IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO.

4, APRIL 2023 1009

Kinematic Control of Serial Manipulators Under

False Data Injection Attack
Yinyan Zhang, Member, IEEE, and Shuai Li, Senior Member, IEEE

Abstract—With advanced communication technologies, cyber- The major classes of attacks in CPS are the false-data injec-
physical systems such as networked industrial control systems can tion attack (FDIA), replay attack, and denial-of-service attack
be monitored and controlled by a remote control center via com-
munication networks. While lots of benefits can be achieved with [1]. In FDIA, false data is injected into the data being trans-
such a configuration, it also brings the concern of cyber attacks to mitted in the communication link, leading to a difference
the industrial control systems, such as networked manipulators between the sent data and the received data. If the data being
that are widely adopted in industrial automation. For such sys- transmitted is replaced with the sent historical data, the attack
tems, a false data injection attack on a control-center-to-manipu-
lator (CC-M) communication channel is undesirable, and has is called a replay attack, for which the attacker needs addi-
negative effects on the manufacture quality. In this paper, we tional effort to record historical data. While the above two
propose a resilient remote kinematic control method for serial classes of attacks are aimed at cheating the sender or the
manipulators undergoing a false data injection attack by leverag- receiver, i.e., destroying integrity, a denial-of-service attack
ing the kinematic model. Theoretical analysis shows that the pro- makes communication resources unavailable to the receiver so
posed method can guarantee asymptotic convergence of the regu-
lation error to zero in the presence of a type of false data injec- that the receiver cannot receive the data sent from the sender,
tion attack. The efficacy of the proposed method is validated via i.e., destroying availability. In the control of CPS, receiving
simulations. bad control commands could be much more harmful to the
Index Terms—Cyber-physical systems, false data injection attack, plant than getting no remote control commands.
manipulators, remote kinematic control. Current research on the security issues of CPS mainly
includes the detection and design of various attacks, and the
I. Introduction design of attack-resilient control methods for corresponding
HE control and monitor of industrial control systems attacks [1]–[8]. For example, by utilizing current and histori-
T composed of sensing, actuation and communication capa-
bilities, which is also a typical type of cyber-physical system
cal data, Ye and Zhang [1] proposed a summation detector for
the detection of FDIA for a class of linear cyber-physical sys-
(CPS) from a remote control center, has been made possible tems with white noise. Kashima and Inoue [2] proposed a
due to the developments in network technologies. Efforts have detection mechanism for replay attacks in CPS systems with
already been made to develop applications of techniques using quantized signals. In terms of the design of attacks, Liang
the Internet of things (IoT) to industrial control systems, et al. [5] reviewed recent advancements in the design of FDIA
which is referred to as industrial IoT. Owing to the vulnerabil- in modern power systems, among which some methods are
ity of network transmission, the security issue in control of also applicable to other CPS. For linear-time invariant plants
CPSs should be addressed. However, traditional research on in CPS, Pang et al. [6] proposed a two-channel FDIA method
security in computer science mainly focusing on the protec- by which an attacker can degrade the output tracking perfor-
tion of information is not enough for the control of CPSs as mance without being detected, which requires knowing the
attacks on CPSs can damage physical plants, which is undesir- corresponding system matrices. Recently, Lu and Yang [7]
able for critical infrastructures. In industrial control systems, a proposed a sparse undetectable attack algorithm to degrade the
properly designed attack that is aimed at causing negative performance of remote state estimation. For CPS, only per-
effects to the quality of products is also undesirable, which forming detection and prevention of attacks is not sufficient.
deserves more attention in the era of unmanned factories. For the safety and proper working of CPS, researchers pro-
Manuscript received July 16, 2022; accepted September 8, 2022. This work posed resilient control methods to mitigate or even eliminate
was supported in part by the National Natural Science Foundation of China the effects of attacks on the control performance [9]–[12]. For
(62206109) and the Fundamental Research Funds for the Central Universities
(21620346). Recommended by Associate Editor Peter Liu. (Corresponding instance, Modares et al. [9 ] proposed a resilient control
author: Yinyan Zhang.) approach for the leader-follower synchronization of linear
Citation: Y. Y. Zhang and S. Li, “Kinematic control of serial manipulators time-invariant multiagent systems subject to attacks on sen-
under false data injection attack,” IEEE/CAA J. Autom. Sinica, vol. 10, no. 4,
pp. 1009–1019, Apr. 2023. sors and actuators, by which bounded synchronization error
Y. Y. Zhang is with the College of Cyber Security, Jinan University, can be achieved. From the literature and the references
Guangzhou, China (e-mail: [email protected]). therein, it can be found that most of the existing results for the
S. Li is with the College of Engineering, Swansea University, Swansea, resilient control of CPS focus on systems that can be modeled
United Kingdom (e-mail: [email protected]).
Color versions of one or more of the figures in this paper are available
as linear time-invariant systems.
online at http://ieeexplore.ieee.org. Due to its important role in industrial automation, the con-
Digital Object Identifier 10.1109/JAS.2023.123132 trol of industrial serial manipulators is widely investigated.

