IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 20, NO.

3, MARCH 2024 4149

Secure Leader–Follower Formation Control of

Networked Mobile Robots Under Replay Attacks
Zhao-Qing Liu , Graduate Student Member, IEEE, Xiaohua Ge , Senior Member, IEEE, Hao Xie ,
Qing-Long Han , Fellow, IEEE, Jinchuan Zheng , Senior Member, IEEE,
and Yu-Long Wang , Member, IEEE

Abstract—This article is concerned with the leader– I. INTRODUCTION

follower formation control problem of multiple networked
ORMATION control of a multiple mobile robot (multi-
mobile robots (MRs), where the information exchanges over
communication networks among the MRs suffer from re-
play attacks. The central aim is to develop an effective se-
F MR) system has found its extensive applications in both
military and civilian areas, such as target tracking and enclos-
cure control scheme for the multiple-MR system such that ing, search and rescue, environmental monitoring, and border
the desired leader–follower formation control objectives
are accomplished regardless of the simultaneous presence patrolling [1], [2], [3]. Specifically, the primary objective of
of replay attacks, network-induced delays, system uncer- formation control of a multi-MR system is to steer a group of
tainties, and external disturbances. Toward this aim, an mobile robots (MRs) into forming a desired geometric pattern
extended state observer is first constructed to estimate in terms of their states (e.g, position and heading), while ac-
the unknown nonlinear terms consisting of system uncer- complishing the coordinated task (e.g., monitoring, tracking,
tainties and external disturbances. Then, leveraging the
time-stamp technique, a dedicated data packet analyzer is enclosing, capturing, and lifting). To achieve this objective,
developed for each MR to detect and handle replay attacks. various formation control strategies have been reported in the
Furthermore, a secure leader–follower formation control literature, including virtual-structure control [4], behavior-based
scheme, consisting of a network-based cooperative kine- control [5], and leader–follower control [6], [7], [8], [9]. In par-
matic control law and two local kinetic control laws, is ticular, leader–follower control strategies have attracted much
designed. Finally, both simulation and experiment results
are provided to validate the efficacy of the proposed secure attention due to their simplicity and scalability [6].
leader–follower formation control scheme. Information exchanges among neighboring robots play an
essential role in achieving the desired coordinated formation
Index Terms—Leader–follower formation (LFF) control,
control objective. In order to realize inter-robot communications
mobile robots (MRs), network-induced delays, replay at-
tacks, secure control. of key information (e.g., position, heading, and/or velocities
of neighboring robots), suitable wireless communication net-
Manuscript received 30 May 2023; revised 5 August 2023; accepted works are usually employed [10], [11], [12], [13], whereas
26 August 2023. Date of publication 12 October 2023; date of cur- the use of wireless communication networks inevitably induces
rent version 23 February 2024. This work was supported in part
by the National Natural Science Foundation of China under Grant certain network-induced constraints such as network-induced
52371372 and Grant 61833011 and in part by the Project of Science delays [14], which may deteriorate significantly the system
and Technology Commission of Shanghai Municipality, China, under performance or even cause instability of the multi-MR control
Grant 20ZR1420200, Grant 21SQBS01600, Grant 22JC1401400, and
Grant 21190780300. Paper no. TII-23-1956. (Corresponding authors: system. It should be mentioned that the communication networks
Qing-Long Han; Yu-Long Wang.) are assumed to be ideal, and the effects of those network-induced
Zhao-Qing Liu is with the School of Science, Computing and Engi- constraints are not incorporated in most of the existing relevant
neering Technologies, Swinburne University of Technology, Melbourne,
VIC 3122, Australia, and also with the School of Mechatronic Engi- studies [10], [11], [12], [13]. How to develop a suitable data
neering and Automation, Shanghai University, Shanghai 200444, China transmission model accounting for those network-induced con-
(e-mail: [email protected]). straints and to further address their adverse effects on formation
Xiaohua Ge, Qing-Long Han, and Jinchuan Zheng are with the
School of Science, Computing and Engineering Technologies, Swin- control performance of a networked multi-MR system deserves
burne University of Technology, Melbourne, VIC 3122, Australia (e-mail: further investigation.
[email protected]; [email protected]; [email protected]). During information exchanges among networked MRs, an-
Hao Xie is with the Shanghai Aerospace Control Technology Insti-
tute, Shanghai 201109, China, and also with the School of Automation other nonnegligible issue is the security of communication chan-
Science and Electrical Engineering, Beihang University, Beijing 100191, nels due to the openness or insufficient protection of the wireless
China (e-mail: [email protected]). network. In this case, the multi-MR system may be vulnerable to
Yu-Long Wang is with the School of Mechatronic Engineering and Au-
tomation, Shanghai University, Shanghai 200444, China, and also with various security threats, such as denial-of-service attacks [15],
the Shanghai Key Laboratory of Power Station Automation Technology, [16], deception attacks [17], and replay attacks [18]. We refer the
Shanghai 200444, China (e-mail: [email protected]). interested readers to [19] for a comprehensive review of secure
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TII.2023.3313651. distributed cooperative control techniques against malicious
Digital Object Identifier 10.1109/TII.2023.3313651 attacks. In this study, we will focus on replay attacks, which

1551-3203 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 4150 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 20, NO. 3, MARCH 2024

represent a type of cyberattack that involves intercepting and

retransmitting data that were previously exchanged between two
parties. In the context of multi-MR formation control, an attacker
may launch a man-in-the-middle attack to intercept and modify
the data exchanged among interacting robots. Alternatively, the
attacker may capture the data by using wireless sniffing tools
or by hacking into the robot’s wireless network. Since robots
rely on the real-time and accurate data from their neighbors to
regulate their motions, replay attacks, if not well tackled, may
result in instability of the multi-MR control system.
To cope with replay attacks on multi-MR control systems, Fig. 1. Illustration of a multi-MR system composed of multiple MRs
a common countermeasure is to implement advanced authenti- and a virtual leader, where Xb -Yb is the body-fixed reference frame and
cation and encryption mechanisms to protect data transmitted X-Y is the earth-fixed reference frame.
among robots. However, these mechanisms do not incorporate
the dynamical behavior of each robot. As such, secure control
schemes are still needed to preserve the stability and perfor- 3) A secure LFF control scheme, featuring a network-based
mance of the multi-MR control system against replay attacks. cooperative kinematic control law and two local kinetic
Up until now, several schemes have been proposed to tackle control laws, is designed to achieve the desired con-
replay attacks in the literature. For example, the effects of replay trol objectives even under the effects of replay attacks,
attacks were analyzed in the context of process noises in [20], network-induced delays, system uncertainties, and exter-
where a watermark signal was injected into the control loop as a nal disturbances. Formal stability analysis of the resultant
private excitation so as to realize the detection of replay attacks. closed-loop multi-MR system is further given.
In [21], a measurement residual-based cumulative sum detector Notation: N denotes the set of positive integers. diag{·}
was proposed to detect replayed LiDAR measurement data. and col{·} represent a diagonal matrix and a column vector,
In [22], replay attacks were described as random network delays, respectively.  ·  denotes the Euclidean norm. The minimum
and a nominal delay-tolerable controller was then designed to and maximum eigenvalues of a symmetric matrix are denoted
mitigate the adverse effects induced by the attack delays. In [18], by λmin (·) and λmax (·), respectively. E{·} is the mathematical
a proportional-integral-observer-based secure control scheme expectation. The transpose of a matrix (vector) is denoted by the
was designed, where two positive scalars and a time-varying superscript T . The inverse of a nonsingular matrix is represented
parameter were used to express the temporal behavior of replay by the superscript −1. Matrices, if not explicitly stated, are
attacks. It is noteworthy that the secure schemes aforementioned dimension-compatible.
are all designed for automated vehicle systems involving only
kinematics [20], [21] or longitudinal dynamics [18], [22]. In II. PROBLEM FORMULATION
practice, the motion of an MR is usually complex in terms of
A. MR Dynamics
multiple degrees of freedom, and its kinematics and kinetics are
intrinsically coupled. Neglecting one or another may jeopardize Consider a multi-MR system with N follower MRs and a
the system-level stability or greatly weaken the desired control virtual leader. Each MR is driven by two actuated wheels, as
performance of the MR. To the best of the authors’ knowledge, shown in Fig. 1. For the ith MR, i = 1, 2, . . . , N , its coupled
the secure formation control problem of a networked multi-MR kinematics and kinetics can be described as [3]
system subject to replay attacks and coupled kinematics and
kinetics has not been adequately addressed, which motivates ẋi (t) = cos(ψi (t))vi (t) (1a)
this study. ẏi (t) = sin(ψi (t))vi (t) (1b)
In this article, we investigate the problem of secure leader–
follower formation (LFF) control for a multi-MR system that ψ̇i (t) = ωi (t) (1c)
suffers from the simultaneous effects of replay attacks, network-
v̇i (t) = σiv (t) + ri (2mvi )−1 uvi (t) (1d)
induced delays, system uncertainties, and external disturbances.
−1 ω
The main contributions of this article are summarized as follows. ω̇i (t) = σiω (t) + ri (2hi mω
i ) ui (t) (1e)
1) Compared with some of the existing networked LFF
system models [10], [11], [12], [13], where the network- where xi (t) and yi (t) are the positions; vi (t) and ωi (t) are
induced constraints are not considered, a network-based the surge and angular velocities, respectively; ψi (t) denotes
the heading angle; mvi and mω i denote the inertia; ui (t) and
v
LFF system model is constructed for the multi-MR sys-
tem to account for the coexistence of replay attacks and ui (t) are the control input torques; ri denotes the wheel ra-
ω

