1. Construct the distributed part of the Lyapunov function candidate as

1  1+q T 1+q
V1 (t) = E 2 diag( p)E 2 . (8.24)
1+q

Taking the time derivative of V1 (t) yields that

 T  T
V̇1 (t) = E q diag( p) Ė = E q (diag( p)L )v (8.25)

where v = Ẋ . Let v∗ = −c2 E q be the virtual control of v, with c2 being a finite

design constant. Upon using Lemma 8.9, we have
 T  T
V̇1 (t) = − c2 E q (diag( p)L )E q + E q (diag( p)L )(v − v∗ )
c2 km  q T q  q T
≤− E E + E (diag( p)L )(v − v∗ )
2
c2 km  2q   
N N N
vi − vi∗
q
=− ei + ji e j , (8.26)
2 i=1 i=1 j=1

1 1
where ji is the ( j, i)th element of diag( p)L. By recalling that δi = (vi ) q − (vi∗ ) q ,
we get from Lemmas 8.3 and 8.4 that


N
 
N 
N 
N
vi − vi∗
q
ji e j ≤ 21−q |δi |q | ji ||e j |
q

i=1 j=1 i=1 j=1

⎡ 2 ⎛ N ⎞2 ⎤

N 
N
1 N 
≤21−q max |δi |q |ei |q ≤ 21−q max ⎣ |δi |q + ⎝ |e j |q ⎠ ⎦
i=1 j=1
2 i=1 j=1
" N #
1  N N 
|ei |2q = 2−q N max
2q 2q
≤21−q × max N |δi |2q + N δi + ei ,
2 i=1 i=1 i=1
(8.27)
 2

N 
N
where max = max j,i∈{1,...,N } | ji |, and the fact that x ≤N x 2 has been
i=1 i=1
used. By substituting (8.27) into (8.26), it thus follows that

N  
c2 km  2q 
N
e + 2−q N
2q 2q
V̇1 (t) ≤ − max δi + ei . (8.28)
2 i=1 i i=1
152 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

Step 2. Define the second part of the Lyapunov function candidate by adding a
power integrator

N $
 vi  
1 1 1
V2 (t) = 1+1/q
ς q − (vi∗ ) q dς (8.29)
21−q c2 i=1 vi∗

which is positive semi-definite and C 1 [2]. Taking the derivative of V2 (t) yields

N  1
 
1 1
V̇2 (t) = 1+1/q
viq − (vi∗ ) q v̇i
21−q c2 i=1
$ vi   1
1 1 1−1 d(−(vi∗ ) q )
+ ς q − (vi∗ ) q dς ·
vi∗ dt
 
1
N
  1
= 1+1/q
δi v̇i + vi − vi∗ · c2q ėi
21−q c2 i=1
⎡ ⎤
1 N
  1 
= ⎣δi v̇i + vi − vi∗ · c2q ai j (vi − v j )⎦ . (8.30)
1+1/q
21−q c2 i=1 j∈N i

Upon using Lemma 8.3, the second term of the right side of (8.30) becomes

1 
N
  1 
1+1/q
vi − vi∗ · c2q ai j (vi − v j )
21−q c2 i=1 j∈N i
 
 
1 

1
N
q 

≤ 1+1/q
2 1−q
|δi | · c2 
q
ai j (vi − v j )
1−q
2 c2 i=1  j∈N i 
 
 
1  q   
N
= |δi |  ai j (vi − v j ) . (8.31)
c2 i=1  j∈N i 


Let ā = max∀i∈{1,...,N } { ai j } and b̄ = max∀i, j{1,...,N } {ai j }. We then see that
j∈N i

  ⎛ ⎞
 
1 N
   1 N 
|δi |q  ai j (vi − v j ) ≤ |δi |q ⎝ā|vi | + b̄ |v j |⎠ . (8.32)
c2 i=1 
j∈N i  c2 i=1 j∈N i

Upon using Lemmas 8.3 and 8.4, we have

8.3 Distributed Adaptive Finite-Time Control Design and Stability Analysis … 153
   
|δi |q |v j | ≤|δi |q v j − v∗j  + |δi |q v∗j  ≤ 21−q |δi |q |δ j |q + c2 |δi |q |e j |q
  c2  2q 
≤2−q |δi |2q + |δ j |2q + |δi | + |e j |2q . (8.33)
2
Upon using (8.33) to (8.32), we have
⎛ ⎞
1  
N
q⎝
|δi | ā|vi | + b̄ |v j |⎠
c2
i=1 j∈N i
N 
1  āc2  2q 
≤ ā21−q |δi |2q + |δi | + |ei |2q
c2 2
i=1
   b̄c   
+ b̄2−q
2
|δi |2q + |δ j |2q + |δi |2q + |e j |2q
2
j∈N i j∈N i
N 
1  āc2  2q  b̄c2 N̄  2q 
≤ ā21−q |δi |2q + |δi | + |ei |2q + 21−q b̄ N̄ |δi |2q + |δi | + |ei |2q
c2 2 2
i=1

ā + b̄ N̄   1−q c2  
N
c2
= 2 + |δi |2q + |ei |2q , (8.34)
c2 2 2
i=1

where N̄ denotes the maximum number of the in-degree and out-degree of each ith
agent for all i ∈ {1, . . . , N }. By combining (8.31), (8.32), and (8.34), the second
term of the right side of (8.30) can be further written as

1 
N
  1 
1+1/q
vi − vi∗ · c2q ai j (vi − v j )
21−q c2 i=1 j∈N i

ā + b̄ N̄ 
N  c2  2q c2 2q 
≤ 21−q + |δi | + |ei | . (8.35)
c2 i=1
2 2

By applying the control laws given in (8.17)–(8.19) to the first term of the right
hand of (8.30), we have


N 
N
δi v̇i = δi (gi u i + f i + f di )
i=1 i=1
N 
 2q
≤ − c1 gi δi − δi gi ĉ f i ϕi tanh(δi ϕi /τi ) − δi gi θ̂i tanh(δi /τi )
i=1

+ |δi |c f i ϕi + |δi |θi

N 
 2q
≤ − c1 gδi − g ĉ f i δi ϕi tanh(δi ϕi /τi ) − g θ̂i δi tanh(δi /τi )
i=1
154 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

+ c f i δi ϕi tanh(δi ϕi /τi ) + 0.2785τi c f i + θi δi tanh(δi /τi ) + 0.2785τi θi

N 
  
2q
= − c1 gδi + 0.2785τi (c f i + θi ) + c f i − g ĉ f i δi ϕi tanh(δi ϕi /τi )
i=1
 
+ θi − g θ̂i δi tanh(δi /τi ) (8.36)

where we have used the fact that 0 ≤ |s| − s · tanh(s/k) ≤ 0.2785k [6].
By substituting (8.35) and (8.36) into (8.30), we arrive at

   
N
2q
N
2q
N
1  
V̇2 (t) ≤k1 δi + k2 ei + 0.2785τi c f i + θi
1−q c1+1/q
i=1 i=1 i=1 2 2
   
+ c f i − g ĉ f i δi ϕi tanh(δi ϕi /τi ) + θi − g θ̂i δi tanh(δi /τi ) (8.37)

where

c1 g ā + b̄ N̄  1−q c2 
k1 = − 1+1/q
+ 2 + ,
21−q c2 c2 2
ā + b̄ N̄
k2 = . (8.38)
2

Step 3. Note that in (8.37) the parameter estimation error of the form •˜ = • −
g •ˆ is involved, which motivates us to introduce the generalized weight parameter
estimation errors c̃ f i (i ∈ {1, . . . , N }) and θ̃i as follows

c̃ f i = c f i − g ĉ f i , θ̃i = θi − g θ̂i (8.39)

with which we introduce the third part of the Lyapunov function candidate as


N
c̃2f i 
N
θ̃i2
V3 (t) = + (8.40)
i=1
2kγ gγ1i i=1
2kγ gγ2i

1+1/q
where kγ = 21−q c2 .
By applying the adaptive laws for ĉ f i and θ̂i given in (8.20), we get the derivative
of V3 as

 ĉ˙ f i  θ̂˙i
N N
c̃ f i θ̃i
V̇3 (t) = (− ) + (− )
i=1
kγ γ1i k
i=1 γ
γ2i
8.3 Distributed Adaptive Finite-Time Control Design and Stability Analysis … 155

N 
c̃ f i 3q−1

= i − δi ϕi tanh(δi ϕi /τi )
σ1i ĉ f1+q
i=1 γ
k
N 
θ̃i 3q−1

+ σ2i θ̂i 1+q − δi tanh(δi /τi )

i=1 γ
k

N
σ1i 3q−1 
N
σ2i 3q−1

= i +
c̃ f i ĉ f1+q θ̃i θ̂i 1+q
i=1
kγ i=1
kγ
 1  
N
− c f i − g ĉ f i δi ϕi tanh(δi ϕi /τi )
k
i=1 γ

 1  
N
− θi − g θ̂i δi tanh(δi /τi ). (8.41)
k
i=1 γ

Sep 4. Define the Lyapunov function candidate as

V (t) = V1 (t) + V2 (t) + V3 (t) (8.42)

where V1 (t), V2 (t), and V3 (t) are given in (8.24), (8.29), and (8.40), respectively.
By combining (8.28), (8.37), and (8.41), we then arrive at


N
2q

N
2q

N
σ1i 3q−1

V̇ (t) ≤ − k3 δi − k4 ei + c̃ f i ĉ f1+q
i
i=1 i=1 i=1
kγ

N
σ2i 3q−1 N
0.2785τi (c f i + θi )
+ θ̃i θ̂i 1+q + (8.43)
i=1
kγ i=1
kγ

where

c1 g ā + b̄ N̄  1−q c2 
k3 = −2−q N max + 1+1/q
− 2 + ,
21−q c2 c2 2
c2 km ā + b̄ N̄
k4 = − 2−q N max − . (8.44)
2 2
1+1/q −1  −q
Thus, c1 and c2 can be chosen as c1 > 21−q c2 g 2 N max +
 1−q c2   
−q
ā+b̄ N̄
c2
2 + 2 and c2 > km 2 N max + 2
2 ā+b̄ N̄
such that k3 and k4 > 0.
Upon using Lemma 8.7, it is straightforward that
  4q
3q−1
1 + q 1−3q 8q(q−1) 2(q−1)
c̃ f i ĉ f1+q
i ≤ g 1+q 2 (1+q)2 − 2 1+q c̃ 1+q
fi
4q
156 8 Finite-Time Consensus for Systems with Second-Order Uncertain …
⎛ −16q(q−1)2
⎞
3q − 1 (1 + q)2 (1+q)3 4q

+ ⎝1 − 2 ⎠ c 1+q
2q−2
1+q + + fi , (8.45)
4q 4q
  4q
3q−1
1 + q 1−3q 8q(q−1) 2(q−1)
θ̃i θ̂i 1+q
≤ g 1+q 2 (1+q)2 − 2 1+q θ̃i1+q
4q
⎛ −16q(q−1)2
⎞
3q − 1 (1 + q)2 (1+q)3 4q

+ ⎝1 − 2 1+q + ⎠ θi 1+q .
2q−2
+ (8.46)
4q 4q

By substituting (8.45) and (8.46) into (8.43), one gets


N
2q

N
2q

N 4q 
N 4q

V̇ (t) ≤ −k3 δi − k4 ei − k5 c̃ f i − k6
1+q
θ̃i 1+q + d (8.47)
i=1 i=1 i=1 i=1

where
1−3q
 2(q−1) 8q(q−1) 
σ 1 (1 + q)g 1+q 2 1+q − 2 (1+q)2
k5 = 1+1/q
,
23−q c2 q
1−3q
 2(q−1) 8q(q−1) 
σ 2 (1 + q)g 1+q 2 1+q − 2 (1+q)2
k6 = 1+1/q
,
23−q c2 q
⎛ ⎞
N 

−16q(q−1)2

⎝1 − 2 1+q + 3q − 1 + (1 + q)2
(1+q)3
⎠ g 1+q
2q−2 1−3q
d=
i=1
4q 4q
4q 4q
 
f i + σ2i θi
(1 + q)(σ1i c 1+q 1+q
) 0.2785τi c f i + θi
× 1+1/q
+ 1+1/q
< ∞, (8.48)
23−q c2 q 21−q c2

where σ 1 = min{σ11 , . . . , σ1N } and σ 2 = min{σ21 , . . . , σ2N }. Note that 2(q−1)