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 1010 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 4, APRIL 2023

Typical control methods for industrial serial manipulators can summarized as follows.
be briefly divided into dynamic control [13]–[15] and kine- 1) For the first time, a novel method is proposed for the
matic control [16]–[21]. Note that, for general serial indus- remote kinematic control of manipulators, which can simulta-
trial manipulators of degree n, when the primary task requires neously detect and remedy a false data injection attack on the
m degrees of freedom with m < n, the manipulators are said to control-center-to-manipulator link.
be redundant. Dynamic control concentrates on the dynamic 2) The performance of the proposed method is theoretically
model of the controlled manipulator, while kinematic control guaranteed, and it can drive the end-effector position and ori-
focuses on the kinematic model. However, the traditional con- entation error to converge to zero in regulation tasks.
trol law design, such as those in [13]–[21], does not consider 3) Simulation results on an ABB IRB 140 industrial manip-
the effect of attacks, which may hinder the reliability of these ulator show that the proposed method outperforms the exist-
control laws for the remote control of industrial manipulators ing method.
in the IoT era. It should be noted that the security issues in the The rest of this paper is organized as follows. The investi-
control of industrial manipulators have already attracted atten- gated problem is formulated in Section II, followed by the
tion in the security community, which are vulnerable to description of the proposed method in Section III. Then, theo-
attacks [22]. Recently, Dong et al. [4 ] proposed novel meth- retical analysis for the proposed method is given in Section
ods for the design and detection of FDIA targeted at the IV, followed by the illustration of a simulative example in
remote dynamic control of manipulators. However, they did Section V. Finally, conclusions are drawn in Section VI.
not provide a countermeasure for realizing the remote
II. Problem Formulation
dynamic control of a manipulator when it is under FDIA. To
our knowledge, there are no existing results addressing the In this section, we formulate the problem investigated in this
remote kinematic control of manipulators under FDIA. paper. We first show the model of the kinematic control of
Motivated by the above discussion, in this paper, we pro- manipulators, followed by the model of the FDIA. Then, the
pose a novel simultaneous detection and control method for remote kinematic control problem under FDIA is described.
the remote kinematic control of manipulators under FDIA. It A. Kinematic Control of Manipulator
should be noted that our objective is to ensure that the manip-
ulator can still complete the given task even when there is For a serial manipulator with n joints, the end-effector posi-
FDIA, which cannot be achieved by using the message tion and orientation vector r = [r1 , r2 , . . . , rm ]T ∈ Rm at time
authentication technique [23]. The role of the attacker in the instant t satisfies
remote control process is depicted in Fig. 1. The implementa- r(t) = f (θ(t)) (1)
tion of kinematic control algorithms on the remote control where θ = [θ1 , θ2 , . . . , θn ∈
]T Rn denotes the vector of joint
center would allow the usage of advanced intelligent control angles, and f (·) : Rn → Rm is the nonlinear mapping from the
methods like deep learning algorithms for complicated tasks, joint space to the workspace. When m < n, the manipulator is
which is very time-consuming and may not be feasible to be redundant. The mapping f (·) can be derived for a serial
deployed in the manipulator owing to limited computational manipulator by following the Denavit and Hartenberg (D-H)
resources. This work is the first step towards secure and intel- convention given the D-H parameters, which specify the kine-
ligent kinematic control of manipulators by leveraging deep matic structure and configuration of the manipulator
learning algorithms for complicated tasks. It is assumed that [24]–[26].
the attacker can read and modify the data sent by the remote The kinematic regulation control problem of manipulators
control center [4]. Our work considers the case that the requires finding a joint vector for a given constant desired
attacker persistently modifies command from the remote con- end-effector configuration rd ∈ Rm such that r = rd is subject
trol center. The case where the communication link between to joint constraints. In fact, for a physical manipulator, the
the manipulator and the remote control center breaks corre- joint angles and their derivatives of any order are limited, due
sponds to the denial-of-service attack [12], which is out of the to the manipulator structure and the constraints on the actua-
scope of this paper. The main contributions of this paper are tors of the joints. In this paper, we consider both joint angle
and joint velocity limits
θ θ θ− ≤ θ(t) ≤ θ+
Communication
network θ̇− ≤ θ̇(t) ≤ θ̇+ (2)
ur us Remote control center
Manipulator where the operator ≤ is element-wise and θ̇ = dθ/dt ∈ Rn
denotes the joint velocity vector; θ− ∈ Rn and θ+ ∈ Rn denote
the lower bound and upper bound of the joint angle vector of
the manipulator; and θ̇− ∈ Rn and θ̇+ ∈ Rn denote the lower
Attacker bound and upper bound of the joint velocity vector of the
manipulator. According to [27], the two constraints (2) can be
Fig. 1. Demonstration of the data flow in the communication process,
unified into the following one:
where θ denotes the joint angle vector of the manipulator, us denotes the con-
trol command sent by the control center, and ur denotes the control command θ̇ ∈ Ω (3)
received by the controlled manipulator. where the set Ω is defined as