network-induced delays. dius; hi denotes the half distance between MRi ’s two actu-
2) Distinct from the replay attack models in [18], [22], and ated wheels; and σiv (t) = (fiv (vi (t), ωi (t)) + r2i dvi (t))/mvi and
[23], where attack detection is not considered, a dedicated σiω (t) = (fiω (vi (t), ωi (t)) + 2h d (t))/mω
ri ω
i i i , with fi (·) and
v

data packet analyzer is proposed for each MR to detect fi (·) being the uncertainties and di (t) and di (t) being the un-
ω v ω

and handle replay attacks simultaneously by verifying the known external disturbances. Denote by pi (t) = [xi (t), yi (t)]T
fresh time stamps attached on the received packets. the position vector of the ith MR.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: SECURE LEADER–FOLLOWER FORMATION CONTROL OF NETWORKED MOBILE ROBOTS UNDER REPLAY ATTACKS 4151

sampled state information can be directly transmitted to MRi .

When j = 0, namely MRj is the virtual leader, MRj ’s state
information (e.g., the reference position and the desired speed)
is updated at discretized instant gϕ. For MRj ’s state information
(no matter when j = 0 or j = 0), the determination of how it
should be transmitted to MRi is made by the Transmitter j. For
this purpose, the sampled MRj ’s state information will further
be encapsulated into a packet and attached with a fresh time
stamp t̂ (i.e., t̂ = gϕ). Then, it will be sent to and eventually
arrive at MRi by virtue of the onboard deployed Receiver i.
Fig. 2. Schematic of secure LFF control for MRi under replay attacks. When the Receiver i receive the packet transmitted from
MRj , the embedded Data Packet Analyzer i will immediately
detect whether the received packet suffers from replay attacks
The position of the virtual leader, denoted as MR0 , is de- by verifying the attached fresh time stamp t̂.
fined as p0 (θ(t)) = [x0 (θ(t)), y0 (θ(t))]T , where x0 (θ(t)) and If t̂ > tk with tk being the time stamp attached on the last ac-
y0 (θ(t)) are the positions, θ(t) stands for a path variable, and cepted packet, it implies that the received packet is not attacked.
θ̇(t) = v0 (t) − χ(t), with v0 (t) being a desired update speed Then, the following steps will be executed by MRi .
and χ(t) denoting a variable to be designed. S1) Accept the received packet.
S2) Save the time stamp t̂ → tk and set γi (t) = 0, where
B. Information Exchange Topology γi (t) is a binary variable to indicate whether the re-
As shown in Fig. 2, the concerned multi-MR system is ceived packet is attacked (namely, “1”“attacked” and
controlled under a network environment, where each MR “0”“not attacked”).
can share its state information with its associated neighbors S3) Send the accepted state information to Buffer i to update
via a wireless network. The information exchange topology the stored information.
among the N follower MRs and MR0 can be described by a S4) Send the accepted state information to Controller i.
digraph G = {VN +1 , E, A0 , A}, where VN +1 = VN ∪ {0} = Otherwise, namely t̂ ≤ tk , this means that the received packet
{0, 1, 2, . . . , N } represents the node set with 0, namely MR0 ; is attacked. Then, the following steps will be executed by MRi .
E ∈ VN × VN describes the information links among follower S1’) Reject the received packet and set γi (t) = 1.
MRs with an arc (i, j) ∈ E indicating that MRi is able to receive S2’) Extract the last accepted state information from Buffer
information from MRj ; A0 = diag{ai0 }N stands for the MR0 ’s i and then send it to Controller i.
weighted adjacency matrix, in which ai0 = 1 being MRi is S3’) Calculate the time interval Δ between t̂ and tk . If the
able to receive information from MR0 and ai0 = 0 otherwise; calculated Δ is greater than the maximum duration of
and A = [aij ]N ×N stands for a weighted adjacency matrix replay attacks (denoted as Δ̄ and its initial value is zero),
with aij = 1 if (i, j) ∈ E and aij = 0 otherwise. Self-loops i.e., Δ = t̂ − tk − ϕ > Δ̄, then update Δ̄ ← Δ.
are excluded in this digraph, namely aii = 0. A path of G Remark 1: From the above steps S1)–S4) and S1’)–S3’), it
is a concatenation of arcs (i, i1 ), (i1 , i2 ), · · · , (ic , j), where can be observed that the data packet analyzer on MRi can detect
i1 , . . . , ic are distinct.Let L = [lij ]N ×N , where lij = −aij replay attacks in real time and further implement a countermea-
when i = j and lij = j∈Ni aij when i = j with Ni = {i, j ∈ sure (discarding or accepting the corresponding data packets) by
VN : (j, i) ∈ E} denoting MRi ’s neighbor set. We further rep- verifying the fresh time stamp t̂ attached on the received packets.
resent by H = L + A0 . Without loss of generality, it is assumed Compared with the replay attack models in [18] and [22], the
that MR0 is globally reachable in the communication digraph G proposed data packet analyzer empowers real-time replay attack
of the multi-MR system. detection. Although detection is also enabled in [20] and [21],
the watermark-based detection method in [20] requires some
specialized hardware or software to embed or extract the water-
C. Data Packet Analyzer Against Replay Attacks mark signal, thereby imposing additional cost and computation
Note that the sampled MRs’ state information may be re- requirement. The residual-based detection method in [21], on the
played with some historical data, namely subject to replay other hand, is sensitive to the timeliness and correctness of the
attacks, when it is shared among the wireless networks. It is, received signal. This means that if the received signal is delayed
thus, our purpose to design a data packet analyzer that can detect or corrupted (e.g., by noises or other attacks), false detection
and cope with replay attacks. may be triggered.
To clarify the reasoning behind the design of the data packet
analyzer, let us consider the data transmission from MRj , j ∈ Ni
or j = 0 (if ai0 = 1), to MRi , i ∈ VN , via wireless networks.
When j = 0, namely MRj is an actual MR and is a neighbor D. Problem
of MRi , MRj ’s state information (e.g., position, heading, and The problem to be addressed is stated as follows: For a
velocities) is sampled by onboard sensors at the time instant gϕ multi-MR system modeled by (1), the central aim is to develop
with the instant index g ∈ N and a constant sampling period a secure LFF control scheme under the above developed data
ϕ > 0, leading to pj (gϕ), ψj (gϕ), vj (gϕ), and ωj (gϕ). Such packet analyzer such that the following two control objectives
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 4152 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 20, NO. 3, MARCH 2024

are satisfied in the presence of replay attacks, network-induced 2) Design of a Networked Cooperative Kinematic Control
delays, system uncertainties, and external disturbances. Law: At first, we define the cooperative LFF error as
1) The multi-MRs are driven to cooperatively follow the
N
virtual leader with a desired time-varying formation pat-
tern qi (t), i.e., limt→∞ pi (t) − p0 (θ(t)) − qi (t) < 1 , zi (t) = aij (pi (t) − pj (t) − qij (t))
where pi (t) = [xi (t), yi (t)]T and 1 > 0 is a scalar. j=0