1+q
−
8q(q−1)
(1+q)2
= 2(q−1)(1−3q)
(1+q)2
= 16(2h−1)
(4h−1)2
and h ∈ Z + . Thus, k5 > 0, and then k6 > 0. By
introducing kd as

kd = min{k3 , k4 , k5 , k6 }, (8.49)

we further represent (8.47) as

N 
 4q 4q

2q 2q
V̇ (t) ≤ −kd δi + ei + c̃ 1+q
f i + θ̃i
1+q
+ d. (8.50)
i=1

Step 5. We prove that there exists a finite time T ∗ > 0 and a bounded constant
0 < ζ < ∞ such that V (t) < ζ when t ≥ T ∗ in the sequel.
8.3 Distributed Adaptive Finite-Time Control Design and Stability Analysis … 157

By recalling the definition of V1 (t) given in (8.24), we have

1  1+q T p̄  1+q
N
1+q
V1 (t) = E 2 diag( p)E 2 ≤ e , (8.51)
1+q 1 + q i=1 i

where p̄ = max{ p1 , . . . , p N }, and p = [ p1 , . . . , p N ]T is the left eigenvector of L

associated with its zero eigenvalue. Upon using Lemma 8.3, we have

N  1
 
1  q  
V2 (t) ≤ v − (v∗ ) q1  vi − v∗ 
1+1/q  i i  i
21−q c2 i=1

1 
N
1 
N
1+q
≤ 1+1/q
|δi | · 21−q |δi |q = 1+1/q
δi . (8.52)
21−q c2 i=1 c2 i=1

From the definition of V3 (t) given in (8.40), it is not hard to see that

1 
N
1 
N
V3 (t) ≤ 1+1/q
c̃2f i + 1+1/q
θ̃i2 , (8.53)
22−q c2 gγ 1 i=1 22−q c2 gγ 2 i=1

with γ 1 = min{γ11 , . . . , γ1N } and γ 2 = min{γ21 , . . . , γ2N }. Let

% &
p̄ 1 1 1
kv = max , 1+1/q , , , (8.54)
1 + q c2 2−q
2 c2
1+1/q 2−q
gγ 1 2 c2
1+1/q
gγ 2

we then have from (8.51) to (8.53) that

N 
 
1+q 1+q
V (t) ≤ kv δi + ei + c̃2f i + θ̃i2 . (8.55)
i=1

Upon using Lemma 8.5, we then arrive at

2q N 
 4q 4q

2q
2q 2q
V (t) 1+q ≤ kv1+q δi + ei + c̃ 1+q
fi + θ̃i
1+q
. (8.56)
i=1

η1 k d
Let c̃ = 2q (kd and kv are given in (8.49) and (8.54), respectively, and
1+q
kv
0 < η1 ≤ 1). It thus follows from (8.50) and (8.56) that
2q
V̇ (t) ≤ −c̃V (t) 1+q + d. (8.57)
158 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

1+q
Let Θ = {(xi , vi ) : V (t) < ( ηd2 c̃ ) 2q , 0 < η2 < 1}. According to Theorem 5.2 in
1+q
[7], for any (xi , vi ) ∈/ Θ for all t ∈ [0, tx ], it holds that V (t) ≥ (d/η2 c̃) 2q , i.e.,
2q
d ≤ η2 c̃V (t) 1+q , for all t ∈ [0, tx ]. This fact, together with (8.57), implies that
2q 2q 2q
V̇ (t) ≤ −c̃V (t) 1+q + η2 c̃V (t) 1+q = −(1 − η2 )c̃V (t) 1+q (8.58)
1+q
for all t ∈ [0, tx ]. Note that V (t) ≥ (d/η2 c̃) 2q > 0 for t ∈ [0, tx ], and it thus follows
2q
1−
V (t0  ) 1+q
from (8.58) that tx < 2q
 according to Lemma 8.1. Therefore for ∀t ≥ T ∗ ,
(1−η2 )c̃ 1− 1+q
with T ∗ satisfying
2q

∗ V (t0 )1− 1+q

T =  , (8.59)
(1 − η2 )c̃ 1 − 1+q2q

we have
  1+q
d 2q
V (t) < = ζ. (8.60)
η2 c̃

Step 6. Derive the estimation for steady-state errors of all agents.

Note that for all i ∈ {1, . . . , N }, we have
 N  1   1+q
1
  1  1+q 1+q 1+q
1+q 1+q
|ei | = ei ≤ ei ≤ V1 (t)
i=1
p
  1+q
1   1+q
1
 1
1+q 1+q d 2q
≤ V (t) ≤ , (8.61)
p p η2 c̃

with p = min{ p1 , . . . , p N } ( p = [ p1 , . . . , p N ]T is the left eigenvector of L associ-

ated with its zero eigenvalue). According to Lemma 8.2, we have |ς 1/q − (vi∗ )1/q | ≥
21−1/q |ς − vi∗ |1/q , and thus, if vi ≥ vi∗ ,

N $

1 vi  1/q
V2 (t) ≥ 1+1/q
21−1/q ς − vi∗ dς
21−q c2 i=1 vi∗

21−1/q    1 +1 v
N
1
=1 × ς − vi∗ q vi∗
q
+ 1 21−q c2
1+1/q
i=1
i

2q− q q
1

N
 1+ 1
= 1+1/q
vi − vi∗ q . (8.62)
c2 (1 + q) i=1
8.3 Distributed Adaptive Finite-Time Control Design and Stability Analysis … 159

If vi < vi∗ , the proof of (8.62) is similar. Then, we have

" N # 1+1/q
1

   1+1/q
1 
|vi − vi∗ | = vi − vi∗ (vi − vi∗ )1+1/q
1+1/q
≤
i=1
" 1+1/q
# 1+1/q
1 " 1+1/q
# 1+q
q
  21
c2 (1 + q) c2 (1 + q) d
≤ V2 (t) ≤ . (8.63)
2 q−1/q q 2 q−1/q q η2 c̃

On the other hand, from (8.61), we see that

  1+q
q
  21
 ∗   1+q d
v  = −c2 eq  ≤ c2 . (8.64)
i i
p η2 c̃

It thus follows from (8.63) to (8.64) that

⎡  1+q  q ⎤
 1+q
1+1/q
q
  21
c (1 + q) 1+q ⎦ d
|vi | ≤ |vi − vi∗ | + |vi∗ | ≤ ⎣ 2 + c2
2q−1/q q p η2 c̃
(8.65)

which then implies, for ∀i, j ∈ {1, . . . , N }, that

⎡  1+q  q ⎤
 1+q
1+1/q
q
  21
c (1 + q) 1+q ⎦ d
|vi − v j | ≤ |vi | + |v j | ≤ 2 ⎣ 2 + c2 .
2q−1/q q p η2 c̃
(8.66)

In addition, for all ∀i ∈ {1, . . . , N },


  N 

|c̃ f i | = c̃ f i ≤ 
2
c̃2f i ≤ 2kγ g γ̄1 V3 (t)
i=1
    1+q
d 4q
≤ 2kγ g γ̄1 V (t) ≤ 2kγ g γ̄1 , (8.67)
η2 c̃

and

   1+q
d 4q
|θ̃i | ≤ 2kγ g γ̄2 , (8.68)
η2 c̃

with γ̄1 = max{γ11 , . . . , γ1N }, and γ̄2 = max{γ21 , . . . , γ2N }.

160 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

From the above analysis, we conclude that under the proposed finite-time control
scheme (8.17)–(8.20), the position and velocity errors between neighbor agents will
converge to a small region Ω1 , defined by

   1+q
1
  2q1   1+1/q  1+q
q

1+q d c2 (1 + q)
Ω1 = |ei | ≤ kv , |vi − v j | ≤ 2
p η2 η1 k d 2q−1/q q

  1+q  

q 2q

1+q  dkv1+q
+ c2 , ∀i, j ∈ {1, . . . , N } , (8.69)
p η2 η1 k d

and the generalized parameter estimation converges to the region Ω1 p given as

    1+q
1+1/q d 4q
Ω1 p = |c̃ f i | ≤ 22−q c2 g γ̄1 kv ,
η2 η1 k d
   1+q 
1+1/q d 4q
|θ̃i | ≤ 22−q c2 g γ̄2 kv , ∀i ∈ {1, . . . , N } (8.70)
η2 η1 k d

in the finite time T ∗ . 

8.3.1 Comparison with Regular State Feedback-Based

Infinite-Time Method Under Directed Topology

Note that if we take the value of the fraction power q as 1, the control scheme (8.17)–
(8.20) reduces to the following regular state feedback-based infinite-time control

u i = −c1 δi − ĉ f i ϕi tanh(δi ϕi /τi ) − θ̂i tanh(δi /τi ), (8.71)

with the updated laws

ĉ˙ f i = −γ1i σ1i ĉ f i + γ1i ϕi tanh(δi ϕi /τi )δi ,

θ̂˙ = −γ σ θ̂ + γ tanh(δ /τ )δ ,
i 2i 2i i 2i i i i (8.72)

where δi = vi − vi∗ with vi∗ = −c2 ei .

With such regular state feedback-based control, the following result is derived.

Theorem 8.2 The second-order uncertain MAS (8.1) under Assumption 8.1 and
directed strongly connected topology with the control scheme (8.71)–(8.72) achieves
cooperative uniformly ultimately bounded (UUB) leaderless consensus in that the
8.3 Distributed Adaptive Finite-Time Control Design and Stability Analysis … 161

leaderless consensus positive errors ei (i = 1, . . . , N ) and velocity errors converge

to a small residual set Ω2 defined by
 '  ' 
2kv d 2
Ω2 = |ei | ≤ , |vi − v j | ≤ 2 2c2 + c2
2
η2 η1 pkd p
' 
dkv
× , ∀i, j ∈ {1, . . . , N } , (8.73)
η2 η1 k d

and the generalized parameter estimate errors c̃ f i and θ̃i converge to a small set Ω1 p
defined by
'

2c22 g γ̄1 kv d
Ω1 p = |c̃ f i | ≤ ,
η2 η1 k d
'

2c22 g γ̄2 kv d
|θ̃i | ≤ , ∀i ∈ {1, . . . , N } (8.74)
η2 η1 k d

where 0 < η1 ≤ 1, 0 < η2 < 1, V (t0 ) is known, and kd and kv are given, respectively,
in (8.49) and (8.54), which are explicitly computable. In addition, all signals in the
closed-loop system remain uniformly bounded.

Proof The proof can be done by letting q = 1 and following the same line as in the
proof of Theorem 8.1. 

8.4 Numerical Simulations

To verify the effectiveness of the proposed finite-time control algorithms, numerical

simulation on a group of four networked autonomous surface vessels (ASVs) [8]
with nonlinear dynamics is performed.
The dynamics of the kth (k = 1, 2, 3, 4) ASV is modeled by
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
m x,k 0 0 v̇x,k f x,k u x,k u x,k
⎣ 0 m y,k 0 ⎦ · ⎣v̇ y,k ⎦ = ⎣ f y,k ⎦ + ⎣u y,k ⎦ + ⎣d y,k ⎦ (8.75)
0 0 m z,k v̇z,k f z,k dx,k dz,k

in which Mk = diag{m x,k , m y,k , m z,k } denotes the mass matrix; rk = [xk , yk , z k ]T ,
vk = ṙk = [vx,k , v y,k , vz,k ]T , and u k = [u x,k , u y,k , u z,k ]T denote the position, veloc-
ity, and control input vector, respectively; Dk = [dx,k , d y,k , dz,k ]T is environment
162 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

disturbance; and f k = [ f x,k , f y,k , f z,k ]T represents coriolis, centripetal, and hydro-
dynamic damping forces and torques acting on the body, where

f k = Δ · vk

with
⎡ ⎤
A x,k + A|x,k| |vx,k | −m y,k vz,k 0
Δ=⎣ m x,k vz,k B y,k + B|y,k| |v y,k | 0 ⎦.
0 0 C z,k + C|z,k| |vz,k |