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 ZHANG AND LI: KINEMATIC CONTROL OF SERIAL MANIPULATORS UNDER FALSE DATA INJECTION ATTACK 1011

Ω = {η− ≤ θ̇ ≤ η+ } For the above attack, we have the following remark.

with η− = max{θ̇− , −κ(θ − θ− )} , η+ = min{θ̇+ , −κ(θ − θ+ )}, and Remark 1: The false data injection attack described as (8) is
parameter κ satisfies a proportional-type attack. It should be noted that propor-
{ + −} tional-type false data injection attacks were also used in [28],
θ̇ − θ̇ [29] for other systems. The considered attack model (8) has
κ > max . (4)
i∈{1,2,...,n} θ+ − θ− two advantages from the perspective of attackers. Firstly, the
To handle both limits, it is better to consider the kinematic attacker does not need any information about the controller or
model at the joint velocity level by calculating the time the kinematic model of the manipulator, which means that
derivative on both sides of (1), by which we have such an attack is easier to deploy than other complicated
attacks on cyber-physical systems, such as those reported in
ṙ(t) = J(θ(t))θ̇(t) (5)
[30]–[32]. Secondly, given β ∈ (0, 1], the attack model ensures
where ṙ = dr/dt and J(θ) = d∂(θ)/∂θ is called the Jacobian that ur ∈ Ω , which means that the received velocity command
matrix of the manipulator kinematics. Let does not violate the velocity limit of the manipulator, making
u(t) = θ̇(t) it more difficult for the remotely controlled manipulator to
denote the control input. The velocity-level kinematic model find the attack compared to the case where ua (t) ∈ R. It should
(5) can be rewritten as be noted that depending on the goal of attacks, the mathemati-
cal properties of attack models differ from general uncertain-
ṙ(t) = J(θ(t))u(t). (6) ties in the control signals. Normally, a high-level attacker
Thus, the kinematic regulation control problem is equiva- would try to attack a system while being not found. Thus, to
lent to finding the input u(t) ∈ Ω with Ω defined in (3) for launch an attack on the system considered in our paper, from
given rd such that r(t) − rd converges to zero. the perspective of the attacker, the modified velocity com-
mand should satisfy the velocity limits of the manipulator.
B. FDIA Model
Otherwise, the manipulator can easily identify that there is a
As shown in Fig. 1 , we assume that the attacker only per- false data injection attack, especially if the received command
form attacks on data sent from the control center to the con- violates the velocity limits of the manipulator. In terms of
trolled manipulator. Let us ∈ Rn denote the control command determining how we deal with the attack, we are fundamen-
sent by the control center, and ur ∈ Rn denote the control com- tally dealing with the kinematic control of manipulators sub-
mand received by the controlled manipulator. The relation- ject to joint limits and false data injection attacks. To the best
ship between us and ur can be described as follows: of our knowledge, such a problem has not been addressed in
ur (t) = us (t) + ua (t) (7) the robotics community, and it is not straightforward to adapt
where ua (t) ∈ Rn
denotes the injected data. Obviously, when the existing controller design with uncertainty in control
ua (t) = 0 , we have ur (t) = us (t), i.e., the manipulator receives inputs to our problem. One example of a considered problem
the correct control command sent by the remote control cen- in the application domain is bilateral teleoperation systems
ter. Generally, an attacker does not want the FDIA to be [4]. Note that [4] only deals with the detection of the attack.
detected by the remotely controlled manipulator. If ur vio- For the convenience of illustration, in the rest of the paper,
lates the joint constraints, then the remotely controlled manip- the argument t is omitted somewhere.
ulator can easily detect the attack and will not execute the con-
C. Remote Kinematic Control Problem Under FDIA
trol command. Thus, a properly designed attack should not
violate the joint constraints. If the attacker does not know the In the remote kinematic control of manipulators, as shown
joint limits, it can be readily verified that the following set- in Fig. 1 , the remote control center sends control commands
ting can guarantee that ur ∈ Ω if us (t) ∈ Ω : us to the manipulator and the manipulator sends its state infor-
mation θ to the remote control center via the communication
ua (t) = (β − 1)us (t) (8) network.
with constant gain parameter β ∈ (0, 1]. In other words, the Assumption 1 [4]: The network condition is perfect, e.g.,
detailed formula of injected data ua (t) with respect to time t is there are no time delays or data losses.
related to the command us (t). Thus, we cannot give a detailed Assumption 1 means that the transmission of control com-
explanation of ua (t) before knowing us (t). The FDIA model mands via the communication network is only affected by
guarantees that the input constraints are always not violated so attacks. This assumption allows us to theoretically analyze the
that the manipulator cannot detect the FDIA by simply check- performance of the proposed scheme.
ing whether the command sent from the remote control center Assumption 2: The feedback signal from the manipulator is
violates the input constraints. In terms of the parameter β , it not manipulated by the attacker.
can be viewed as a measure for the strength of the FDIA. In Assumption 2 serves as a basis for designing resilient con-
particular, when β = 1, it means that there is no FDIA. For the trol laws for the manipulator under FDIA. If the feedback sig-
false data injection attack in this paper, when ua (t) , 0, the nal from the manipulator is also manipulated by the attacker,
attack frequency is 100% since there always exists FDIA it means that there is not a correct feedback signal. From the
according to (7) and (8). Meanwhile, the attack intensity of the perspective of control systems, the realization of a stable
FDIA considered in this paper is characterized by β ∈ (0, 1]. closed-loop control system relies on the feedback signal. If we