2) The derivative of θ(t) converges to a desired speed v0 (t) + ai0 (pi (t) − p0 (θ(t)) − qi (t))
as limt→∞ θ̇(t) − v0 (t) < 2 , where 2 > 0 is a scalar.
where qij (t) = qi (t) − qj (t).
III. MAIN RESULTS Taking the time derivative of zi (t) yields

A. Design of the Desired Secure LFF Control Scheme żi (t) = aςi αi (t) + aςi αei (t) − ai0 pθ0 (t)(v0 (t) − χ(t))
1) Design of an Extended State Observer (ESO): To esti- N N
mate the nonlinear terms σiv (t) and σiω (t), the following ESO is − aij vj (t)Ωj (ψj (t)) − aij q̇ij (t) (4)
constructed: j=1 j=0
⎧
⎪
⎪ v̂˙ i (t) = −ki1 ṽi (t) + σ̂iv (t) + 2m
ri
v ui (t)
v
 ∂p0 (θ(t))
j=0 aij , p0 (t) = ∂θ(t) , Ωj (ψj (t)) = [cos(ψj (t)),
θ
⎪ v
⎨ i where aςi = N
σ̂˙ i (t) = −ki2 ṽi (t) sin(ψj (t))] , αi (t) = vi (t)Ωi (ψi∗ (t)), and αei (t) = vi (t)
T ∗
˙ (2)
⎪
⎪ω̂i (t) = −ki3 ω̃i (t) + σ̂i (t) + 2hmωi ui (t)
ω ri ω
Ωi (ψi (t)) − vi∗ (t)Ωi (ψi∗ (t)), with vi∗ (t) and ψi∗ (t) being the
⎪
⎩˙ω
σ̂ (t) = −ki4 ω̃i (t)
i commanded surge velocity and heading angle, respectively, and
Ωi (ψi∗ (t)) = [cos(ψi∗ (t)), sin(ψi∗ (t))]T .
where ṽi (t) = v̂i (t) − vi (t), ω̃i (t) = ω̂i (t) − ωi (t), and kin for For t ∈ [tk + τk , tk+1 + τk+1 )  Dk , where τk and τk+1 are
n = 1, 2, 3, 4 are positive scalars. the duration of network-induced delays at instants tk and tk+1 ,
Then, the following observer error system can be derived: respectively, the update law of χ(t) is constructed as
Ėi1 (t) = Ai Ei1 (t) − Bi1 σ̇iv (t) − Bi2 σ̇iω (t) (3) N
χ̇(t) = − μ1 χ(t) + μ2 ai0 (pθ0 (t))T ((1 − γi (t))
where Ei1 (t)=[ṽi (t),σ̃iv (t),ω̃i (t),σ̃iω (t)]T with σ̃iv (t) = σ̂iv (t) −
i=1
σiv (t) and σ̃iω (t) = σ̂iω (t) − σiω (t), Bi1 =[0,1,0,0] 
T
,  Bi2 =
−ki1 1
[0,0,0,1]T , and Ai =diag {Ai,11 ,Ai,22 } with Ai,11 = −ki2 0 and ×zi (tk ) + γi (t)zi (tk−1 )) , t ∈ Dk (5)
 
−ki3 1
Ai,22 = −ki4 0 . Given that Ai is a Hurwitz matrix, there is a
positive-definite matrix Pi1 such that ATi Pi1 + Pi1T Ai = −ρi I, where μ1 and μ2 are positive scalars. Meanwhile, the virtual
where ρi > 0 is a scalar. Then, the following lemma is given to control law αi (t) is developed as
analyze the stability of the observer error system (3).
Lemma 1: Given suitable positive scalars ρi and kin for i ∈ αi (t) = a−1
ςi {−Ki1 ((1 − γi (t))zi (tk ) + γi (t)zi (tk−1 ))
VN and n = 1, 2, 3, 4, the observer error system (3) is input-to- N N
state stable (ISS). + aij q̇ij (t) + aij ((1 − γi (t))vj (tk )
Proof: Choose a Lyapunov candidate function as j=0 j=1

1 ×Ωj (ψj (tk )) + γi (t)vj (tk−1 )Ωj (ψj (tk−1 )))

Vio (t) = Ei1 (t)T Pi1 Ei1 (t).
2
+ ai0 (1 − γi (t))pθ0 (tk )v0 (tk )
The time derivation of Vio (t) along the error system (3) can be 
derived as +γi (t)pθ0 (tk−1 )v0 (tk−1 ) , t ∈ Dk (6)

V̇io (t) = Ei1 (t)T Pi1 (Ai Ei1 (t) − Bi1 σ̇iv (t) − Bi2 σ̇iω (t)) where Ki1 is a positive controller gain matrix.
Let αi (t) = [αix (t), αiy (t)]T . Then, design ψi∗ (t) and vi∗ (t)
which can be further estimated as as ψi∗ (t) = 2Fi (t)π + atan2(αiy (t), αix (t)), with Fi (t) being an
ρi ρi i integer-valued signal and
V̇io (t) ≤ − (1 − i )Ei1 (t)2 − Ei1 (t)2 + Ei1 (t)
2 2
vi∗ (t) = αi (t) cos ((1 − γi (t))ψi (tk )
× (Pi1 Bi1 |σ̇iv (t)| + Pi1 Bi2 |σ̇iω (t)|)
+γi (t)ψi (tk−1 ) − ψi∗ (t)) , t ∈ Dk . (7)
where 0 < i < 1.
It follows that V̇io (t) ≤ − ρ2i (1 − i )Ei1 (t)2 for any 3) Design of Local Kinetic Control Laws: The design of local
Ei1 (t) ≥ ρi2i (Pi1 Bi1 |σ̇iv (t)| + Pi1 Bi2 |σ̇iω (t)|). kinetic control laws is further divided into two parts: Part i) surge
According to [24, Th. 4.6], it can be concluded that the velocity control law design and Part ii) heading angle control
observer error system (3) is ISS. The proof is completed.  law design.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: SECURE LEADER–FOLLOWER FORMATION CONTROL OF NETWORKED MOBILE ROBOTS UNDER REPLAY ATTACKS 4153