In the simulation, the physical parameters are taken as: Mk = diag{600+

6(−1)k + 6Δm(t), 1000 + 10(−1)k + 10Δm(t), 800 + 8(−1)k + 8Δm(t)} with
Δm(t) = sin(π t/50 − π ), A x,k = −1 + 0.1(−1)k , A|x,k| = −25 + 2.5(−1)k ,
B y,k = −10 + (−1)k , B|y,k| = −200 + 20(−1)k , C z,k = −0.5 + 0.05(−1)k , and
C|z,k| = −1500 + 150(−1)k for k = 1, 2, 3, 4. The external disturbance is taken
as Dk = [3 + 3(−1)k sin(t/50) + 2 sin(t/10), −1 + 3(−1)k sin(t/20 − π/6)+
2 sin(t), −5(−1)k sin(0.1t) − sin(t + π/3)]T . The networked communication topol-
ogy is directed and strongly connected as shown in Fig. 8.1. Each edge weight is taken
as 0.1. The left eigenvector of L associated with eigenvalue 0 is [2, 1, 1, 1]T .
The simulation objective is that the four ASVs are required to achieve con-
sensus in a finite time by using the proposed control law given in (8.17)–(8.20)
under the directed communication topology condition. The initial condition of
the vessels is r1 (0) = (−1.5m, 0m, π/3rad), r2 (0) = (−1m, 2m, −π/4rad), r3 (0) =
(1.5m, 2m, −π/9rad), r4 (0) = (2m, 1m, π/4rad), vk (0) = (0, 0, 0) (k = 1, 2, 3, 4),
respectively. The control parameters are taken as: s = 2, c1 = 2000 and c2 = 3. In
addition, the initial values of the estimates are chosen as ĉ f x,k = ĉ f y,k = ĉ f z,k = 0
and θ̂x,k = θ̂ y,k = θ̂z,k = 0 for k = 1, 2, 3, 4. The simulation runs for 15s. The trajec-
tories of the ASVs are represented in Fig. 8.2, which shows the trajectory of each ASV

Fig. 8.1 Directed

communication topology 2
among the four ASVs

3 1

4
8.4 Numerical Simulations 163

2.5

1.5

0.5

−0.5
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5

Fig. 8.2 Trajectory of each ASV from the initial position to the final position

from the initial position to the finial position. The position error convergence results
in x-direction, y-direction, and z-direction under the finite-time control scheme pro-
posed in (8.17)–(8.20) are represented in Fig. 8.3. In addition, the velocity error
convergence results in x-direction, y-direction, and z-direction under the proposed
finite-time control scheme are given in Fig. 8.4. From both Figs. 8.3 and 8.4, we see
that the consensus is achieved under the proposed finite-time control scheme with
high accuracy and fast convergence rate. Figure 8.5 shows the control input signals
of the four ASVs under the proposed finite-time control scheme.
To show the effectiveness of our proposed scheme, we make a comparison on
convergence time between the proposed control scheme (8.17)–(8.20) and the typ-
ical infinite-time-based adaptive scheme (8.71)–(8.72). The two control schemes
are applied to the same group of ASVs (8.75) with the same unknown parameters
mentioned above. The convergence comparison results are shown in Fig. 8.6. It is
observed from the comparison results that the convergence rates of the four ASVs,
including the position error convergence rates and velocity error convergence rates,
are faster under the finite-time controller compared with the infinite-time controller.
On the other hand, to see the effects of the control parameters on the error con-
vergence time, we consider two groups of control parameters: s = 2, c1 = 2000
and c2 = 3; s = 2, c1 = 5000 and c2 = 4, respectively. The position and velocity
error convergence comparison results are depicted in Fig. 8.7, which indicates that
the finite-time T ∗ specified in (8.22) can be adjusted smaller by choosing larger
controller parameters c1 and c2 , but it will result in larger initial control effort.
164 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

0.4 ex,1 ex,2 ex,3 ex,4

0.2
Ex

−0.2

−0.4
0 5 10 15
Time (s)
(a) E in x-direction.
0.2 ey,1 ey,2 ey,3 ey,4

0.1
Ey

−0.1

−0.2
0 5 10 15
Time (s)
(b) E in y-direction.
ez,1 ez,2 ez,3 ez,4
0.2

0.1
Ez

−0.1

−0.2

0 5 10 15
Time (s)
(c) E in z-direction.
Fig. 8.3 Position error convergence of the four ASVs under the proposed finite-time control scheme
8.4 Numerical Simulations 165

0.3
ėx,1 ėx,2 ėx,3 ėx,4
0.2

Ėx 0.1

−0.1

−0.2

0 5 10 15
Time (s)
(a) Ė in x-direction.
0.2
ėy,1 ėy,2 ėy,3 ėy,4

0.1
Ėy

−0.1

−0.2
0 5 10 15
Time (s)
(b) Ė in y-direction.
0.2
ėz,1 ėz,2 ėz,3 ėz,4

0.1
Ėz

−0.1

−0.2
0 5 10 15
Time (s)
(c) Ė in z-direction.
Fig. 8.4 Velocity error convergence of the four ASVs under the proposed finite-time control scheme
166 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

3000 ux,1 ux,2 ux,3 ux,4

2000

Ux 1000

−1000

−2000

−3000
0 5 10 15
Time (s)
(a) U in x-direction.
3000 uy,1 uy,2 uy,3 uy,4

2000

1000
Uy

−1000

−2000

−3000
0 5 10 15
Time (s)
(b) U in y-direction.
2000 uz,1 uz,2 uz,3 uz,4

1000
Uz

−1000

−2000
0 5 10 15
Time (s)
(c) U in z-direction.
Fig. 8.5 Control inputs of the four ASVs under the proposed finite-time control scheme
8.4 Numerical Simulations 167

0.3
ex,2 under finite-time control scheme
ex,3 under finite-time control scheme
0.2 ex,2 under traditional infinite-time control scheme
ex,3 under traditional infinite-time control scheme
0.1
Ex

−0.1

−0.2
0 5 10 15
Time (s)
(a) E in x-direction.

0.15
ėx,2 under finite-time control scheme
ėx,3 under finite-time control scheme
0.1
ėx,2 under traditional infinite-time control scheme
ėx,3 under traditional infinite-time control scheme
0.05
Ėx

−0.05

−0.1

0 5 10 15
Time (s)
(b) Ė in x-direction.
Fig. 8.6 Error convergence comparison with x-direction under the two control schemes

8.5 Notes

The work in this chapter explicitly addressed the finite-time leaderless consensus
control of nonlinear MAS with second-order dynamics in the presence of unknown
time-varying gain and non-parametric uncertainties under the one-way directed com-
munication topology. It is shown that under the proposed control scheme, not only
the finite-time convergence is realized but also the unknown time-varying gain and
the non-parametric uncertainties are compensated successfully.
The main contributions of the work in this chapter are summarized as follows:
(1) The finite-time consensus control scheme is derived for a group of networked
nonlinear systems with second-order uncertain dynamics under the directed commu-
nication topology. It is shown that all the internal signals are ensured to be uniformly
bounded and the consensus configuration error uniformly converges to a small resid-
ual set in finite time. (2) The one-way directed communication condition is taken
168 8 Finite-Time Consensus for Systems with Second-Order Uncertain …

0.2

Ex 0

−0.2 ex,1 with c1 = 2000, c2 =2

ex,4 with c1 = 2000, c2 =2
ex,1 with c1 = 5000, c2 =4
ex,4 with c1 = 5000, c2 =4
−0.4
0 2 4 6 8 10
Time (s)
(a) E x-direction .

ėx,1 with c1 = 2000, c2 =2

0.4
ėx,4 with c1 = 2000, c2 =2
ėx,1 with c1 = 5000, c2 =4
0.3
ėx,4 with c1 = 5000, c2 =4
0.2
Ėx

0.1

−0.1
0 2 4 6 8 10
Time (s)
(b) Ė in x-direction.
Fig. 8.7 Error convergence comparison with x-direction under the finite-time control scheme with
different control parameters

into consideration. The technical difficulty arising from the directed topology is tack-
led by deriving the general algebraic connectivity properties established in Lemmas
8.8–8.9 with the newly defined Laplacian matrix Q. (3) By introducing the skill-
fully defined weighting parameter estimate error and by establishing an important
inequality in Lemma 8.7, the lumped uncertainties in the system with unknown and
time-varying control gains are compensated gracefully.

References

1. Bhat, S.P., Bernstein, D.S.: Continuous finite-time stabilization of the translational and rotational
double integrators. IEEE Trans. Autom. Control 43(5), 678–682 (1998)
2. Qian, C., Lin, W.: Non-Lipschitz continuous stabilizer for nonlinear systems with uncontrollable
unstable linearization. Syst. Control Lett. 42(3), 185–200 (2001)
References 169

3. Hardy, G., Littlewood, J., Polya, G.: Inequalities. Cambridge University Press, Cambridge (1952)
4. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Fault-tolerant finite time consensus for multiple
uncertain nonlinear mechanical systems under single-way directed communication interactions
and actuation failures. Automatica 63, 374–383 (2016)
5. Zhang, H., Lewis, F.L., Qu, Z.: Lyapunov, adaptive, and optimal design techniques for coop-
erative systems on directed communication graphs. IEEE Trans. Ind. Electron. 59, 3026–3041
(2012)
6. Polycarpout, M.M., Ioannout, P.A.: A robust adaptive nonllinear control design. In: American
Control Conference, pp. 1365–1369 (1993)
7. Bhat, S.P., Bernstein, D.S.: Finite-time stability of continuous autonomous systems. SIAM J.
Control Optim. 38(3), 751–766 (2000)
8. Almeida, J., Silvestre, C., Pascoal, A.M.: Cooperative control of multiple surface vessels with
discrete-time periodic communications. Int. J. Robust. Nonlinear Control 22(4), 398–419 (2012)
Chapter 9
Finite-Time Leaderless Consensus
Control for Systems with High-Order
Uncertain Dynamics

The majority of the research on finite-time consensus control has been focused on
MAS described by single or double integrator under undirected topology. This work
explicitly addressed the problem of finite-time consensus for MAS with high-order
nonlinear uncertain dynamics under directed communication constraints. Finite-time
leaderless consensus for networked multi-agent systems (MAS) with high-order non-
linear uncertain dynamics under local communication condition is an challenging
problem due to the involvement of high-order nonlinear uncertain dynamics, local
communication constraints, and unknown time-varying control effectiveness gain.
In this chapter, based upon the locally defined consensus error and fractionally com-
posed virtual error, a number of useful intermediate results are derived, with which,
the finite-time consensus solutions are established for networked MAS with high-
order uncertain dynamics under single-way directed communication topology. By
including fraction power integration as part of the Lyapunov function candidate and
by using inductive analysis, it is shown that the proposed distributed solution is able
to achieve consensus in finite time with sufficient accuracy. The benefits and effec-
tiveness of the developed algorithm are also confirmed by numerical simulations.
Section 9.1 addresses the finite-time leaderless consensus control problem for
the high-order multi-agent systems with unknown non-affine dynamics under local
directed communication topology. Section 9.2 gives some useful lemmas and also
derives a number of useful intermediate results that are related to the finite-time
stability analysis for the high-order uncertain multi-agent systems. In Sect. 9.3, the
locally defined neighborhood error and the fractionally composed virtual error are
introduced into the control scheme, where the fraction power feedback law and
fraction power adaptive updated law for the parameter estimation are designed. The
main control results are given in Sect. 9.4, in which rigorous proof of the stability is
carried out by inductive analysis upon adding the fraction power integration term to

Part of this chapter has been reproduced from Wang, Y. J., Song, Y. D., Krstic, M., Wen, C. Y.:
Adaptive finite time coordinated consensus for high-order multi-agent systems: adjustable fraction
power feedback approach. Information sciences, vol. 372, pp. 392–406, 2016 © 2016 Elsevier Inc.,
reprinted with permission.

© Springer Nature Switzerland AG 2019 171

Y. Song and Y. Wang, Cooperative Control of Nonlinear
Networked Systems, Communications and Control Engineering,
https://doi.org/10.1007/978-3-030-04972-0_9
172 9 Finite-Time Leaderless Consensus Control for Systems …

the Lyapunov function candidate. Section 9.5 shows the comparison results between
the infinite-time method and the finite-time method. Section 9.6 gives two simulation
examples to verify the effectiveness of the developed algorithm. Section 9.7 gives
a discussion on the finite-time consensus control problem of high-order uncertain
MAS.