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 1012 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 4, APRIL 2023

cannot have the correct feedback signal, we cannot design the θ

corresponding controller to stabilize the system. FDIA detection
The kinematics of the remotely controlled manipulator is Controlled
system (12)
manipulator
described as follows: β (9)
ṙ = J(θ)ur . (9) Control law us ur
Attacker (7)
(11)
With (7) and (9) the kinematics of the remotely controlled
manipulator under FDIA is further obtained as follows:
rd
ṙ = J(θ)(us + ua ). (10)
Remote control center Communication network
The remote kinematic control problem investigated in this
paper requires the remote control center to find the control Fig. 2. Demonstration of the information flow of the proposed resilient
input us for a given constant rd such that r − rd converges to remote kinematic control method for manipulators under FDIA described by
zero under FDIA described by (7) and (8). (7) and (8), where θ is the joint angle measurement of the remotely controlled
Regarding the effect of FDIA described by (7) and (8) on manipulator, ur is the control command received by the remotely controlled
regulation control performance, we provide the following manipulator, us is the control command sent by the remote control center, rd
remark. is the desired end-effector configuration of the remotely controlled manipula-
Remark 2: For regulation control tasks, it can be expected tor, and β̂ is the estimation of the FDIA parameter β.
that the FDIA described by (7) and (8) will make the remotely
controlled manipulator cost more time to complete the task Regarding the FDIA detection system, we provide the fol-
given by the remote control center, i.e., make the conver- lowing remark.
gence of e = r − rd to 0 slower, thus degrading remote control Remark 3: The design of the FDIA detection system (12)
efficiency. fully utilizes the kinematic model of the remotely controlled
manipulator under normal operation, for which ṙ = J(θ)us. As
III. Controller Design
seen from (12), given that β̂(0) = 1 and r̂(0) = f (θ(0)), the state
In this section, the control command design for the remote
variable r̂ essentially monitors whether the remotely con-
kinematic control problem of manipulators under FDIA is pre-
trolled manipulator works as expected. In other words, if
sented.
ur = us , then we have r̂ = r . If there is a difference between r̂
As joint constraints need to be satisfied, we first recall the
and r , then the state variable β̂ is updated with r̂ − f (θ) as
definition of the projection operator as follows.
Definition 1: The projection of vector y ∈ R p on the set S of feedback so as to drive r̂ to converge to r. The feedback term
p-dimension real-valued vectors is denoted as PS (y) , which is −s(β̂ − PΓ (β̂)) forces β̂ to be within the set Γ. Thus, if β = 1, it
calculated as follows: means that there is no FDIA; otherwise, it means that there is
FDIA.
PS (y) = argminx∈S ∥y − x∥2
In terms of the implementation of the proposed approach,
where ∥ · ∥2 denotes the Euclidean norm (i.e., 2-norm). we have the following remark.