Part i) Surge velocity control law design: Denoting vie (t) = τ1 (t)))+γi (t)vj (t − τ2 (t))Ωj (ψj (t − τ2 (t))), and p̆θ0 (t)v̆0
v̂i (t) − vi∗ (t), it follows that: (t) = (1−γi (t))pθ0 (t−τ1 (t))v0 (t−τ1 (t))+γi (t)pθ0 (t−τ2 (t))
ri v v0 (t − τ2 (t)).
v̇ie (t) = −ki1 ṽi (t) + σ̂iv (t) + u (t) − v̇ ∗ (t). (8) Based on (4), (8)–(11), and (14)–(16), the following network-
2mvi i
based LFF error system can be derived:
Design the surge velocity control law as ⎧
2 mvi ⎪żi (t) = −K
⎪
⎪ N
i1 z̆i (t) + aςi αei (t) + ai0 δi (t)
uvi (t) = (−Ki2 vie (t) − σ̂iv (t) + v̇i∗ (t)) , t ∈ Dk (9) ⎪
⎪
ri ⎪
⎨ + j=1 aij j (t) + ai0 pθ0 (t)χ(t)
v̇ie (t) = −Ki2 vie (t) − ki1 ṽi (t) (17)
where Ki2 is a positive controller gain. ⎪
⎪
Part ii) heading angle control law design: Denoting ψie (t) = ⎪
⎪ ψ̇ie (t) = −Ki3 ψie (t) + ωie (t) − ω̃i (t)
⎪
⎪
ψi (t) − ψi∗ (t), one has that ψ̇ie (t) = ωi (t) − ψ̇ ∗ (t). Further ⎩
ω̇ie (t) = −Ki4 ωie (t) − ki3 ω̃i (t) + ω̃id (t) − ψie (t)
define ωie (t) = ω̂i (t) − ωi∗ (t), with ωi∗ (t) being designed as
ωi∗ (t) = −Ki3 ψie (t) + ψ̇i∗ (t). It leads to where t ∈ Dk , j (t) = v̆j (t)Ωj (ψ̆j (t)) − vj (t)Ωj (ψj (t)),
δi (t) = p̆θ0 (t)v̆0 (t) − pθ0 (t)v0 (t), and ω̃id (t) = ω̂id (t) − ω̇i∗ (t).
ψ̇ie (t) = −Ki3 ψie (t) + ωie (t) − ω̃i (t) (10) The following lemma is given to analyze the stability of the
where Ki3 is a positive controller gain. LFF error system (17).
Taking the time derivative of ωie (t), we obtain that Lemma 2: Given suitable scalars τ1∗ ≥ 0, Δ̄ > 0, ϕ > 0, γ̄i ∈
∗
[0, 1], βi > 0, μ1 > 0, μ2 > 0, and positive controller gains Kin ,
ω̇ie (t) = −ki3 ω̃i (t) + σ̂iω (t) + uω
i (t)/Υi − ω̇i (t) (11) n = 1, 2, 3, 4, the LFF error system (17) is stochastic ISS if there
where Υi = 2hi mω are positive-definite matrices Wi1 , Qi1 , Qi2 , Ui1 , and Ui2 such
i /ri . Then, we design the following heading
angle control law: that
∗ Θi < 0 ∀ i ∈ VN (18)
i (t) = Υi (−Ki4 ωie (t) − σ̂i (t) + ω̇i (t) − ψie (t))
uω ω
(12)
where t ∈ Dk and Ki4 is a positive controller gain. where
⎡ ⎤
To estimate ω̇i∗ (t), the following second-order tracking dif- Θ11
i 0 0 Θ14i Θ15
i Θ16
i Θ17
i Θ18
i
ferentiator is constructed: ⎢  Θ 22
Θi23
0 0 Θi Θ27
26
Θ28 ⎥
 ∗ ⎢ i i i ⎥
⎢   Θ33 Θ38 ⎥
ω̂˙ i (t) = ω̂id (t) ⎢ i 0 0 0 0 i ⎥
⎢    Θ44 0 0 0 0 ⎥
ω̂˙ id (t) = −ι2 (ω̂ ∗ (t) − ω ∗ (t)) − 2ιi ω̂id (t) Θi = ⎢
⎢ 
i ⎥
i i i
⎢    Θ 55
i 0 0 0 ⎥⎥
where ω̂i∗ (t) and ω̂id
∗
(t) are, respectively, the estimates of ωi∗ (t) ⎢      Θ66 0 0 ⎥
⎢ i ⎥
∗
and ω̇i (t); ιi > 0 denotes a tuning parameter. Based on [25], ⎣       Θ77 i 0 ⎦
there exist positive constants ω̄ ∗ and ω̄id ∗
such that |ω̂i∗ (t) −        Θ88
i
∗ ∗ ∗ ∗ ∗
ωi (t)| ≤ ω̄ and |ω̂id (t) − ω̇i (t)| ≤ ω̄id .
Then, for t ∈ Dk , (12) can be redesigned as with Θ11 i = Qi1 + Qi2 − 4Ui1 − 4Ui2 , Θi = −2Ui1 , Θi =
14 15

−2Ui2 , Θi = 6Ui1 , Θi = 6Ui2 , Θi = Wi1 , Θi = −2(1 −

16 17 18 22

i (t) = Υi (−Ki4 ωie (t) − σ̂i (t) + ω̂id (t) − ψie (t)) . (13)
uω ω
γ̄i )2 Ki1 , Θ23i = −2(γ̄i − γ̄i )Ki1 , Θi = 6Ui1 , Θi = 6Ui2 ,
2 26 27

Θi = −(1 − γ̄i )(1 + βi Ki1 ), Θi = −2γ̄i Ki1 , Θi = −γ̄i (1

28 33 2 38

B. Stability Analysis + βi Ki1 ), Θ44 i = −Qi1 − 4Ui1 , Θi = −Qi2 − 4Ui2 , Θi =

55 66
∗ 2 ∗ 2
Let τ1 (t) = t − tk and τ2 (t) = t − tk−1 for t ∈ Dk . Obvi- −12Ui1 , Θi = −12Ui2 , and Θi = (τ1 ) Ui1 + (τ2 ) Ui2 −
77 88

ously, 0 ≤ τk ≤ τ1 (t) ≤ τ1∗ and 0 < ϕ ≤ τ2 (t) < τ2∗ with τ1∗ = 2βi I.
max{tk } and τ2∗ = Δ̄ + ϕ + τ1∗ . Then, (5)–(7) are converted Proof: See the Appendix.
into The following theorem presents the stability analysis conclu-
sion of the closed-loop control system cascaded by subsystems
N
(3) and (17).
χ̇(t) = − μ1 χ(t) + μ2 ai0 (pθ0 (t))T z̆i (t) (14)
Theorem 1: Consider MR dynamics (1) together with the
i=1
ESO (2), the path regulation law (14), the networked cooperative
kinematic control law (15), the surge velocity control law (9),
αi (t) = a−1
ςi − Ki1 z̆i (t) + ai0 p̆θ0 (t)v̆0 (t) and the heading angle control law (13). The closed-loop control
⎫ system for LFF of multi-MRs is stochastic ISS.
N N ⎬ Proof: By combining [3, Th. 1] and the proofs of Lemmas 1
+ aij q̇ij (t) + aij v̆j (t)Ωj (ψ̆j (t)) (15) and 2 in this article, one can derive that the closed-loop control
⎭ system is stochastic ISS. The detailed proof is omitted here for
j=0 j=1
brevity. 
vi∗ (t) = αi (t) cos(ψ̆i (t) − ψi∗ (t)) (16)
for t ∈ Dk , where z̆i (t) = (1 − γi (t))zi (t − τ1 (t))+γi (t)zi C. LFF Control Scheme Implementation Algorithm
(t − τ2 (t)), ψ̆i (t) = (1 − γi (t))ψi (t − τ1 (t)) + γi (t)ψi (t − With the proposed data packet analyzer, the maximum
τ2 (t)), v̆j (t)Ωj (ψ̆j (t)) = (1 − γi (t))vj (t − τ1 (t))Ωj (ψj (t − allowable duration of replay attacks can be recorded as Δ̄,
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 4154 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 20, NO. 3, MARCH 2024

which will be further adopted to compute the upper bound

Algorithm 1: LFF Control Scheme Implementation.
τ2∗ = Δ̄ + ϕ + τ1∗ of the artificial delay τ2 (t), where ϕ > 0
is the sampling period and τ1∗ ≥ 0 is the upper bound of the /% Offline Design Part %/
network-induced delay τk . It is clearly seen from Lemma 2 Observer gains design: Given scalars ρi > 0 and observer
that the scalar Δ̄ plays an important role in determining the gains kin > 0, n = 1,2,3,4, solve the following Riccatti
desired controller gains Kin , n = 1, 2, 3, 4. As such, the de- equation ATi Pi1 + Pi1T Ai = −ρi I to determine a feasible
sired controller gains can be designed in an offline manner. positive definite matrix Pi1 . If feasible, continue;
Specifically, one may use Δ̄ as a tuning parameter for the linear otherwise, reset ρi and kin , and repeat the above process;
matrix inequality conditions in (18) to find feasible controller Controller gains design: Given suitable scalars τ1∗ ≥ 0,
gains Kin . In this sense, the system designers may regard the Δ̄ > 0, ϕ > 0, γ̄i ∈ [0, 1], βi > 0, μ1 > 0, μ2 > 0, and
scalar Δ̄ as a metric for evaluating the different replay attack controller gains Kin > 0, n = 1, 2, 3, 4, solve (18) to
scenarios and designing the relevant attack-resilient controller determine a feasible solution of positive-definite matrices
gains. It should be mentioned that one may also turn to solve Wi1 , Qi1 , Qi2 , Ui1 , Ui2 . If feasible, continue; otherwise,
the following optimization problem: max{Δ̄} subject to (18), reset τ1∗ , Δ̄, ϕ, γ̄i , βi μ1 , μ2 and Kin , and repeat the above
to assess the worst case replay attack scenario that the resilient process;
LFF control scheme can accommodate. Under those scenarios, /% Online Implementation Part %/
the designed resilient control schemes will be implemented Initialization: Set the mission runtime of Tr seconds and
during the entire multi-MR LFF mission without any redesign the network channel monitoring window size of Tm
or reconfiguration of its gains. However, it is understandable seconds;
that the LFF control performance of the multi-MR system be- for g = 0 : 1 : Tϕr  do
comes relatively conservative since the network transmission 1. Run Data Packet Analyzer i to update Δ̄;
condition is assumed to be always the worst during controller 2. Implement the designed ESO (2) and control laws (9),
implementation. To achieve a tradeoff between satisfactory LFF (13), and (15) to each MRi ;
control performance and real-time attack severity, the proposed 3. If mod(g, Tϕm ) == 0, employ Lemma 2 under the
secure LFF control scheme can also be implemented in an online latest Δ̄ to find alternative Kin , n = 1,2,3,4; Else
manner. This can be done by employing the proposed data packet continue;
analyzer over some sliding monitoring windows. To draft the end
idea, an implementation algorithm for each MRi , i ∈ VN , is
briefly outlined in Algorithm 1.