9.1 Finite-Time Consensus Formulation for High-Order

Uncertain Systems

In this chapter, we consider a group of networked systems consisting of N subsystems

with high-order non-affine dynamics, modeled by

ẋi,m = xi,m+1 , m = 1, . . . , n − 1
ẋi,n = f i (x̄i , u i ) + f di (x̄i , t),
yi = xi,1 , i = 1, . . . , N , (9.1)

where x̄i = [xi,1 , . . . , xi,n ]T ∈ Rn , xi,m ∈ R, u i ∈ R, and yi ∈ R are system state,

control input, and control output, respectively, f i (·) : Rn × R → R is a smooth but
unknown non-affine function, and f di (·) denotes all the uncertainties and distur-
bances acting on the ith subsystem.
Let the communication network among the N subsystems be represented by
a directed graph G = (ι, ε) and each subsystem can be seen as a node in such
graph [1], where ι = {ι1 , ι2 , . . . , ι N } denotes the set of nodes, ε ⊆ ι × ι is the set
of edges between two distinct nodes. The set of neighbors of node ιi is denoted by
Ni = {ι j ∈ ι|ε ji ∈ ε}, in which ε ji = (ι j , ιi ) ∈ ε indicates that node i can obtain
information from node j. The graph topology is represented by the weighted
adjacency matrix A = [ai j ], where ai j > 0 ⇔ ε ji ∈ ε, and ai j = 0, otherwise.
Moreover, it is assumed that aii = 0 for all i = 1, . . . , N . The in-degree matrix
B = diag(B1 , . . . , B N ) ∈ R N ×N is defined as Bi = j∈N i ai j such that the Lapla-
cian matrix is defined by L = [li j ] = B − A ∈ R N ×N .
The local position error is defined as

ei = ai j (yi − y j ), i, j ∈ {1, . . . , N } (9.2)
j∈N i

The control objective is to establish distributed consensus protocols for gen-

eral high-order nonlinear MAS in the presence of non-affine dynamics and non-
parametric uncertainties as described by (9.1) via finite-time control method such
that not only the stable cooperative output consensus is achieved, but also the finite-
time convergence is ensured with sufficient accuracy, i.e., there exists a finite time
T ∗ that can be explicitly expressed such that for all t ≥ T ∗ and all i = 1, . . . , N ,
|ei (t)| < ε̄, where ei (t) denotes the local neighborhood error, ε̄ > 0 is a constant
9.1 Finite-Time Consensus Formulation for High-Order Uncertain Systems 173

which can be made sufficiently small, and meanwhile, all the internal signals in the
closed-loop systems are bounded.
Note that by non-affine dynamics, the control action enters into the system through
an implicit and uncertain way, making the controller design much more challenging.
To proceed, the non-affine dynamic model (9.1) is first transformed into an affine
form by using the mean value theorem [2], that is, there exists a constant ξi ∈ (0, u i )
(i ∈ {1, . . . , N }) such that

∂ f i (x̄i , ξi )
f i (x̄i , u i ) = f i (x̄i , 0) + ui . (9.3)
∂u i

Let gi (x̄i , ξi ) = ∂ f i (x̄i , ξi )/∂u i denote the control effectiveness gain,

and L f i (x̄i , t) = f i (x̄i , 0) + f di (x̄i , t) denote the lumped uncertainties. By substi-
tuting (9.3) into (9.1), one gets

ẋi,n = gi (x̄i , ξi )u i + L f i (x̄i , t). (9.4)

The following fundamental assumptions are in order for the controller design.
Assumption 9.1
a. The directed communication network G is strongly connected.
b. The control effectiveness gain gi (·) is unknown and time-varying but bounded
away from zero, that is, there exist unknown constants gli and gui such that
0 < gli ≤ |gi (·)| ≤ gui < ∞ and gi (·) is sign-definite (we assume sgn(gi ) = +1
in this paper without loss of generality).
c. For the lumped uncertainties L f i (·), there exists an unknown constant αi ≥ 0 and
a known scalar function ψi (x̄i ) ≥ 0 such that |L f i (·)| ≤ αi ψi (·). In addition, if
x̄i is bounded, so is ψi (x̄i ).
It should be addressed that in Assumption 9.1-b, although gli is unknown, we can
always estimate its lower bound, i.e., g < gli for some constant g > 0. In addition,
Assumption 9.1-b is commonly imposed in most existing works in addressing the
tracking control problem, for instance [3–8]. Assumption 9.1-c is related to the
extraction of the core information from the nonlinearities of the system, which can
be readily done for any practical system with only crude model information [9].

9.2 Useful Lemmas and Intermediate Results

In this section, we represent some useful lemmas and derive a number of useful
intermediate results that play a crucial role in the finite-time stability analysis for the
high-order uncertain multi-agent systems.
Lemma 9.1 ([10]) Suppose there exists a continuously differentiable function
V (x, t) : U × R+ → R (U ⊂ R N is an open neighborhood of the origin), a real
174 9 Finite-Time Leaderless Consensus Control for Systems …

number c > 0 and 0 < α < 1, such that V (x, t) is positive definite and V̇ (x, t) +
cV (x, t)α ≤ 0 on U0 (U0 ⊂ U), then V (x, t) is locally in finite-time convergent with
a finite settling time satisfying T ∗ ≤ V (x(0))
1−α

c(1−α)
, such that for any given initial state
x(t0 ) ∈ U0 /{0}, limt→T ∗ V (x, t) = 0 and V (x, t) = 0 for t ≥ T ∗ .

Lemma 9.2 ([11]) Let c, d be positive real numbers and γ (x, y) > 0 a real-valued
−c/d
function. Then, |x|c |y|d ≤ cγ (x,y)|x| + dγ (x,y)c+d |y| .
c+d c+d

c+d

Lemma 9.3 ([11]) If h = h 2 / h 1 ≥ 1, where h 1 , h 2 > 0 are odd integers, then |x −

y|h ≤ 2h−1 |x h − y h |.

Lemma 9.4 ([11]) If 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are odd integers, then
|x h − y h | ≤ 21−h |x − y|h .
N
Lemma 9.5 ([12]) For xi ∈ R, i = 1, 2, . . . , N , 0 < h ≤ 1, then ( i=1 |xi |)h ≤
N  N
i=1 |x i | ≤ N ( i=1 |xi |) .
h 1−h h

Lemma 9.6 ([7]) For x ∈ R, y ∈ R, 0 < h = h 1 / h 2 ≤ 1, where h 1 , h 2 > 0 are pos-

itive odd integers, then x h (y − x) ≤ 1+h
1
(y 1+h − x 1+h ).

Lemma 9.7 Let Ξ = 1/2(P L + L T P), where P = diag{ p} and p = [ p1 , . . . , p N ]T

is the left eigenvector of L associated with the zero eigenvalue. With the graph sat-
isfying Assumption 9.1-a, ∀E s = 0 N , there exists a constant ks > 0 such that

(E s )T Ξ E s ≥ ks (E s )T E s , (9.5)

where s = (4ln − 1)/(4ln + 1), l ∈ Z + .

Proof Before moving on, we first give an important property on Ξ , which has been
proved in our previous work [7]: ∀X = 0 N , X T Ξ X = 0 if and only if X = c1 N with c
XT Ξ X
being a nonzero constant, and moreover, 0 < min XT X
≤ λ2 (Ξ ) (we denote
X =c1 N ,X =0 N
by λi (Ξ ) the eigenvalue of Ξ ). In the following, we prove that indeed E s = c1 N , and
then min (E(E h) )TΞEEh > 0. Let ks = min (E(E h) )TΞEEh , then it holds that (E h )T Ξ E h ≥
h T h h T h

E h =0 N E h =0 N
ks (E h )T E h . Note that p T E = p T L(y − ) = 0, i.e., p1 e1,1 + · · · + p N e N ,1 = 0,
thus for E = 0 N , it is impossible that sgn(ei,1 ) = 1 (or sgn(ei,1 ) = −1) for all i =
1, 2, . . . , N due to the fact that pi > 0 ([13]). Hence, E = c1 N with c being a nonzero
constant. Note that sgn(ei,1 ) = sgn(ei,1 h
) for i = 1, . . . , N , it is readily established
that E h = c1 N . 

Lemma 9.8 Suppose V : D → R is a continuous function satisfying:

(i) V is positive definite.
(ii) There exist real numbers c̃ > 0, α ∈ (0, 1) and 0 < ς < ∞, and an open neigh-
borhood U ⊆ D of the origin such that

V̇ (x, t) ≤ −c̃V (x, t)α + ς, x ∈ U/{0}. (9.6)

9.2 Useful Lemmas and Intermediate Results 175

Then there exists a finite time T ∗ (called the settling time) satisfying

V (x, 0)1−α
T∗ ≤ , 0 < μ1 < 1 (9.7)
(1 − μ1 )c̃(1 − α)

such that for ∀t ≥ T ∗ , we have V (x, t) is bounded by

1
V (x, t) < (ς/(μ1 c̃)) α . (9.8)

Proof Let Δ = {x : V (x, t) < ( μς1 c̃ ) α , 0 < μ1 < 1}. According to Theorem 5.2 in
1

1
[14] for any x ∈ / Δ for all t ∈ [0, tx ], it holds that V (x, t) ≥ (ς/μ1 c̃) α , i.e., ς ≤
α
μ1 c̃V (x, t) , for all t ∈ [0, tx ]. This fact, together with (9.6), implies that

V̇ (x, t) ≤ − c̃V (x, t)α + μ1 c̃V (x, t)α

= − (1 − μ1 )c̃V (x, t)α (9.9)

for all t ∈ [0, tx ]. Note that V (x, t) ≥ (ς/η1 c̃)α > 0 for t ∈ [0, tx ]. According to
V (x,0)1−α
Lemma 9.1 and (9.9), we know tx < (1−μ 1 )c̃(1−α)
= T ∗ . Thus (9.8) holds for ∀t ≥ T ∗ .


Lemma 9.9 Consider a continuous function of the form

ϑ̇(t) = −kϑ ϑ(t)h + qς (t), (9.10)

where kϑ > 0, 0 < h = h 1 / h 2 ≤ 1 (h 1 , h 2 > 0 are positive odd integers), q > 0 is

a constant, and ς (t) ≥ 0 is a scalar function. Then, for any given bounded initial
condition ϑ(0) ≥ 0, we have ϑ(t) ≥ 0, ∀t ≥ 0.

Proof The assertion is vacuously true if ς (t) ≡ 0. In fact, in this case (9.10) becomes

ϑ̇(t) = −kϑ ϑ(t)h . (9.11)

By direct integration, we obtain the solution to (9.11) is

  1
ϑ(t) = |ϑ(0)|1−h − kϑ (1 − h)t 1−h , t < T̄ , ϑ(0) = 0,
ϑ(t) = 0, otherwise (9.12)

where T̄ = |ϑ(0)|
1−h

kϑ (1−h)
, which implies that ϑ(t) ≥ 0 for ∀t ≥ 0 and any given bounded
initial condition ϑ(0) ≥ 0.
We now consider the case that ς (t) is not always equal to 0. If there exists some
time constant t1 > 0 such that ς (t1 ) = 0, then by continuity there exists a small
neighborhood [t1 , t1 + Δt1 ) such that ϑ(t) ≥ 0 on [t1 , t1 + Δt1 ) (Δt1 > 0) from
the above analysis; if there exists some time constant t2 > 0 such that ς (t2 ) > 0
176 9 Finite-Time Leaderless Consensus Control for Systems …

  h1
2)
and ϑ(t2 ) > qς(t kϑ
> 0, then by continuity there exists a small neighborhood
[t2 , t2 + Δt2 ) such that ϑ(t) > 0 on [t2 , t2 + Δt2 ) (Δt2 > 0); if there exists some time
  h1
3)
constant t3 > 0 such that ς (t3 ) > 0 and ϑ(t3 ) = qς(t kϑ
> 0, then by (9.10) it holds
  h1 h
3)
that ϑ̇(t3 ) = −kϑ qς(t kϑ
+ qς (t3 ) = 0, and by continuity there exists a small
  h1
3)
neighborhood [t3 , t3 + Δt3 ) such that ϑ(t) = qς(t kϑ
> 0 on [t3 , t3 + Δt3 ) (Δt3 >
0); and if there exists some time constant t4 > 0 such that ς (t4 ) > 0 and 0 < ϑ(t4 ) <
 1  1 h
qς(t4 ) h qς(t4 ) h
kϑ
, it follows from (9.10) that ϑ̇(t 4 ) > −k ϑ kϑ
+ qς (t4 ) = 0, and
then there exists a small neighborhood [t4 , t4 + Δt4 ) such that ϑ(t) > ϑ(t4 ) > 0 on
[t4 , t4 + Δt4 ) (Δt4 > 0) by continuity. By global continuity of the function ϑ(t),
we have ϑ(t) ≥ 0 ∀t ≥ 0 for any given bounded initial condition ϑ(0) ≥ 0 in this
case. 