To solve the remote kinematic control problem with FDIA Remark 4: In practice, for the sake of data transmission, we
described by (7) and (8), we propose the following law of con- can discretize the proposed detection system via the Euler dif-
trol command: ference rule [4]. Let the sampling gap be denoted by τ > 0 . Let
( T ) k = 0, 1, 2, . . ., denotes the sampling index. By the Euler differ-
αJ (θ)(rd − f (θ))
u s = PΩ (11) ence rule, the discrete-time implementation of the proposed
β̂
approach is given as follows:
where PΩ (·) is the projection operator for which α > 0 ∈ R is a ( T k )
parameter to scale the feedback strength, and β̂ ∈ R is the αJ (θ )(rd − f (θk ))
u s = PΩ
k
detection variable of the following FDIA detection system: β̂k
r̂˙ = β̂J(θ)u s − k1 (r̂ − f (θ)) r̂k+1 = r̂k + τβ̂k J(θk )uks − k1 (r̂k − f (θk ))
β̂˙ = −k2 uTs J T (θ)(r̂ − f (θ)) − s(β̂ − PΓ (β̂)) (12) s J (θ )(r̂ − f (θ )) − s(β̂ − PΓ (β̂ ))
β̂k+1 = β̂k − τk2 ukT T k k k k k

with s ∈ R defined as follows: where uks = u s (kτ), r̂k = r̂(kτ), and β̂k = β̂(kτ). Thus, at each



 1, if β̂ ∈ Γ time instant, the remote control center reads the received joint


 angle vector of the remotely controlled manipulator, updates
s= |k2 uTs J T (θ)(r̂ − f (θ))| (13)


 , if β̂ < Γ and sends the control command, and then updates the vari-
 |β̂ − P (β̂)|
Γ ables of the FDIA detection system. Without loss of general-
where | · | denotes the absolute value, r̂ ∈ Rm and ûa ∈ Rn are ity, we have u0 = 0 ∈ Rn , r̂0 = f (θ0 ), and β̂0 = 1. Note that
state variables of the FDIA detection system, and Γ = {x ∈ designing control laws in the continuous domain and imple-
R|ϵ ≤ x ≤ 1}, where ϵ > 0 is set to be sufficiently small. The menting them via the Euler difference rule are widely consid-
block diagram of the proposed control approach is shown in ered to be feasible (see [33]–[35] with experimental valida-
Fig. 2. tions for example). The only requirement is that we need to set

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 ZHANG AND LI: KINEMATIC CONTROL OF SERIAL MANIPULATORS UNDER FALSE DATA INJECTION ATTACK 1013

the sampling gap τ > 0 to be small, e.g., τ = 0.001 s. As long Thus, according to the chain rule for the calculation of
as the hardware allows, with smaller sampling gaps, better derivatives,
control performance can be expected. ∂L ˙ ˙
L̇ = ( )β̂ = β̂(β̂ − PΓ (β̂)).
IV. Theoretical Analysis ∂β̂
In this section, theoretical results are presented for the pro- Evidently, if β̂ ∈ Γ , then β̂ − PΓ (β̂) = 0 , by which L̇ = 0 ; if
posed resilient remote control method for manipulators under β̂ < Γ , then, β̂ − PΓ (β̂) , 0 . With the FDIA detection system
the false data injection attack. described as (12) and (13), we further have
Before showing the main theoretical result, we present three
lemmas. L̇ = (−k2 uTs J T (θ)(r̂ − f (θ)) − s(β̂ − PΓ (β̂)))(β̂ − PΓ (β̂))
Lemma 1: If y ∈ Γ, then (x − y)(x − PΓ (x)) ≥ 0, ∀x ∈ R . (
Proof: From Definition 1 and the definition of Γ, we have, | − k2 uTs J T (θ)(r̂ − f (θ))|
= − k2 uTs J T (θ)(r̂ − f (θ)) −
∀x ∈ R , |β̂ − PΓ (β̂)|
 )