IV. ILLUSTRATIVE EXAMPLES

In this section, both numerical simulation and experimental
case studies are conducted to validate the control performance of
the designed LFF control scheme even in the presence of replay
attacks.
Fig. 3. Communication digraphes of multiple MRs. (a) Digraph
adopted in the simulation case. (b) Digraph adopted in the experiment
A. Numerical Example Case case.

Consider the multi-MR system containing five MRs, i.e.,

N = 5. The dynamics parameters of MRi are given as mvi = TABLE I
2 DESIRED TIME-VARYING FORMATION PATTERNS
4.5 kg, mω i = 0.56 kg · m , hi = 0.186 m, and ri = 0.042 m
for i = 1, 2, . . . , 5. The position and heading of MR0 are gen-
erated by p0 (θ(t)) = [x0 (θ(t)), y0 (θ(t))]T and ψ0 (θ(t), respec-
tively, with x0 (θ(t)) = 0.12θ(t), y0 (θ(t)) = 40 sin(0.01θ(t)),
and ψ0 (θ(t)) = atan2(ẏ0 (θ(t)), ẋ0 (θ(t))). The adapted com-
munication digraph is described in Fig. 3(a). Moreover,
ϕ = 0.01 s and max{τk } = 0.02 s for k ∈ N. The rest
parameters are given as γ̄4 = 0.38, ιi = 100, μ1 = μ2 =
10, Ki1 = diag{0.8, 0.8}, Ki2 = Ki3 = Ki4 = ki1 = 20, ki2 = from MR2 to MR4 . The simulated replay attacks on MR4 occur
100, ki3 = 300, and ki4 = 1000 for i = 1, 2, . . . , 5. in the following time periods: [30 s, 40 s), [41 s, 45 s), [60 s,
In this numerical case, the formation of the multi-MR system 80 s), [95 s, 105 s), [120 s, 125 s), and [135 s, 143 s).
is switched at 45 and 115 s, respectively. The desired time- Implementing the developed secure LFF control scheme, the
varying formation pattern of each formation is described in corresponding numerical simulation results under the replay
Table I. To simulate replay attacks, we assume that MR2 ’s attacks are shown in Figs. 4–6. To be specific, Fig. 4 shows
position-heading and velocity information is randomly recorded the resulting time-varying LFF shaped by the five MRs and
by an adversary. Then, the adversary resend the recorded in- the virtual leader MR0 under replay attacks. Fig. 5 shows the
formation to MR4 through the communication network channel position snapshots of the multi-MR system at t = 0 s, t = 20 s,

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: SECURE LEADER–FOLLOWER FORMATION CONTROL OF NETWORKED MOBILE ROBOTS UNDER REPLAY ATTACKS 4155

Fig. 4. LFF performance under replay attacks.

Fig. 6. LFF errors under replay attacks in the numerical example case.
(a) The position errors in the X-axis. (b) The position errors in the Y-axis.
(c) The heading errors.

Fig. 7. Experimental platform.

reason lies in that their arrival times at the overlapping region

are different. Notice that MR5 also deviates from its desired
path when MR4 suffers from replay attacks. This is because
Fig. 5. Position snapshots of the five MRs and the virtual leader.
(a) t = 0 s. (b) t = 20 s. (c) t = 60 s. (d) t = 140 s.
MR5 obtains information from MR4 directly [see Fig. 3(a)].
Therefore, the motion of MR5 is straightforwardly influenced
by MR4 , whereas, by virtue of the designed secure LFF control
scheme, MR4 and MR5 can successfully recover and get back
t = 60 s, and t = 140 s, respectively. Fig. 6 shows the following to their desired paths in a relatively short period of time after
errors between MRi (i = 1, 2, . . . , 5) and MR0 . As shown in the end of each attack. As is clearly shown in Figs. 4–6, the
Fig. 5, each desired formation is achieved during the numerical designed secure LFF control scheme offers certain security and
simulation. In addition, from Figs. 4 and 6, one can see that resilience guarantees for the concerned multi-MR system under
MRi ’s motion path and following errors experience obvious replay attacks.
vibrations at the instants of formation switching, whereas the
durations and amplitudes of these vibrations are constrained
B. Experiment Case
to acceptable levels by using the designed secure LFF control
scheme. We next examine the LFF control performance under replay
To cope with replay attacks on MR4 , the time stamps attached attacks through an experimental case study.
on the transmitted and received packets are compared by virtue As shown in Fig. 7, the experiment platform for the LFF multi-
of the designed data packet analyzer. If the time stamp attached MR system containing two actual MRs and a supervisory com-
on the received packet is not fresh, it means that the newly puter. Each MR is equipped with an NI myRIO-1900 controller,
received packed is outdated. In this situation, the received packed two Maxon dc-motors integrated with encoders, and two Maxon
will be discarded actively. Although this method can eliminate motor drivers. In addition, a local area network is built by using a
the replayed data packets, it leads to a sustained increase in TP-LINK router such that the real-time data transmission among
τ2 (t). As a result, MR4 ’s path deviation and position-heading the two MRs and the supervisory computer can be achieved.
errors persistently increase, which can be seen from Figs. 4 and The model parameters of the two MRs are same as the ones in
6. Especially, when the duration of a replay attack reaches 20 s, the previous case. The communication digraph among the two
namely the attack during [60 s, 80 s), the motion path of MR4 MRs and a virtual leader (i.e., MR0 ) is described in Fig. 3(b).
is deviated considerably and even overlapped with the paths of The formation patterns are given as q1 (t) = [0.4; 0] and q2 (t) =
MR0 and MR2 . It should be mentioned that the collision between [−0.4; 0]. The position and heading of MR0 are generated by
MR4 and MR2 does not occur during this time period. The p0 (θ(t)) = [x0 (θ(t)), y0 (θ(t))]T and ψ0 (θ(t), respectively, with

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 4156 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 20, NO. 3, MARCH 2024

Fig. 8. Position snapshots of the two physical MRs. (a) At t = 5 s.

(b) At t = 15 s. (c) At t = 25 s.

Fig. 10. LFF errors under replay attacks in the experiment case.
(a) The position errors in the X-axis. (b) The position errors in the Y-axis.
(c) The heading errors.