Lemma 9.10 Let ζ̃ = ζ − ζ̂ . It then holds that


3s−1 1 + s  8s(s−1) 2(s−1)
 4s 2s−2
ζ̃ ζ̂ 1+s ≤ 2 (1+s)2 − 2 1+s ζ̃ 1+s + 1 − 2 1+s
4s
−16s(s−1)2
3s − 1 (1 + s)2 (1+s)3 4s
+ + ζ 1+s , (9.13)
4s 4s

where s = (4ln − 1)/(4ln + 1), l ∈ Z + .

Proof Upon using Lemma 9.6, it is deduced from Lemma 9.4 that
  3s−1    1+s
1+s 1+s 4s
4s

ζ̃ ζ − ζ̃ ≤ ζ 1+s − ζ − ζ̃
4s

1+s 4s
3s−1
1+s
= ζ 1+s − ζ − ζ̃ ζ − ζ̃
4s
1+s 4s 2s−2 3s−1 3s−1

≤ ζ 1+s − 2 1+s ζ − ζ̃ ζ 1+s − ζ̃ 1+s . (9.14)
4s

Note that s = 4ln−1

, straightforward derivation yields that 3s−1 = 2ln−1 , and there-
4ln+1
3s−1 3s−1
   3s−1 3s−1
 1+s 2ln
fore, ζ − ζ̃ ζ 1+s − ζ̃ 1+s = ζ − ζ̃ ζ 1+s − ζ̃ 1+s . By using Lemma 9.2, it
then holds that
9.2 Useful Lemmas and Intermediate Results 177

  3s−1
1+s
ζ̃ ζ − ζ̃
1+s 4s 2s−2 4s 2s−2 3s−1 2s−2 3s−1 2s−2 4s

≤ ζ 1+s − 2 1+s ζ 1+s + 2 1+s ζ̃ ζ 1+s + 2 1+s ζ ζ̃ 1+s − 2 1+s ζ̃ 1+s
4s
 2s−2  4s
 (1 + s) 2 1+s ζ̃
1+s 4s
1+s 4s 2s−2 4s (3s − 1)ζ 1+s
≤ ζ 1+s −2 1+s ζ 1+s + +
4s 4s 4s
 s−1 2
 4s  2s−2
 4s
(1 + s) 2−4( 1+s ) ζ
1+s 1+s
(3s − 1) 2 1+s ζ̃ 2s−2 4s
+ + − 2 1+s ζ̃ 1+s
4s 4s
⎡ ⎛ ⎞ ⎤
−16s(s−1)2
1 + s ⎢ 8s(s−1) 2(s−1) 4s ⎜ 2s−2 3s − 1 (1 + s)2 (1+s)3 4s ⎥
⎟ 1+s
≤ ⎣ 2 (1+s)2 − 2 1+s ζ̃ 1+s + ⎝1 − 2 1+s + + ⎠ζ ⎦.
4s 4s 4s

(9.15)

9.3 Inductive Controller Design

The focus in this section is on deriving the finite-time output consensus control
scheme for MAS with dynamics (9.1). We first introduce the neighborhood error
and local virtual error and then conduct the inductive control design and stability
analysis.
To develop the distributed finite-time control scheme for the high-order MAS on
graph, we define the following local position error ei,1 and local virtual errors ei,m
(m = 2, . . . , n) as

ei,1 = ai j (yi − i − yj + j ), i, j ∈ {1, . . . , N }
j∈N i
1/q ∗1/q
ei,m = xi,m m − xi,m m , i ∈ {1, . . . , N }, m = 2, . . . , n (9.16)

where i are constants denoting the final consensus configuration such that yi −
y j = i − j (i, j ∈ {1, . . . , N }); qm = (4ln + 3 − 2m)/(4ln + 1) > 0 (l ∈ Z + );
∗ ∗ qm
xi,m (m = 2, . . . , n) is the virtual control given by xi,m = −βi,m−1 ei,m−1 , where
βi,m−1 is positive constant to be designed. Denote y = [y1 , . . . , y N ] ∈ R ,
T N
=
[ 1 , . . . , N ]T ∈ R N , and E = [e1,1 , . . . , e N ,1 ]T ∈ R N , then it holds that E =
L(y − ).
It is noted that unlike the existing method [15] in defining the fraction power qm ,
here we introduce the integer l into qm . Such treatment leads to the fraction power
feedback control capable of adjusting the control precision and finite convergence
time, as shown in sequel.
With the intermediate results and lemmas as represented in Sect. 9.2, we are ready
to address the finite-time adaptive consensus control design for the high-order non-
affine MAS (9.1).
178 9 Finite-Time Leaderless Consensus Control for Systems …

The design procedure consists of n steps. At each of the first m (m = 1, 2, . . . , n −

∗
1) step, a virtual control xi,m+1 is designed, and finally, the true control law u i
is derived at step n. In the following, we denote by xk = [x1,k , . . . , x N ,k ] for k =
1, . . . , n.
Step 1: Construct the first part of Lyapunov function candidate as

1 1+s 1+s
V1 (x1 ) = (E 2 )T P E 2 , (9.17)
(1 + s)ks

where s and ks are defined as in Lemma 9.7. With simple computation, it is obtained
that
1 s T 1
V̇1 (x1 ) = (E ) P Ė = (E s )T P L x2
ks ks
1 s T 1
= (E ) P L x2∗ + (E s )T P L(x2 − x2∗ ), (9.18)
ks ks

where x2 = [x1,2 , . . . , x N ,2 ]T ∈ R N . By introducing the virtual control x2∗ = −β1 E s

(β1 > 0 is design constant) into (9.18), we have

β1 s T 1
V̇1 (x1 ) = − (E ) P L E s + (E s )T P L(x2 − x2∗ )
ks ks
1  
N N
β1 ∗
= − (E s )T Ξ E s + (xi,2 − xi,2 )  ji (e j,1 )s (9.19)
ks ks i=1 j=1

where Ξ = 1/2(P L + L T P) defined as in Lemma 9.7, and  ji is the ( j, i)th ele-

ment of (P L). Let max = max j,i∈{1,...,N } | ji |, a straightforward derivation by using
Lemmas 9.2 and 9.4 yields

1   1  1−s 
N N N N
∗
(xi,2 − xi,2 )  ji (e j,1 )s ≤ 2 |ei,2 |s | ji ||e j,1 |s
ks i=1 j=1
k s i=1 j=1
  2  N 2
1   1  
N N N
≤2 max
1−s
|ei,2 | s
|e j,1 | ≤
s
|ei,1 | s
+ C̃2 |ei,2 | s
ks i=1 j=1
2N i=1 i=1

1 
N N
≤ (ei,1 )2s + C2 (ei,2 )2s (9.20)
2 i=1 i=1

2

N 
N
in which the fact xi ≤N xi2 has been used, and C2 = N C̃2 is a fixed
i=1 i=1
constant.
In light of Lemma 9.7 and (9.20), it is deduced from (9.19) that
9.3 Inductive Controller Design 179


N
1
N  N
V̇1 (x1 ) ≤ −β1 (ei,1 )2s + (ei,1 )2s + C2 (ei,2 )2s . (9.21)
i=1
2 i=1 i=1

Step 2: Construct the second part of Lyapunov function candidate as

V2 (x1 , x2 ) = V1 (x1 ) + W2 (x1 , x2 ), (9.22)

with
N 
 xi,2 1+s−q2
1 ∗ q1
W2 (x1 , x2 ) = τ q2 − xi,22 dτ. (9.23)
∗
i=1 xi,2

It is obvious that W2 (· · · ) is C 1 , then V2 (·) is C 1 , and moreover,

 V2 (·) is positive
N
definite and V2 (·) ≤ k̄b i=1 (ei,1
1+s
+ ei,2
1+s
), where k̄b = max (1+s)k
1
s
, 2 . Note that

∂ W2 1+s−q
= ei,2 2 = ei,2 ,
∂ xi,2
∗ q1 
∂ W2 ∂ xi,22 xi,2 1 ∗ q1
s−q2
= −(1 + s − q2 ) τ q2 − xi,22 dτ
∂ x j,1 ∂ x j,1 ∗
xi,2
∗ q1
∂ xi,22 !
=− ∗
xi,2 − xi,2 , j ∈ N¯i = Ni ∪ i. (9.24)
∂ x j,1

The derivative of W2 (t) is thus derived as

⎡ ⎤
∗1

N    ∂ x q2
⎢ ∗ i,2 ⎥
Ẇ2 (·) = ⎣ei,2 ẋi,2 − xi,2 − xi,2 ẋ j,1 ⎦
∂ x j,1
i=1 j∈N¯i
⎡ ⎤

N 
≤ ⎣ei,2 (xi,3 − x ∗ ) + ei,2 x ∗ + 21−q2 β 1/q2 · |ei,2 |q2 | ai j (xi,2 − x j,2 )|⎦ .
i,3 i,3 1
i=1 j∈Ni
(9.25)

According to Lemma 9.5, we have

q2
∗1/q2
|xi,2 | ≤ ei,2 + xi,2 ≤ |ei,2 |q2 + β1 |ei,1 |q2 . (9.26)

Denote by r1 = max∀i∈J { ai j } and r2 = max∀i, j∈J {ai j }. It then follows from
j∈N i
(9.26) that
180 9 Finite-Time Leaderless Consensus Control for Systems …

1/q2

N 
21−q2 β1 |ei,2 |q2 ai j (xi,2 − x j,2 )
i=1 j∈N i
⎛ ⎞

N 
|ei,2 |q2 ⎝r1 |xi,2 | + r2 |x j,2 |⎠
1/q
≤21−q2 β1 2
i=1 j∈N i
N 

1/q2
≤21−q2 β1 r1 |ei,2 |2q2 + r1 β1 |ei,2 |q2 |ei,1 |q2
i=1

+ r2 |ei,2 |q2 (|e j,2 |q2 + β1 |e j,1 |q2 ) . (9.27)
j∈N i

Note that

2(1+1/q2 ) 2 1
r1 |ei,2 |q2 |ei,1 |q2 ≤ 22(1−q2 ) β1
1+1/q2
21−q2 β1 r1 |ei,2 |2q2 + |ei,1 |2q2 ,
4
 
|e j,2 |q2 ≤ 2−q2 β1 2 r2
1/q 1/q
21−q2 β1 2 r2 |ei,2 |q2 (|ei,2 |2q2 + |e j,2 |2q2 ),
j∈N i j∈N i
1+1/q2

2 1−q2
β1 r2 |ei,2 |q2 |e j,1 |q2

j∈N i
 2(1+1/q2 ) 2 1
≤ (22(1−q2 ) β1 r2 cN |ei,2 |2q2 + |e j,1 |2q2 ), (9.28)
4cN
j∈N i

in which cN = maxi∈{1,...,N } {dim{Ni }} with dim{Ni } being the number of the ele-
ments in Ni .
By substituting (9.28) into (9.27) we have

1/q2

N  1  2q2
N  2q N
21−q2 β1 |ei,2 |q2 · ai j (xi,2 − x j,2 ) ≤ ei,1 + C̄2 ei,22 (9.29)
i=1
2 i=1 i=1
j∈N i

2(1+1/q ) 2(1+1/q2 )
with C̄2 = 21−q2 β1 2 r1 + 22(1−q2 ) β1 r12 + 21−q2 β1 2 r2 cN +22(1−q2 ) β1
1/q 2 1/q
2 2
r2 cN .
By combining (9.22), (9.21), (9.25), and (9.29), we see that


N 
N
V̇2 ≤ − (β1 − 1) (ei,1 )2s + (C2 + C̄2 ) (ei,2 )2s
i=1 i=1


N
 
∗ ∗
+ ei,2 (xi,3 − xi,3 ) + ei,2 xi,3 . (9.30)
i=1
9.3 Inductive Controller Design 181

Note that q3 = q2 − 2/(4ln + 1), and then 1 + q3 = 2q2 = 2s. Thus by choosing
∗
β1 > n − 1 + kc with kc > 0 being a design parameter and the virtual control xi,3 =
q3
−βi,2 ei,2 with βi,2 ≥ n − 2 + kc + C2 + C̄2 , we get


N 
2 
N
∗
V̇2 ≤ −(n − 2 + kc ) (ei,m )2s + ei,2 (xi,3 − xi,3 ). (9.31)
i=1 m=1 i=1