1, if x > 1 × (β̂ − PΓ (β̂)) (β̂ − PΓ (β̂))

PΓ (x) = 
 x, if ϵ ≤ x ≤ 1


ϵ, if x < ϵ. = −k2 uTs J T (θ)(r̂ − f (θ))(β̂ − PΓ (β̂))
Thus, given that y ∈ Γ, i.e., ϵ ≤ y ≤ 1, we have the following (β̂ − PΓ (β̂))2 |k2 uTs J T (θ)(r̂ − f (θ))|
results. −
|β̂ − PΓ (β̂)|
1) If x > 1, then x − y > 1 − y ≥ 0 and x − PΓ (x) = x − 1 > 0,
by which (x − y)(x − PΓ (x)) > 0. ≤ | − k2 uTs J T (θ)(r̂ − f (θ))||β̂ − PΓ (β̂)|
2) If ϵ ≤ x ≤ 1, then x − PΓ (x) = x − x = 0, by which
(x − y)(x − PΓ (x)) = 0. (β̂ − PΓ (β̂))2 |k2 uTs J T (θ)(r̂ − f (θ))|
−
3) If x < ϵ , then x − y < ϵ − y ≤ 0 and x − PΓ (x) = x − ϵ < 0 , |β̂ − PΓ (β̂)|
by which (x − y)(x − PΓ (x)) > 0.
= | − k2 uTs J T (θ)(r̂ − f (θ))||β̂ − PΓ (β̂)|
By summarizing the above results, it is concluded that, if
y ∈ Γ, then − |β̂ − PΓ (β̂)||k2 uTs J T (θ)(r̂ − f (θ))|
(x − y)(x − PΓ (x)) ≥ 0, ∀x ∈ R. ■ = 0.
Lemma 2: ∀x ∈ Rn,
we have Ω (−x) ≤ 0 .
xT P It follows hat V̇ ≤ 0 . Together with V(0) = 0 since β̂(0) ∈ Γ,
Proof: According to Definition 1, ∀x ∈ Rn, and V(t) ≥ 0, it is concluded that β̂(t) ∈ Γ for all t > 0 . ■
PΩ (−x) = argminy∈Ω ∥ − x − y∥2 Based on Lemmas 1–3, we have the following theorem
about the performance of the proposed remote kinematic con-
by which we have, ∀y ∈ Ω, trol approach.
∥ − x − PΩ (−x)∥2 ≤ ∥ − x − y∥2 . Theorem 1: If the remote control center adopts a control law
(11) with FDIA detection system (12), under FDIA described
By the definition of Ω, we have 0 ∈ Ω. Thus, we further
by equations (7) and (8), given that β̂(0) ∈ Γ, and J(θ) is full-
have
rank, for constant rd , the regulation error e = r − rd of the
∥ − x − PΩ (−x)∥2 ≤ ∥ − x∥2 remotely controlled manipulator (9) globally asymptotically
with y = 0. It follows that: converges to zero. Besides, if β ∈ Γ and there exists t2 > t1 > 0
such that
∥ − x − PΩ (−x)∥22 ≤ ∥ − x∥22
w t2
i.e., uTs (t)J T (θ(t))J(θ(t))u s (t)dt > 0
t1
∥ − x∥22 + ∥PΩ (−x)∥22 + 2xT PΩ (−x) ≤ ∥ − x∥22
then the state variable β̂ of the FDIA detection system (12)
which yields
asymptotically converges to β.
2xT PΩ (−x) ≤ −∥PΩ (−x)∥22 ≤ 0. (14) Proof: In view of r = f (θ), the FDIA detection system (12)
can be rewritten as
It follows that, ∀x ∈ Rn,Ω (−x) ≤ 0 .
xT P ■
Lemma 3: If β̂(0) ∈ Γ, given that β ∈ Γ , for the FDIA detec-
r̂˙ = β̂J(θ)u s − k1 (r̂ − r)
tion system (12), β̂(t) ∈ Γ for all t > 0 .
Proof: Consider the following candidate function: β̂˙ = −k2 uTs J T (θ)(r̂ − r) − s(β̂ − PΓ (β̂)). (15)
L = (β̂ − PΓ (β̂))2 /2 ≥ 0. Let
According to Appendix E.9 of [36], r̃ = r̂ − r,
∂L β̃ = β̂ − β.
= β̂ − PΓ (β̂).
∂β̂ Then, (15) is further written as