Fig. 9. LFF performance under replay attacks.

x0 (θ(t)) = 0.6 cos(0.4θ(t)), y0 (θ(t)) = 0.6 sin(0.4θ(t)) + 1,

and ψ0 (θ(t)) = atan2(ẏ0 (θ(t)), ẋ0 (θ(t))). The other parameters Fig. 11. Control inputs of left and right motors in the experiment case.
are chosen as ιi = 5, μ1 = μ2 = 0.1, Ki1 = diag{0.8, 0.8}, (a) Left motor input. (b) Right motor input.
Ki2 = Ki3 = Ki4 = ki1 = 20, ki2 = 10, ki3 = 30, and ki4 =
10 for i = 1, 2. To simulate replay attacks, we consider that
V. CONCLUSION
an adversary can randomly record MR1 ’s position-heading and
velocity information, and such recorded information is then In this article, the LFF control was tackled for multi-MR
periodically resent and injected into MR2 through the communi- systems under replay attacks. First, an ESO was established to
cation network channel from MR1 to MR2 during t ∈ [15 s, 25 s]. estimate the unknown nonlinear dynamics of MRs. Then, a data
The experimental results are depicted in Figs. 8–11. It is packet analyzer was designed to detect and cope with replay
clear from Figs. 8–10 that the desired LFF objectives are well attacks. By simultaneously taking into account network-induced
achieved. Moreover, MR2 ’s motion path and position-heading delays and replay attacks, a networked LFF error system model
errors experience continuous vibrations under replay attacks was developed. Furthermore, to accomplish the formation track-
from 15 to 25 s. Accordingly, the control inputs of MR2 ’s ing task successfully even under replay attacks, a secure LFF
two dc motors also suffer from significant fluctuations during control scheme was designed. Finally, numerical simulation and
attacks, as shown in Fig. 11. However, owing to the resilience experimental case studies were conducted to substantiate the
and security of the proposed LFF control scheme, MR2 does not efficacy of the proposed control scheme.
significantly deviate from its desired path under the considered
attack scenario. APPENDIX A
The simulation and experiment results above demonstrate PROOF OF LEMMA 2
that the designed secure LFF control scheme is well suited for
Consider the following Lyapunov candidate:
networked multi-MR systems under replay attacks. In addition,
as compared with the existing networked control scheme, such Ve (t, zi (t), ψie (t), vie (t), ωie (t), χ(t))
as in [10], [11], [12], and [13], the designed secure LFF control
scheme can not only drive a group of MR to perform the desired
N

= Vzi (t, zi (t)) + Vψie (t, ψie (t)) + Vvie (t, vie (t))
formation task under the coexistence of unknown disturbances i=1
and system uncertainties but also provide certain resilience to 
against network-induced delays and replay attacks. +Vωie (t, ωie (t)) + Vχ (t, χ(t))

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: SECURE LEADER–FOLLOWER FORMATION CONTROL OF NETWORKED MOBILE ROBOTS UNDER REPLAY ATTACKS 4157

where Similarly, one has

 t
Vzi (t, zi (t)) = ziT (t)Wi1 zi (t) + ziT (s)Qi1 zi (s)ds V̇ψie (t, ψie (t)) ≤ −2Ki3 ψie
2
(t) + 2ψie (t)ωie (t)
t−τ1∗
 + 2|ψie (t)||ω̃i (t)|
t
+ ziT (s)Qi2 zi (s)ds V̇vie (t, vie (t)) ≤ −2Ki2 vie
2
(t) + 2ki1 |vie (t)||ṽi (t)|
t−τ2∗
 0  t V̇ωie (t, ωie (t)) ≤ −2Ki4 ωie
2
(t) + 2ki3 |ωie (t)||ω̃i (t)|
+ τ1∗ żiT (s)Ui1 żi (s)dsdϑ
−τ1∗ t+ϑ + 2|ωie (t)||ω̃id (t)| − 2ωie (t)ψie (t)
   
0 t   2χ(t)
N
+ τ2∗ żiT (s)Ui2 żi (s)dsdϑ E V̇χ (t, χ(t)) = − χ(t) + μ2 ai0 (pθ0 (t))T z̄i (t) .
−τ2∗ t+ϑ μ2 i=1
Vψie (t, ψie (t)) = T
ψie (t)ψie (t), Vvie (t, vie (t)) = vie
T
(t)vie (t)
Then, the following inequality can be derived:
Vωie (t, ωie (t)) = ωie
T
(t)ωie (t), Vχ (t, χ(t)) = χ2 (t)/(μ1 μ2 ).  
E V̇e (t, zi (t), ψie (t), vie (t), ωie (t), χ(t))
In order to facilitate the stability analysis, while guaranteeing
that the designed data packet analyzer can detect and handle N

various replay attack scenarios with different attack frequen- ≤ λmin (Θi1 )Ei2 (t)2 + 2(βi + 1)Ei2 (t)
cies and durations, the mathematical expectation is introduced i=1
to describe the probability of γi (t), i.e., Prob{γi (t) = 1} = ⎛ ⎞
N
E{γi (t)}  γ̄i , where 0 ≤ γ̄i ≤ 1. To carry on, the time deriva- × ⎝aςi αei (t) + ai0 δ̄i (t) + ¯ j (t)⎠
aij 
tive of Vzi (t, zi (t)) is obtained as j=1

V̇zi (t, zi (t)) 0 Ei2 (t)|χ(t)| − 2Ki3 ψie (t) + 2|ψie (t)||ω̃i (t)|
+ 2βi ai0 pθ∗ 2

= 2ziT (t)Wi1 żi (t) + ziT (t)(Qi1 + Qi2 )zi (t) − 2Ki2 vie
2
(t) + 2ki1 |vie (t)||ṽi (t)| + 2ki3 |ωie (t)||ω̃i (t)|

− ziT (t − τ1∗ )Qi1 zi (t − τ1∗ ) − ziT (t − τ2∗ )Qi2 zi (t − τ2∗ ) − 2Ki4 ωie
2
(t) + 2|ωie (t)||ω̃id (t)| − 2χ2 (t)/μ2
+ żiT (t)[(τ1∗ )2 Ui1 + (τ2∗ )2 Ui2 ]żi (t) + ζi1 + ζi2
where pθ∗0 is a bound satisfying p0 (t) ≤ p0 .
θ θ∗

∗ t

∗ t Let j = min{−λmin (Θi1 ), 2Ki2 , 2Ki3 , 2Ki4 , 2/μ2 },  = j −
where ζi1 = −τ1 t−τ1∗ żiT (ϑ)Ui1 żi (ϑ)dϑ and ζi2 = −τ2 t−τ2∗
0 > 0, E2 (t) = col{Ei2 (t)}N , ψe (t) = col{ψie (t)}N ,
2βi ai0 pθ∗
żiT (ϑ)Ui2 żi (ϑ)dϑ. ve (t) = col{vie (t)}N , ωe (t) = col{ωie (t)}N , and Γ(t) =
By using the Wirtinger-based integral inequality, we obtain col{E2 (t), ψe (t), ve (t), ωe (t), χ(t)}. Then, we have
ζi1 ≤ −ΠTi1 Ui1 Πi1 − 3ΠTi2 Ui1 Πi2 and ζi2 ≤ −ΠTi3 Ui2 Πi3 −
3ΠTi4 Ui2 Πi4 , where Πi1 = zi (t) − zi (t − τ1∗ ), Πi2 = zi (t) +  
t E V̇e (t, zi (t), ψie (t), vie (t), ωie (t), χ(t))
zi (t − τ1∗ ) − τ2∗ t−τ ∗ zi (ϑ)dϑ, Πi3 = zi (t) − zi (t − τ2∗ ), and
1 1
t
Πi4 = zi (t) + zi (t − τ2∗ ) − τ2∗ t−τ ∗ zi (ϑ)dϑ. ≤ −(1 − ε)Γ(t)2 − εΓ(t)2 + Γ(t)
2 2

Note that E{2(βi żiT (t)+ z̆iT (t))(−Ki1 z̆i (t)+ N j=1 aij j N

(t)+ai0 δi (t)+aςi αei (t)+ai0 p0 (t)χ(t)− żi (t))} = 0. Let Ei2

θ × {2i Ei1 (t) + 2|ω̃id (t)| + 2(βi + 1)
(t) = [ziT (t), ziT (t − τ1 (t)), ziT (t − τ2 (t)), ziT (t − τ1∗ ), ziT (t − i=1
t t 
τ2∗ ), τ1∗ t−τ ∗ ziT (ϑ)dϑ, τ1∗ t−τ ∗ ziT (ϑ)dϑ, żiT (t)]T . Then, the × a∗ςi 
¯ i (t) + ai0 δ̄i (t) + aςi αei (t)
1 1 2 2
following inequality can be derived:

  where i = maxi=1,2,...,N {ki1 , ki3 , 1}, a∗ςi = N j=1 aji , and
N
E V̇zi (t, zi (t)) ≤ Ei2 T
(t)Θi1 Ei2 (t) + 2 βi żiT (t) + z̄iT (t) 0 < ε < 1. As Γ(t) ≥ ε i=1 {i Ei1 (t) + |ω̃id (t)| +
2