Step k (3 ≤ k ≤ n − 1): To proceed, we conduct the proof by using an induc-

tive argument. Suppose at step k − 1, there exists a C 1 Lyapunov function Vk−1
(x1 , . . . , xk−1 ), which is positive definite and satisfies


N
Vk−1 (·) ≤ k̄b (ei,1
1+s
+ · · · + ei,k−1
1+s
) (9.32)
i=1

such that


N 
k−1 
N
1+s−qk−1 ∗
V̇k−1 (·) ≤ −(n − k + 1 + kc ) 2s
ei,m + ei,k−1 (xi,k − xi,k ). (9.33)
i=1 m=1 i=1

We claim that (9.32) and (9.33) also hold at step k. To prove this, consider

Vk (x1 , . . . , xk ) = Vk−1 (·) + Wk (x1 , . . . , xk ), (9.34)

with
N 
 xi,k 1+s−qk
1 ∗ q1
Wk (x1 , . . . , xk ) = τ qk − xi,k k dτ. (9.35)
∗
i=1 xi,k

It is not hard to check that Wk (·) is C 1 , and moreover,

∂ Wk 1+s−q
= ei,k k ,
∂ xi,k
∗ q1 
∂ Wk ∂ xi,k k xi,k 1 ∗ q1
s−qk
= −(1 + s − qk ) τ qk − xi,k k dτ, m = 1, . . . , k − 1,
∂ xi,m ∂ xi,m ∗
xi,k
(9.36)
N
Therefore, Vk (·) is C 1 , positive definite and satisfies Vk (·) ≤ k̄b i=1 (ei,1
1+s
+ ··· +
1+s
ei,k ).
By recalling (9.33), we derive the derivative of Vk (·) as
182 9 Finite-Time Leaderless Consensus Control for Systems …


N 
k−1 
N
!
1+s−qk−1 ∗
V̇k (·) ≤ − (n − k + 1 + kc ) 2s
ei,m + ei,k−1 xi,k − xi,k
i=1 m=1 i=1
⎛ ⎞

N N 
k−1
∂ W  ∂ Wk
+
1+s−q
ei,k k xi,k+1 + ⎝ k
ẋi,m + ẋ j,1 ⎠ . (9.37)
∂ x i,m ¯
∂ x j,1
i=1 i=1 m=2 j∈N i

To continue, we examine the second term and the fourth term on the right-hand
side of (9.37). First, according to Lemma 9.4, it holds that
2s
1+s−qk−1 ∗
ei,k−1
ei,k−1 (xi,k − xi,k ) ≤ 21−qk |ei,k−1 |1+s−qk−1 |ei,k |qk ≤ + Ci,k ei,k
2s
(9.38)
2
with Ci,k being a constant. As for the fourth term on the right-hand side of (9.37),
we have
⎛ ⎞
N 
k−1
∂ Wk  ∂ Wk
⎝ ẋi,m + ẋ j,1 ⎠
∂ x i,m ¯
∂ x j,1
i=1 m=2 j∈N i
∗ 1
∗ 1
N 
k−1
∂ xi,k k
q
 ∂ xi,kqk
≤ (1 + s − qk )21−qk |ei,k |s · ẋi,m + ẋ j,1 . (9.39)
∂ xi,m ¯
∂ x j,1
i=1 m=2 j∈N i

Now we show that

∗ 1
∗ 1  k 

N 
k−1
∂ xi,k k
q
 ∂ xi,kqk 
N 
ẋi,m + ẋ j,1 ≤ ρki,m |ei,m | s
, (9.40)
∂ xi,m ∂ x j,1
i=1 m=2 j∈N ¯i i=1 m=1

for a constant ρki,m ≥ 0. This is done by inductive argument. First of all, it is deduced
from (9.24)–(9.26) that

∗ 1  2 

N 
∂ xi,22
q

N
1/q
 
N 
ẋ j,1 ≤ β1 2 ai j (xi,2 − x j,2 ) ≤ ρ2i,m |ei,m | s
,
i=1 j∈N ¯i
∂ x j,1 i=1 i=1 m=1
j∈N i
(9.41)

where ρ2i,m ≥ 0 is a constant. Assume that for k − 1, it holds,

∗ 1
∗ 1  k−1 

N 
k−2
∂ xi,k−1
qk−1
 ∂ xi,k−1
qk−1

N 
ẋi,m + ẋ j,1 ≤ ρ(k−1)i,m |ei,m | s
(9.42)
∂ xi,m ∂ x j,1
i=1 m=2 j∈N ¯i i=1 m=1
9.3 Inductive Controller Design 183

with ρ(k−1)i,m ≥ 0 (m = 1, . . . , k − 1) being constants. By recalling that ei,m =

1/q ∗1/q ∗ qm
xi,m m − xi,m m and xi,m = −βi,m−1 ei,m−1 , it thus holds that

∗1/qm 1/q
xi,m = −βi,m−1
m
ei,m−1 , (9.43)
qm
∗1/qm
|xi,m | ≤ ei,m + xi,m ≤ |ei,m |qm + βi,m−1 |ei,m−1 |qm , (9.44)

which, together with (9.42), yields that

∗ 1
∗ 1

N 
k−1
∂ xi,k k
q
 ∂ xi,kqk
ẋi,m + ẋ j,1
∂ xi,m ∂ x j,1
i=1 m=2 j∈N ¯i
⎛ ⎞

N
1/qk ⎝

k−1
∂e  ∂e
ẋ j,1 ⎠
i,k−1 i,k−1
= −βi,k−1 ẋi,m +
∂ x i,m ¯
∂ x j,1
i=1 m=2 j∈N i
1
−1 ∗ 1
∗ 1

N
1/qk
qk−1
xi,k−1 
k−2
∂ xi,k−1
qk−1
 ∂ xi,k−1
qk−1

≤ βi,k−1 ẋi,k−1 − ẋi,m − ẋ j,1

qk−1 ∂ xi,m ∂ x j,1
i=1 m=2 j∈N ¯i


N 
1/q 1 1−q
1
−1 1−q
≤ βi,k−1 · ei,k−1k−1 + βi,k−2
k−1 q
k
ei,k−2k−1
i=1
qk−1

k−1
· (|ei,k |qk + |ei,k−1 |qk βi,k−1 ) + ρ(k−1)i,m |ei,m |s
m=1


N 
k
≤ ρki,m |ei,m |s . (9.45)
i=1 m=1

Therefore,
⎛ ⎞
N 
k−1
∂ W  ∂ Wk
⎝ k
ẋi,m + ẋ j,1 ⎠
∂ x i,m ∂ x j,1
i=1 m=2 j∈N ¯i
 k 

N 
≤ (1 + s − qk )2 1−qk
|ei,k | s
ρki,m |ei,m | s

i=1 m=1

1 
N 
k−1 
N
≤ 2s
ei,m + 2s
C̄i,k ei,k (9.46)
2 i=1 m=1 i=1

with C̄i,k > 0 being a constant.

By substituting (9.38) and (9.46) into (9.37), we arrive at
184 9 Finite-Time Leaderless Consensus Control for Systems …


N 
k−1 
N
! 2s
V̇k (·) ≤ − (n − k + kc ) 2s
ei,m + Ci,k + C̄i,k ei,k
i=1 m=1 i=1


N
! N
1+s−qk ∗ 1+s−q ∗
+ ei,k xi,k+1 − xi,k+1 + ei,k k xi,k+1 . (9.47)
i=1 i=1

∗ q
By introducing the virtual control xi,k+1 = −βi,k ei,kk+1 with βi,k ≥ n − k + kc +
Ci,k + C̄i,k > 0, we get from (9.47) that


N 
k 
N
!
1+s−qk ∗
V̇k (·) ≤ − (n − k + kc ) 2s
ei,m + ei,k xi,k+1 − xi,k+1 . (9.48)
i=1 m=1 i=1

Step n: Consider the nth part of Lyapunov function candidate

Vn (x1 , . . . , xn ) = Vn−1 (·) + Wn (x1 , . . . , xn ) (9.49)

with
N 
 xi,n 1+s−qn
1 ∗ 1

Wn (x1 , . . . , xn ) = τ qn − xi,nqn dτ. (9.50)

∗
i=1 xi,n

N
Clearly, Vn (·) so defined is C 1 , positive definite and satisfies Vn (·) ≤ k̄b i=1 (ei,1
1+s
+
· · · + ei,n
1+s
).
From the above inductive argument, one can conclude that


N 
n−1 
N
1+s−qn

N
! 2s
V̇n (·) ≤ −kc 2s
ei,m + ei,n ẋi,n + Ci,n + C̄i,n ei,n . (9.51)
i=1 m=1 i=1 i=1

where Ci,n and C̄i,n are finite positive constants.

The actual control u i is designed as

q ζ̂i 2 1+s−qn
u i = −βi,n ei,nn+1 − ψ e (9.52)
ηi2 i i,n

with the updated law

σi
ζ̂˙i = −σi γi ζ̂i 1+s + 2 ψi2 ei,n
3s−1
2(1+s−qn )
, (9.53)
ηi

where ζ̂i is the estimation of ζi (here ζi is a virtual parameter to be defined later),

ψi (·) is a scalar and readily computable function, βi,n , σi , γi , and ηi are positive
design parameters chosen arbitrarily by the designer. Now we are ready to present
the main control result in the next section.
9.4 The Finite-Time Stability Analysis for the Uncertain High-Order MAS 185

9.4 The Finite-Time Stability Analysis for the Uncertain

High-Order MAS

In this section, we give the main control result and conduct the finite-time stability
analysis for the uncertain non-affine multi-agent systems (9.1).

Theorem 9.1 Consider the networked high-order non-affine uncertain MAS (9.1)
under Assumption 9.1. If the distributed adaptive control scheme (9.52)–(9.53) with
∗
the virtual control xi,m (m = 1, . . . , n) is applied, then for any initial conditions sat-
isfying ζ̂i (0) ≥ 0 and V (0) ≤ μ (μ > 0 is a finite constant), the finite-time leaderless
consensus for system (9.1) is achieved in that
(1) the neighborhood error E converges to a small residual set Θ0 , namely
⎧   2s1 ⎫
⎨ [(1 + s)ks kb ]
2s
1+s ς ⎬
Θ0 = E ≤ , (9.54)
⎩ μ1 μ2 ka ⎭

and the parameter estimate error ζ̃ converges to a small region Θ1 , given by

⎧ ⎡ ⎤ 1+s ⎫
⎪  1+s
2s 4s ⎪
⎨ 2σmax kb ς ⎦ ⎬
Θ1 = ζ̃  ≤ ⎣ (9.55)
⎪
⎩ g μ1 μ2 ka ⎪
⎭

in a finite time T ∗ that satisfies

2s
V (0)1− 1+s
T∗ ≤ , (9.56)
(1 − μ1 )c̃(1 − 2s
1+s
)

where
)  2(s−1) * ) *
1+s 8s(s−1)  g
ka = min kc , gγi 2 1+s − 2 (1+s)2
, kb = max k̄b , ,
4s 2σi
⎛ −16s(s−1)2
⎞
N
ηi
2 N
1+s 3s − 1 (1 + s)2 (1+s)3 4s
gγi ⎝1 − 2 1+s + ⎠ ζi 1+s
2s−2
ς= + +
i=1
4 i=1
4s 4s 4s

0 < μ1 < 1, 0 < μ2 ≤ 1, ζ̃ = [ζ̃1 , . . . , ζ̃ N ]T , σmax = max{σ1 , . . . , σ N }.

(9.57)
186 9 Finite-Time Leaderless Consensus Control for Systems …

(2) all of the signals in the controlled system are semi-globally uniformly ultimately
bounded.