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 1014 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 4, APRIL 2023

r̂˙ = β̂J(θ)u s − k1 r̃ −k1 r̃T r̃ = −k1 ∥r̃∥22 ≤ 0. (23)

β̂˙ = −k2 uTs J T (θ)r̃ − s(β̂ − PΓ (β̂)). (16) It follows from (20)–(23) that:
Since β is a constant, we have β̇ = 0, by which β̃˙ = β̂˙ . Mean- V̇ ≤ 0.
while, the kinematic model (9) of the remotely controlled In the following, LaSalle’s invariance principle [37] is
manipulator under FDIA described by (7) and (8) can be adopted to find the largest invariant set.
rewritten as Let V̇ = 0 , and from (20)–(23), we have
ṙ = βJ(θ)u s (17) r̃ = 0
which together with (16) yields
sβ̃(β̂ − PΓ (β̂)) = 0
r̃˙ = β̃J(θ)u s − k1 r̃ ( )
α T −αJ T (θ)e
e J(θ)PΩ =0 (24)
β̃˙ = −k2 uTs J T (θ)r̃ − s(β̂ − PΓ (β̂)). (18) β̂ β̂
By substituting the proposed control law (11) into the kine- at the invariant set.
matic model (17) under FDIA, with e = r − rd , the closed-loop In the invariant set, in view of the third equation in (24) and
kinematic model of the manipulator is further obtained as fol- (14), we have
lows: ( )
( ) −αJ T (θ)e
−αJ T (θ)e 0 ≤ − PΩ ≤0
ṙ = βJ(θ)PΩ . (19) β̂ 2
β̂
which indicates that
Consider the following Lyapunov candidate function:
( )
r̃T r̃ β̃2 eT e −αJ T (θ)e
V= + + . PΩ = 0.
2 2k2 2 β̂
Obviously, V is radially unbounded. Note that for constant By the definition of Ω, 0 is not on the boundary of set Ω and
rd , we have ė = ṙ . In view of (18) and (19), the time deriva- 0 ∈ Ω. Thus, by Definition 1, with α > 0 and β̂ > 0, we further
tive of V along the state trajectory of the system satisfies conclude that J T (θ)e = 0. As a result, given that J is full-rank,
β̃β̃˙ in view of J T (θ) ∈ Rn×m with n > m, we further have e = 0.
V̇ = r̃T r̃˙ + + eT ė In the invariant set, based on Lemma 3, sβ̃(β̂−PΓ (β̂)) = 0
k2
means β̃ = 0 or β̂ = PΓ (β̂), from both of which it is concluded
β̃ ( that β̂ ∈ Γ given that β ∈ Γ . It follows that, in view of (18),
= r̃T (β̃J(θ)u s − k1 r̃) + − k2 uTs J T (θ)r̃ − s(β̂
k2 r̃˙ = 0, we further have:
) ( )
−αJ T (θ)e β̃J(θ)u s = 0
− PΓ (β̂)) + βe J(θ)PΩ
T
β̂
β̃˙ = 0.
s
= −k1 r̃T r̃ − β̃(β̂ − PΓ (β̂)) As a result, β̃ is a constant in the invariant set, and satisfies
k2
( ) β̃J(θ)u s = 0 , by which we have
ββ̂ α T −αJ T (θ)e
+ e J(θ)PΩ . (20) β̃uTs J T (θ)J(θ)u s = 0.
α β̂ β̂
Given that β ∈ Γ , according to Lemma 1, we have Integrating both sides of the above equation at time region
[t1 , t2 ] with t2 > t1 > 0 yields
β̃(β̂ − PΓ (β̂)) = (β̂ − β)(β̂ − PΓ (β̂)) ≥ 0. w t2
Together with k2 > 0 and s > 0, we further have β̃ uT (t) s J T (θ(t))J(θ(t))u s (t)dt = 0.
t1
s It follows that, if there exists t2 > t1 > 0 such that:
− β̃(β̂ − PΓ (β̂)) ≤ 0. (21)
k2 w t2
Since β̂(0) ∈ Γ, according to Lemma 3, we have β̂ > 0, uT (t) s J T (θ(t))J(θ(t))u s (t)dt > 0
t1
∀t > 0 . Then, according to Lemma 2, with x = αJ T (θ)e/β̂ , we
then we have β̃ = 0.
have
By summarizing the above analysis, according to the
( )
α T −αJ T (θ)e LaSalle’s invariance principle, it is concluded that, given that
e J(θ)PΩ ≤0 J(θ) is full-rank, rd is a constant, β̂(0) ∈ Γ, β ∈ Γ , and there
β̂ β̂
exists t2 > t1 > 0 such that
which together with α > 0 and β > 0 yields
w t2
( ) uT (t) s J T (θ(t))J(θ(t))u s (t)dt > 0
ββ̂ α T −αJ T (θ)e
e J(θ)PΩ ≤ 0. (22) t1
α β̂ β̂ the regulation error e = r − rd of the remotely controlled
Besides, because k1 > 0 , we have manipulator (9) globally asymptotically converges to zero,