⎧ ⎫ (βi + 1)(a∗ςi  ¯ i (t) + ai0 δ̄i (t) + aςi αei (t))}, we have E
⎨N ⎬
{V̇e (t, zi (t), ψie (t), vie (t), ωie (t), χ(t))} ≤ −(1 − ε)Γ(t)2 .
× aij  ¯ j (t) + ai0 δ̄i (t) + aςi αei (t) + ai0 pθ0 (t)χ(t)
⎩ ⎭ Based on [26, Th. 2], the following inequality can be
j=1
derived: Prob{Γ(t) < κ̄2 (Γ(t0 ), t)+Ξ(t)} ≥ 1−κ, where
where z̄i (t) = (1 − γ̄i )zi (t − τ1 (t)) + γ̄i zi (t − τ2 (t)),  ¯ j (t) κ > 0, κ̄2 (·) is a KL function, and Ξ(t) = 2(ε)−1
N
= v̄j (t)Ωj (ψ̄j (t)) − vj (t)Ωj (ψj (t)), and δ̄i (t) = p̄θ0 (t)v̄0 (t) − i=1 Pi2 {i Ei1 (t) + |ω̃id (t)| + (βi + 1)(a∗ςi i (t) +
$
pθ0 (t)v0 (t) with v̄j (t)Ωj (ψ̄j (t)) = (1 − γ̄i )vj (t − τ1 (t))Ωj aςi αei (t) + ai0 δi (t))}, with Pi2 = λmax (Pi2 )/λmin (Pi2 )
(ψj (t − τ1 (t))) + γ̄i vj (t − τ2 (t))Ωj (ψj (t − τ2 (t))) and p̄θ0 (t) and Pi2 = diag{Wi1 , Qi1 , Qi2 , τ1∗ Ui1 , τ2∗ Ui2 , I, (μ1 μ2 )−1 }.
v̄0 (t) = (1 − γ̄i )pθ0 (t − τ1 (t))v0 (t − τ1 (t)) + γ̄i pθ0 (t − τ2 (t)) According to [27, Def. 3], one can conclude that the LFF error
v0 (t − τ2 (t)). system (17) is stochastic ISS. The proof is completed. 

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 4158 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 20, NO. 3, MARCH 2024

REFERENCES [22] X. Li, X. Xu, L. Chen, H. Zhang, Z. Ju, and W. Guo, “Longitudinal speed
tracking control for an electric connected vehicle with actuator saturation
[1] R. S. Sharma, A. Mondal, and L. Behera, “Tracking control of mobile subject to a replay attack,” Nonlinear Dyn., vol. 111, pp. 1369–1383,
robots in formation in the presence of disturbances,” IEEE Trans. Ind. Jan. 2023.
Inform., vol. 17, no. 1, pp. 110–123, Jan. 2021. [23] W. Xu, J. Kurths, G. Wen, and X. Yu, “Resilient event-triggered con-
[2] B. Ning, Q.-L. Han, Z. Zuo, L. Ding, Q. Lu, and X. Ge, “Fixed-time trol strategies for second-order consensus,” IEEE Trans. Autom. Control,
and prescribed-time consensus control of multi-agent systems and its vol. 67, no. 8, pp. 4226–4233, Aug. 2022.
applications: A survey of recent trends and methodologies,” IEEE Trans. [24] H. K. Khalil, Nonlinear Control. London, U.K.:Pearson Educ., 2015.
Ind. Inform., vol. 19, no. 2, pp. 1121–1135, Feb. 2023. [25] B. Guo and Z. Zhao, “On convergence of tracking differentiator,” Int. J.
[3] N. Gu, Z. Peng, D. Wang, and F. Zhang, “Path-guided containment Control, vol. 84, no. 4, pp. 693–701, Apr. 2011.
maneuvering of mobile robots: Theory and experiments,” IEEE Trans. [26] P. Zhao, W. Feng, and Y. Kang, “Stochastic input-to-state stability of
Ind. Electron., vol. 68, no. 8, pp. 7178–7187, Aug. 2021. switched stochastic nonlinear systems,” Automatica, vol. 48, no. 10,
[4] Q. Zhang, L. Lapierre, and X. Xiang, “Distributed control of coordinated pp. 2569–2576, Oct. 2012.
path tracking for networked nonholonomic mobile vehicles,” IEEE Trans. [27] S.-J. Liu, J.-F. Zhang, and Z.-P. Jiang, “A notion of stochastic input-
Ind. Inform., vol. 9, no. 1, pp. 472–484, Feb. 2013. to-state stability and its application to stability of cascaded stochastic
[5] G. Antonelli, F. Arrichiello, and S. Chiaverini, “Experiments of formation nonlinear systems,” Acta Math. Appl. Sinica, vol. 24, no. 1, pp. 141–156,
control with multirobot systems using the null-space-based behavioral Jan. 2008.
control,” IEEE Trans. Control Syst. Technol., vol. 17, no. 5, pp. 1173–1182,
Sep. 2009.
[6] S.-L. Dai, S. He, X. Chen, and X. Jin, “Adaptive leader-follower for-
mation control of nonholonomic mobile robots with prescribed transient
and steady-state performance,” IEEE Trans. Ind. Inform., vol. 16, no. 6, Zhao-Qing Liu (Graduate Student Member,
pp. 3662–3671, Jun. 2020. IEEE) received the M.Sc. degree in control engi-
[7] Z. Miao, H. Zhong, Y. Wang, H. Zhang, H. Tan, and R. Fierro, “Low- neering from the Jiangsu University of Science
complexity leader-following formation control of mobile robots using only and Technology, Zhenjiang, China, in 2018.
FOV-constrained visual feedback,” IEEE Trans. Ind. Inform., vol. 18, no. 7, He is currently working toward the Ph.D. de-
pp. 4665–4673, Jul. 2022. gree in control science and engineering with
[8] Z. Xu, T. Yan, S. X. Yang, and S. A. Gadsden, “Distributed leader follower the School of Science, Computing and Engi-
formation control of mobile robots based on bioinspired neural dynamics neering Technologies, Swinburne University of
and adaptive sliding innovation filter,” IEEE Trans. Ind. Inform., early Technology, Melbourne, VIC, Australia, and also
access, doi: 10.1109/TII.2023.3272666. with the School of Mechatronic Engineering
[9] X. Ge, Q.-L. Han, X.-M. Zhang, and D. Ding, “Communication resource- and Automation, Shanghai University, Shang-
efficient vehicle platooning control with various spacing policies,” hai, China.
IEEE/CAA J. Autom. Sinica, early access, doi: 10.1109/JAS.2023.123507. His research interests include motion control for unmanned surface
[10] J. Zhang, X. Shao, W. Zhang, and J. Na, “Path-following control capable vehicles and mobile robots, networked control systems, and multiagent
of reinforcing transient performances for networked mobile robots over a systems.
single curve,” IEEE Trans. Instrum. Meas., vol. 72, pp. 1–12, May 2023.
[11] S. J. Yoo and B.-S. Park, “Connectivity preservation and collision avoid-
ance in networked nonholonomic multi-robot formation systems: Uni-
fied error transformation strategy,” Automatica, vol. 103, pp. 274–281, Xiaohua Ge (Senior Member, IEEE) received
May 2019. the Ph.D. degree in computer engineering from
[12] Z. Liu, B. Wu, J. Dai, and H. Lin, “Distributed communication-aware Central Queensland University, Rockhampton,
motion planning for networked mobile robots under formal specifications,” QLD, Australia, in 2014.
IEEE Trans. Control Netw. Syst., vol. 7, no. 4, pp. 1801–1811, Dec. 2020. From 2015 to 2017, he was a Research Fel-
[13] P. Lu, W. Yu, G. Chen, and X. Yu, “Leaderless consensus of ring-networked low with the Griffith School of Engineering, Grif-
mobile robots via distributed saturated control,” IEEE Trans. Ind. Elec- fith University, Gold Coast, QLD. He is cur-
tron., vol. 67, no. 12, pp. 10723–10731, Dec. 2020. rently a Senior Lecturer with the School of Sci-
[14] X.-M. Zhang, Q.-L. Han, and X. Ge, “A novel approach to H∞ perfor- ence, Computing and Engineering Technolo-
mance analysis of discrete-time networked systems subject to network- gies, Swinburne University of Technology, Mel-
induced delays and malicious packet dropouts,” Automatica, vol. 136, bourne, VIC, Australia. His research interests in-
Feb. 2022, Art. no. 110010. clude networked, secure, and intelligent control and estimation theories,
[15] Z. Han, W. Wang, J.-S. Huang, and Z. Wang, “Distributed adaptive forma- and their applications in electric vehicles, intelligent robotic vehicles,
tion tracking control of mobile robots with event-triggered communication connected automated vehicles, intelligent transportation systems, and
and denial-of-service attacks,” IEEE Trans. Ind. Electron., vol. 70, no. 4, so on.
pp. 4077–4087, Apr. 2023. Dr. Ge is an Associate Editor for IEEE TRANSACTIONS ON INDUSTRIAL
[16] X. Ge, Q.-L. Han, Q. Wu, and X.-M. Zhang, “Resilient and safe pla- INFORMATICS, IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS:
tooning control of connected automated vehicles against intermittent SYSTEMS, IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, IEEE TRANS-
denial-of-service attacks,” IEEE/CAA J. Autom. Sinica, vol. 10, no. 5, ACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, and IEEE/CAA
pp. 1234–1251, May 2023. JOURNAL OF AUTOMATICA SINICA.
[17] W. Fu, J. Qin, W. X. Zheng, Y. Chen, and Y. Kang, “Resilient cooperative
source seeking of double-integratormulti-robot systems under deception
attacks,” IEEE Trans. Ind. Electron., vol. 68, no. 5, pp. 4218–4227,
May 2021.
[18] M. Xie, D. Ding, X. Ge, Q. -L. Han, H. Dong, and Y. Song, “Distributed Hao Xie received the B.Eng. degree in robotics
platooning control of automated vehicles subject to replay attacks based on and mechatronics and the Ph.D. degree in con-
proportional integral observers,” IEEE/CAA J. Autom. Sinica, early access, trol science and engineering from the School of
doi: 10.1109/JAS.2022.105941. Software and Electrical Engineering, Swinburne
[19] W. He, W. Xu, X. Ge, Q.-L. Han, W. Du, and F. Qian, “Secure control University of Technology, Melbourne, VIC, Aus-
of multi-agent systems against malicious attacks: A brief survey,” IEEE tralia, in 2019 and 2023, respectively.
Trans. Ind. Inform., vol. 18, no. 6, pp. 3595–3608, Jun. 2022. He is currently an Engineer with the Shanghai
[20] W. H. Ko, B. Satchidanandan, and P. R. Kumar, “Dynamic watermarking- Aerospace Control Technology Institute, Shang-
based defense of transportation cyber-physical systems,” ACM Trans. hai, China, and a Postdoctoral Fellow with the
Cyber-Phys. Syst., vol. 4, no. 1, pp. 1–21, Nov. 2019. School of Automation Science and Electrical
[21] Q. Liu, Y. Mo, X. Mo, C. Lv, E. Mihankhah, and D. Wang, “Secure pose Engineering, Beihang University, Beijing, China.
estimation for autonomous vehicles under cyber attacks,” in Proc. IEEE His research interests include trajectory tracking and cooperative guid-
Intell. Veh. Symp., 2019, pp. 1583–1588. ance of aerial and ground vehicles.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: SECURE LEADER–FOLLOWER FORMATION CONTROL OF NETWORKED MOBILE ROBOTS UNDER REPLAY ATTACKS 4159