Proof Choose the Lyapunov function candidate as


N
g
V = Vn + ζ̃i2 . (9.58)
i=1
2σi

Taking the derivative of V along (9.4) yields that


N 
n−1 
N
1+s−qn !
V̇ (·) ≤ − kc 2s
ei,m + ei,n gi u i + L f i
i=1 m=1 i=1


N
! 2s  N
g
+ Ci,n + C̄i,n ei,n + ζ̃i − ζ̂˙i . (9.59)
i=1 i=1
σi

According to Assumption 9.1 and Young’s inequality, we have, for any ηi > 0,
that

1+s−qn αi2 2 2(1+s−qn ) ηi2

ei,n L f i ≤ αi ψi |ei,n |1+s−qn ≤ ψ e + . (9.60)
ηi2 i i,n 4

Let ζi = g −1 αi2 . By using (9.52) and (9.60), it is deduced from (9.59) that


N 
n−1 
N
! 2s
V̇ (·) ≤ − kc 2s
ei,m − gi βi,n − Ci,n − C̄i,n ei,n
i=1 m=1 i=1
 

N
gi ζ̂i 2(1+s−qn ) gζi 2(1+s−qn )
+ − 2 ψi2 ei,n + 2 ψi2 ei,n
i=1
ηi ηi

N
η2 
N
g ˙
+ i
+ ζ̃i − ζ̂i . (9.61)
i=1
4 i=1
σi

According to Lemma 9.9 and (9.53), we see that ζ̂i (t) ≥ 0 for any given initial
estimate ζ̂i (0) ≥ 0. If βi,n is chosen such that βi,n g − Ci,n − C̄i,n > kc , it thus follows
from (9.61) that


N 
n 
N
g 2(1+s−qn )
V̇ (·) ≤ −kc 2s
ei,m + ψi2 ei,n (ζi − ζ̂i )
i=1 m=1 i=1
ηi2

N
η2 
N
g ˙
+ i
+ ζ̃i (− ζ̂i ). (9.62)
i=1
4 i=1
σi
9.4 The Finite-Time Stability Analysis for the Uncertain High-Order MAS 187

By applying the adaptive law for ζ̂i given in (9.53), one has


N 
n 
N 3s−1 
N
η2
V̇ (·) ≤ −kc 2s
ei,m + gγi ζ̃i ζ̂i 1+s
+ i
. (9.63)
i=1 m=1 i=1 i=1
4

According to Lemma 9.10, one gets

3s−1 1 + s  2(s−1) 8s(s−1)  4s
ζ̃i ζ̂i 1+s ≤ − 2 1+s − 2 (1+s)2 ζ̃i 1+s
4s ⎛ ⎞
−16s(s−1)2
1+s ⎝ 2s−2 3s − 1 (1 + s)2 (1+s)3 ⎠ 1+s 4s
+ 1 − 2 1+s + + ζi , (9.64)
4s 4s 4s

2(s−1) 8s(s−1)
where 2 1+s − 2 (1+s)2 > 0. Let
⎛ −16s(s−1)2
⎞

N
η2 
N
1+s 3s − 1 (1 + s)2 (1+s)3 4s
gγi ⎝1 − 2 ⎠ ζi 1+s .
2s−2
ς= i
+ 1+s + +
i=1
4 i=1
4s 4s 4s
(9.65)

By inserting (9.64) into (9.63), it then holds that


N 
n 
N
1+s  2(s−1) 8s(s−1)  4s
V̇ (·) ≤ −kc 2s
ei,m − gγi 2 1+s − 2 (1+s)2 ζ̃i 1+s + ς. (9.66)
i=1 m=1 i=1
4s

By denoting that
)  2(s−1) *
1+s 8s(s−1) 
ka = min kc , gγi 2 1+s − 2 (1+s)2 , (9.67)
i∈{1,...,N } 4s

(9.66) can be expressed as

+ ,

N 
n 
N 2 2s
V̇ (·) ≤ −ka 2s
ei,m + ζ̃i 1+s
+ ς. (9.68)
i=1 m=1 i=1

In the sequel, we prove that there exists a finite time T ∗ > 0 and a bounded constant
∗
cV such that V (t) < cV when t g≥T .
Let kb = maxi∈{1,...,N } k̄b , 2σi , and thus,

+ N n ,
 
N 2 1+s
V (·) ≤ kb 1+s
ei,m + ζ̃i 1+s
(9.69)
i=1 m=1 i=1
188 9 Finite-Time Leaderless Consensus Control for Systems …

which further implies, upon using Lemma 9.5, that

 N n 
2s 2s  
N 4s
V 1+s (·) ≤ kb 1+s 2s
ei,m + ζ̃i 1+s
. (9.70)
i=1 m=1 i=1

μ2 k a
Let c̃ = 2s (0 < μ2 ≤ 1). It is straightforward from (9.68) and (9.70) that
kb1+s

2s
V̇ (·) ≤ −c̃V 1+s (·) + ς. (9.71)

Upon using Lemma 9.8, it is seen that there exists a finite time T ∗ satisfying
2s
∗ V (0)1− 1+s
T ≤ , (9.72)
(1 − μ1 )c̃(1 − 2s
1+s
)

such that for ∀t ≥ T ∗ ,

1+s 1+s
ς 2s ς 2s
V (·) < = kb = cV . (9.73)
μ1 c̃ μ1 μ2 ka

Next, we examine the steady-state errors of all agents. Note that

 2s
 2s1
1 1 [(1 + s)ks kb ] 1+s ς
E ≤[(1 + s)ks V1 (·)] 1+s ≤ [(1 + s)ks V (·)] 1+s ≤ .
μ1 μ2 ka
(9.74)

Let ζ̃ = [ζ̃1 , . . . , ζ̃ N ]T ∈ R N and σmax = max{σ1 , . . . , σ N }. Then we have from

(9.58) that

- ⎡  1+s
2s ⎤ 1+s
4s
2σmax ⎣ 2σ max k b ς ⎦
ζ̃  ≤ V (·) ≤ . (9.75)
g g μ1 μ2 ka

Thus, it has been established that under the proposed finite-time control scheme,
the neighborhood error converges to a small region Θ0 given as
⎧   2s1 ⎫
⎨ [(1 + s)ks kb ]
2s
1+s ς ⎬
Θ0 = E ≤ , (9.76)
⎩ μ1 μ2 ka ⎭

and the parameter estimate errors converge to the region Θ1 ,

9.4 The Finite-Time Stability Analysis for the Uncertain High-Order MAS 189
⎧ ⎡ ⎤ 1+s ⎫
⎪  1+s
2s 4s ⎪
⎨ 2σmax kb ς ⎦ ⎬
Θ1 = ζ̃  ≤ ⎣ (9.77)
⎪
⎩ g μ1 μ2 ka ⎪
⎭

in the finite time T ∗ as given in (9.72).

9.5 Comparison with Regular State Feedback-Based

Infinite-Time Method

Note that if we set the fraction power s = 1, q1 = q2 = · · · = qn = 1, the proposed

finite-time control scheme reduces to the regular infinite-time control and the con-
vergence result is in infinite-time sense. In this case, the local virtual error ei,m
∗
(i ∈ {1, . . . , N }, m = 2, . . . , n) becomes ei,m = xi,m − xi,m , with the virtual control
∗
xi,m given as

∗
xi,m = −βi,m−1 ei,m−1 , (9.78)

and the actual control u i becomes

ζ̂i 2
u i = −βi,n ei,n − ψ ei,n (9.79)
ηi2 i

with the updated law

σi
ζ̂˙i = −σi γi ζ̂i + 2 ψi2 ei,n (9.80)
ηi

then the cooperative uniformly ultimately bounded (CUUB) consensus is achieved.

More specifically, under this control scheme, the neighborhood error converges to a
set given as
. 1 /
2kb ς 2
Θ2 = E ≤ , (9.81)
μ1 μ2 ka

and the parameter estimate errors converge to a set

⎧   21 ⎫
⎨ 2σmax kb ς ⎬
Θ3 = ζ̃  ≤ (9.82)
⎩ gμ1 μ2 ka ⎭

with c̃ and ς corresponding to the values with s = 1 and q1 = · · · = qn = 1.

190 9 Finite-Time Leaderless Consensus Control for Systems …

Remark 9.1 It is seen that in these two control schemes, i.e., (9.52)–(9.53) and
(9.79)–(9.80), the steady-state errors are controlled by the convergence regions,
i.e., Θ0 , Θ1 , Θ2 , and Θ3 , respectively, all of which can be reduced as small as
desired by choosing the control parameters. Nevertheless, due to control satura-
tion constraint, the control parameters are not allowed to be set too large. While in
the proposed method, one can choose the fractional power s = (4ln − 1)/(4ln + 1)
properly to enhance the disturbance rejection performance and increase the control
precision without the need for excessively large control gain. For instance, suppose
2s
[(1+s)ks kb ] 1+s ς
μ1 μ2 k a
< 1 is satisfied (this is allowed because all the parameters in ka and
kb , i.e., kc , γi , and σi are free design parameters), note that 1/s is much larger than
2s
1
2s   21
[(1+s)ks kb ] 1+s ς
1, which makes μ1 μ2 k a
 μ2k1 μb k2skςa , thus leading to better control pre-
cision as compared with regular state feedback-based infinite-time control method.

Remark 9.2 It is worth noting that the proposed finite-time control method is essen-
tially different from the regular state feedback-based UUB control method, here the
feedback error and the adaptive compensation are not used directly in building the
control scheme and the adaptive updated law, instead the fractional power of the
3s−1
q
feedback error of the form ei,mm
and the fractional power of adaptive term ζ̂i 1+s are
embedded into the algorithm, it is such special treatment that makes it possible to
achieve sufficient control precision in finite time by the adaptive approach. In other
words, for the proposed control method, the convergence precision is adjustable and
the convergence time is easily adjustable without the need for high feedback gain
to reach the given precision, whereas for regular state-based UUB control, although
it can reach a bound in a finite time, both the bound and the finite time cannot be
prescribed and one can only enlarge the control gain to get a better control precision.

Remark 9.3 Note that if letting

 the number of the agents N be equal to 1 and the
neighborhood error ei,1 =  j∈N i ai j (xi,1 − x j,1 ) be replaced by xi,1 , and accord-
ingly r1 = max∀i∈{1,...,N } { ai j } = 1 and r2 = max∀i, j∈{1,...,N } {ai j } = 0, then the
j∈N i
proposed control scheme is immediately applicable to single (non-networked) sys-
tems with high-order nonlinear dynamics. Compared with the existing works on
single nonlinear systems by [15–17], the control scheme presented herein is able to
deal with much less restrictive uncertainties including non-parametric uncertainties
and external disturbances gracefully. In addition, in defining the fraction power qm ,
we introduce the integer l into qm here, such treatment makes the fraction power feed-
back control capable of adjusting the control precision and finite convergence time,
leading to a more favorable solution for finite-time control of high-order systems.

Remark 9.4 Compared with the finite-time control for the system with second-order
dynamics, the work proposed in this chapter extends the result to high-order case,
which poses significant technical challenge. In fact, the main hindrance stems from
the fact that in the presence of high-order dynamics, the commonly used filtering
technique [18] cannot be used with Babarlat lemma or UUB theory to derive finite
9.5 Comparison with Regular State Feedback-Based Infinite-Time Method 191

time convergence because the convergence of the filtered error to zero or to a small
value cannot ensure the finite-time convergence of the original error. Compared
with the other two works [19, 20] on adaptive finite-time consensus for nonlinear
MAS, where (i) the systems are second-order; (ii) the communication is undirected;
and (iii) the uncertainties are assumed to exhibit the linear parametric property, the
work proposes in this chapter gives an adaptive distributed control solution to the
problem of the cooperative consensus of nonlinear MAS with general high-order
dynamics involving non-parametric uncertainties and non-vanishing disturbances
under directed communication, by focusing on achieving consensus in finite time
with sufficient accuracy.

9.6 Numerical Simulations

In order to demonstrate the feasibility of the proposed finite-time control, two numer-
ical examples are given.

Example 9.1 We simulate on a group of third-order MAS with non-affine dynamics

in the form of (9.1) with

ẋi,3 = (0.6 + 0.1exp(−xi,2

2
− xi,3
2
))u i + 0.1 sin(u i ) + f di (·), (9.83)

for i = 1, 2, 3, 4, where the system uncertainties f di are taken as f di (·) = di +

2
0.5(−1)i xi,1 + 0.5(−1)i (xi,2
2
) + 0.5xi,3
2
, and di denotes disturbances, taken as ran-
dom and bounded by |di | ≤ 0.1. The communication topology is directed and con-
nected as shown in Fig. 9.1. Each edge weight is taken as 0.1. The left eigenvector
of L associated with eigenvalue 0 is [1, 1, 1, 1]T .