Qing-Long Han (Fellow, IEEE) received the Yu-Long Wang (Member, IEEE) received the
B.Sc. degree in mathematics from Shandong B.Sc. degree in computer science and tech-
Normal University, Jinan, China, in 1983, and nology from Liaocheng University, Liaocheng,
the M.Sc. and Ph.D. degrees in control engi- China, in 2000, and the M.Sc. and Ph.D. de-
neering from the East China University of Sci- grees in control science and engineering from
ence and Technology, Shanghai, China, in 1992 Northeastern University, Shenyang, China, in
and 1997, respectively. 2006 and 2008, respectively.
He is currently a Pro Vice-Chancellor (Re- He was a Postdoctoral Research Fellow and a
search Quality) and a Distinguished Professor Research Fellow with Central Queensland Uni-
with the Swinburne University of Technology, versity, North Rockhampton, QLD, Australia. He
Melbourne, VIC, Australia. He held various aca- was an Academic Visitor with the University of
demic and management positions with Griffith University, Gold Coast Adelaide, Adelaide, SA, Australia. He was a Professor with the Jiangsu
QLD, Australia, and Central Queensland University, Rockhampton, QLD, University of Science and Technology, Zhenjiang, China. In 2017, he
Australia. His research interests include networked control systems, was appointed as an Eastern Scholar by the Municipal Commission of
multi-agent systems, time-delay systems, smart grids, unmanned sur- Education, Shanghai, China, and joined Shanghai University, Shanghai,
face vehicles, and neural networks. China, where he is currently a Professor. His current research interests
Prof. Han was the recipient of the 2021 Norbert Wiener Award (the include deep reinforcement learning and the motion control for marine
Highest Award in systems science and engineering, and cybernetics) vehicles.
and the 2021 M. A. Sargent Medal (the Highest Award of the Electrical
College Board of Engineers Australia). He was also the recipient of the
IEEE Systems, Man, and Cybernetics Society Andrew P. Sage Best
Transactions Paper Award in 2022, 2020, and 2019, the IEEE/CAA
JOURNAL OF AUTOMATICA SINICA Norbert Wiener Review Award in 2020,
and the IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS Outstanding
Paper Award in 2020. He is a Member of the Academia Europaea (the
Academy of Europe). He is a Fellow of the International Federation
of Automatic Control, the Institution of Engineers Australia, and the
Chinese Association of Automation. He is a Highly Cited Researcher in
both Engineering and Computer Science (Clarivate). He was an AdCom
Member of the IEEE Industrial Electronics Society (IES), a Member of
IEEE IES Fellows Committee, a Member of the IEEE IES Publications
Committee, and the Chair of the IEEE IES Technical Committee on
Networked Control Systems. He is the Editor-in-Chief for IEEE/CAA
JOURNAL OF AUTOMATICA SINICA, the Co-Editor-in-Chief for IEEE TRANS-
ACTIONS ON INDUSTRIAL INFORMATICS, and the Co-Editor for Australian
Journal of Electrical and Electronic Engineering.

Jinchuan Zheng (Senior Member, IEEE) re-

ceived the B.Eng. and M.Eng. degrees in
mechatronics engineering from Shanghai Jiao
Tong University, Shanghai, China, in 1999 and
2002, respectively, and the Ph.D. degree in elec-
trical and electronic engineering from Nanyang
Technological University, Singapore, in 2006.
In 2005, he joined the Australian Research
Council Centre of Excellence for Complex Dy-
namic Systems and Control, School of Electri-
cal and Computer Engineering, the University
of Newcastle, Callaghan, NSW, Australia, as a Research Academic.
From 2011 to 2012, he was a Staff Engineer with the Western Digital
Hard Disk Drive R&D Center, Singapore. Since 2012, he has been
with the Swinburne University of Technology, Melbourne, VIC, Australia,
where he is currently an Associate Professor with the School of Sci-
ence, Computing and Engineering Technologies. His research interests
include high-precision motion control systems, electric vehicle control
technology, mobile robots, and biomechatronic devices.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on January 27,2025 at 19:20:25 UTC from IEEE Xplore. Restrictions apply.