The simulation runs for 20 s. The initial conditions are x1 (0) = [−0.2, −0.3, 0,
0.2]T , x2 (0) = [0.2, 0.3, 0, −0.2]T , and x3 (0) = 04 . The simulation is conducted
by applying the control laws given in (9.52)–(9.53), where the control parameters
are taken as: l = 1, β1 = 4, βi,2 = 12, βi,3 = 40, σi = 1, γi = 1, and ηi = 0.8. The
scalar function ψi (x̄i ) = 0.2 + xi,1
2
+ xi,2
2
+ xi,3
2
. In addition, the initial values of the
estimates are chosen as ζ̂i (0) = 0.5 for i = 1, 2, 3, 4. The error convergence result of
the four agents is depicted in Fig. 9.2. Figure 9.3 represents the parameter estimation
ζ̂i (i = 1, 2, 3, 4).
To show that better performance is achieved with the proposed control scheme,
we also tested the convergence property between the proposed finite-time adaptive
control scheme (9.52)–(9.53) and the typical infinite-time-based adaptive scheme
(9.79)–(9.80). Both the two control laws are applied to system (9.83) by using the
same design parameters. It is observed from Fig. 9.4 that the convergence rate is
faster and the error precision is better with the finite-time controller compared with
the infinite-time controller.
192 9 Finite-Time Leaderless Consensus Control for Systems …

Fig. 9.1 Directed communications among the four agents

0.05
e11 e21 e31 e41
e i,1

−0.05
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Fig. 9.2 Position neighborhood errors under the finite-time controller

5
χ1 χ2 χ3 χ4
4

3
χi

0
0 5 10 15 20
Time (s)

Fig. 9.3 Parameter estimation under the finite time controller

9.6 Numerical Simulations 193

0.03
e21 under finite time control
e41 under finite time control
0.02
e21 under infinite time control
e41 under infinite time control
0.01
e i,1

−0.01

−0.02
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Fig. 9.4 Position neighborhood errors under the two different controllers

Example 9.2 We consider a one-link manipulator [21] with non-affine actuation

system,

Di q̈i + Bi q̇ + Ni sin(qi ) = f (u i ), (9.84)

for i = 1, 2, 3, 4, where qi , q̇i , and q̈i denote the link angular position, velocity,
and acceleration of the ith robot, respectively; f (u i ) = u i + 0.5cos(u i ) is the driv-
ing torque in which u i is the motor current, G g (t) = −0.1 sin(t) is the external
disturbance. For convenience, we consider the same communication condition as
in Example 1 with each edge weight being 0.1. We set the final consensus con-
figuration i as = 0.15[π/2, 5π/8, 3π/4, 7π/8]T . The initial conditions are
x1 (0) = [−0.1, −0.2, 0, 0.1]T , x2 (0) = 04 , and ζ̂ (0) = 0.514 .

Figure 9.5 is the position trajectories of the four robots, and Fig. 9.6 is the neighbor-
hood errors. For comparison, we also compare the convergence property of the pro-
posed control with the typical infinite-time-based control (9.79)–(9.80). Figure 9.7
demonstrates that the convergence property including the convergence rate and the
error precision is better with the finite-time controller as compared with the infinite-
time method, as theoretically predicted.

9.7 Notes

Finite-time convergence behavior is of special importance in cooperative control

of MAS, but the vast majority research on finite-time control of MAS has been
focused on linear single or double integrator ([22–25], to just name a few). Although
effort on finite-time consensus of MAS with first- or second-order feedback lin-
earizable nonlinear dynamics has been made by [20, 26–28] by using homogeneous
(local) approximation, results on finite-time distributed control of MAS with general
194 9 Finite-Time Leaderless Consensus Control for Systems …

0.5
Robot 1 Robot 2 Robot 3 Robot 4

xi,1

−0.5
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Fig. 9.5 Position trajectories of the four agents under the finite-time controller

0.02
e11 e21 e31 e41

0.01
e i,1

−0.01

−0.02
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Fig. 9.6 Position neighborhood errors under the finite-time controller

0.02
e11 under finite time contol
e21 under finite time control
e11 under infinite time control
0.01
e21 under infinite time control
e i,1

−0.01

−0.02
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Fig. 9.7 Position neighborhood errors under the two different controllers
9.7 Notes 195

high-order nonlinear dynamics are scarce, and extending the finite-time control meth-
ods for single or double integrator to higher-order case encounters significant tech-
nical challenge. The main hindrance stems from the fact that in the presence of high-
order dynamics, the commonly used filtering technique [18] cannot be used with
Babarlat lemma or uniformly ultimately bounded (UUB) theory to derive finite-time
convergence because the convergence of the filtered error to zero or to a small value
cannot ensure the finite-time convergence of the original error.
The underlying problem becomes even more challenging when non-parametric
uncertainties and unknown time-varying control effectiveness gains (or non-affine
dynamics) are involved. In fact, the existence of non-parametric uncertainties liter-
ally makes it impossible to use continuous control to achieve zero steady-state error
in finite time even for non-networked (single) system. Although zero error conver-
gence is no longer feasible for such situation, convergence with sufficient steady-state
accuracy is acceptable for most practical applications. Motivated by the work of [14],
where it is shown that the finite-time convergence systems with nonlinearities possess
many good features including faster convergence rate, better disturbance rejection,
and robustness against uncertainties, it is thus practically meaningful and plausible
to pursue achieving finite-time convergence with sufficient accuracy. However, new
issues and challenges arise along with this effort. One is that if the control effec-
tiveness gain is unknown and time-varying, literally any designed protocol will be
polluted or disguised, making the control impact on the agent uncertain and unclear
such that the widely used negative feedback control method is inapplicable for direct
cancelation. Furthermore, to guarantee global finite-time stable control with suffi-
cient accuracy for high-order nonlinear MAS, the derivative of the Lyapunov function
V (x, t) must satisfy the relation of V̇ (x, t) ≤ −cV (x, t)α + ς for suitable constants
c > 0, 0 < α < 1 and bounded constant ς > 0, which brings about another tech-
nical difficulty to adaptive control design because it is non-trivial to synthesize the
adaptive compensation to fulfill such relation. The problem under consideration is
made further complicated by the local communication topology, especially the local
one-way directed communication topology, which imposes significant challenge to
extend existing finite-time adaptive control methods to distributed consensus control
of networked MAS. As such, the problem of distributed finite-time consensus con-
trol of MAS with high-order nonlinear dynamics has not been well addressed; this is
particularly true when there involve non-parametric uncertainties and disturbances
in the model and the communication is local and especially one-way directed.
In this chapter, we present a solution to such problem by focusing on achieving
consensus of MAS in finite time with sufficient accuracy. To circumvent the afore-
mentioned technical obstacles, we first introduce the locally defined neighborhood
error and the fractionally composed virtual error, which are embedded into the con-
trol scheme; and then, we design the fraction power feedback law and fraction power
adaptive updated law for the parameter estimation, which renders the crucial relation
V̇ (x, t) ≤ −cV (x, t)α + ς to be satisfied. To carry out rigorous proof of the stability
196 9 Finite-Time Leaderless Consensus Control for Systems …

by inductive analysis, we add the fraction power integration term to the Lyapunov
function candidate. By integrating the adjustable fractional power feedback con-
trol method with adaptive control method, not only the inherent uncertainties with
unknown and time-varying control gains can be effectively compensated but also the
cooperative consensus can be achieved in finite time with sufficient accuracy.

References

1. Olfati-Saber, R., Murray, R.: Consensus problems in networks of agents with switching topol-
ogy and time-delays. IEEE Trans. Autom. Control 49(9), 1520–1533 (2004)
2. Oldham, K.B.: The fractional calculus. Elsevier (1974)
3. Gao, R.Z., Wang, Y.J., Lai, J.F., Gao, H.: Neuro-adaptive fault-tolerant control of high speed
trains under traction-braking failures using self-structuring neural networks. Inf. Sci. 367–368,
449–462 (2016)
4. Meng, W., Yang, Q., Jagannathan, S., Sun, Y.: Adaptive neural control of high-order uncertain
nonaffine systems: a transformation to affine systems approach. Automatica 23(2), 1473–1480
(2014)
5. Song, Y.D., Wang, Y.J., Wen, C.Y.: Adaptive fault-tolerant PI tracking control with guaranteed
transient and steady-state performance. IEEE Trans. Autom. Control 62(1), 481–487 (2017)
6. Wang, Y.J., Song, Y.D., Lewis, F.L.: Robust adaptive fault-tolerant control of multi-agent
systems with uncertain non-identical dynamics and undetectable actuation failures. IEEE Trans.
Ind. Electron. 62(6), 3978–3988 (2015)
7. Wang, Y.J., Song, Y.D., Krstic, M., Wen, C.Y.: Fault-tolerant finite time consensus for multiple
uncertain nonlinear mechanical systems under single-way directed communication interactions
and actuation failures. Automatica 63, 374–383 (2016)
8. Wang, Y.J., Song, Y.D.: Fraction dynamic-surface-based neuroadaptive finite-time containment
control of multiagent systems in nonaffine pure-feedback form. IEEE Trans. Neural Netw.
Learn. Syst. 28(3), 678–689 (2016)
9. Song, Y.D., Huang, X.C., Wen, C.Y.: Tracking control for a class of unknown nonsquare MIMO
nonaffine systems: A deep-rooted information based robust adaptive approach. IEEE Trans.
Autom. Control 61(10), 3227–3233 (2016)
10. Bhat, S.P., Bernstein, D.S.: Continuous finite-time stabilization of the translational and rota-
tional double integrators. IEEE Trans. Autom. Control 43(5), 678–682 (1998)
11. Qian, C., Lin, W.: Non-Lipschitz continuous stabilizer for nonlinear systems with uncontrol-
lable unstable linearization. Syst. Control Lett. 42(3), 185–200 (2001)
12. Hardy, G., Littlewood, J., Polya, G.: Inequalities. Cambridge University Press, Cambridge
(1952)
13. Qu, Z.: Cooperative Control of Dynamical Systems: Applications to Autonomous Vehicles.
Springer, London (2009)
14. Bhat, S.P., Bernstein, D.S.: Finite-time stability of continuous autonomous systems. SIAM J.
Control Optim. 38(3), 751–766 (2000)
15. Huang, X., Lin, W., Yang, B.: Global finite-time stabilization of a class of uncertain nonlinear
systems. Automatica 41, 881–888 (2005)
16. Hong, Y.: Finite-time stabilization and stabilizability of a class of controllable systems. Syst.
Control Lett. 46, 231–236 (2002)
17. Huang, J., Wen, C., Wang, W., Song, Y.D.: Design of adaptive finite-time controllers for nonlin-
ear uncertain systems based on given transient specifications. Automatica 69, 395–404 (2016)
18. Slotine, J.E., Sastry, S.S.: Tracking control of nonlinear systems using sliding surfaces with
application to robot manipulators. Int. J. Control 38(2), 132–135 (1983)
References 197

19. Yu, H., Shen, Y., Xia, X.: Adaptive finite-time consensus in multi-agent networks. Syst. Control
Lett. 62(10), 880–889 (2013)
20. Huang, J., Wen, C.Y., Wang, W., Song, Y.D.: Adaptive finite-time consensus control of a group
of uncertain nonlinear mechanical systems. Automatica 51, 292–301 (2015)
21. Chen, M., Tao, G., Jiang, B.: Dynamic surface control using neural networks for a class of
uncertain nonlinear systems with input saturation. IEEE Trans. Neural Netw. Learn. Syst.
26(9), 2086–2097 (2015)
22. Xiao, F., Wang, L., Chen, J., Gao, Y.: Finite-time formation control for multi-agent systems.
Automatica 45(11), 2605–2611 (2009)
23. Wang, L., Xiao, F.: Finite-time consensus problems for networks of dynamic agents. IEEE
Trans. Autom. Control 55(4), 950–955 (2010)
24. Li, S., Du, H., Lin, X.: Finite-time consensus algorithm for multi-agent systems with double-
integrator dynamics. Automatica 47(8), 1706–1712 (2011)
25. Zhang, Y., Yang, Y.: Finite-time consensus of second-order leader-following multi-agent sys-
tems without velocity measurements. Phys. Lett. A 337(3–4), 243–249 (2013)
26. Cao, Y., Ren, W.: Finite-time consensus for multi-agent networks with unknown inherent
nonlinear dynamics. Automatica 50(10), 2648–2656 (2014)
27. Li, C.Y., Qu, Z.H.: Distributed finite-time consensus of nonlinear systems under switching
topologies. Automatica 50(6), 1626–1631 (2014)
28. Du, H., He, Y., Cheng, Y.: Finite-time synchronization of a class of second-order nonlinear
multi-agent systems using output feedback control. IEEE Trans. Circuits Syst. I 61(6), 1778–
1788 (2014)