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 ZHANG AND LI: KINEMATIC CONTROL OF SERIAL MANIPULATORS UNDER FALSE DATA INJECTION ATTACK 1015

and state variable β̂ of the FDIA detection system (12) asymp- TABLE I
totically converges to β. ■ D-H Parameters of the ABB IRB 140 Manipulator
For the case that rd is time-varying, the bounded tracking Link ai (m) di (m) αi (rad)
error can be guaranteed by following similar steps in the proof 1 0.07 0.352 −π/2
of Theorem 1. It should be noted that stability analysis of most 2 0 0
0.36
of the existing works on kinematic control of manipulators is
3 0 0 π/2
conducted under the assumption that rd is a constant (see [20],
[24], [27] for example). 4 0 0.38 −π/2
In terms of the conditions stated in Theorem 1, we have the 5 0 0 π/2
following remark. 6 0 0.065 π/2
Remark 5: Theorem 1 requires that J(θ) ∈ Rm×n is full-rank.
This is required in all the existing kinematics control appro-
aches [16]–[20]. Generally, for manipulators, this assumption
is weak. For example, for a 6 degrees-of-freedom manipula-
tor with the dimension of workspace being 3, J(θ) is a 3-by-6
matrix, for which the full-rank requirement can be readily sat-
isfied. Meanwhile, the condition where t2 > t1 > 0 such that
r t2 T
t1 u (t) s J (θ(t))J(θ(t))u s (t)dt > 0 is often referred to as per-
T

sistent excitation, which is standard for guaranteeing exact

parameter convergence [38]. The persistent excitation condi-
tion can be interpreted as the richness of the signals. The con-
dition can be satisfied by adding noises into the signals. Fur-
thermore, the condition that β ∈ Γ can be satisfied by setting
the value of ϵ > 0 to be sufficiently small. In addition, the con-
dition that β̂(0) ∈ Γ can be readily satisfied since the set Γ is
known to the remote control center.
According to the above analysis, the time it takes to detect Fig. 3. Demonstration of the ABB IRB 140 industrial manipulator in V-
and address situations with a false data injection attack REP.
depends on the convergence of β. To have faster convergence,
we can set large values of k1 and k2. For the illustrative exam- TABLE II
ple shown in Section V, it takes less than 2 seconds. Joint Constraints of the ABB IRB 140 Manipulator

V. Simulations and Comparison

Joint i θi+ (°) θi− (°) θ̇i+ (°/s) θ̇i− (°/s)
1 180 −180 200 −200
In this section, simulations are presented to show the effec-
tiveness and advantages of the proposed approach for the 2 110 −90 200 −200
remote kinematic control of manipulators under FDIA. 3 50 −230 260 −260
4 200 −200 360 −360
A. Static Reference Output rd
5 115 −115 360 −360
We consider a 6-axis ABB IRB 140 industrial manipulator,
6 180 −180 450 −450
whose D-H parameter is shown in Table I [39]. A figure of the
manipulator in V-rep is shown in Fig. 3. The manipulator has
six revolute joints with θ ∈ R6 . The joint limits of the manipu- TABLE III
lator is listed in Table II . Based on Table II and (4), we set Parameter Setting and Initial Values of the Resilient
Kinematic Control Method
κ = 5. In the simulation, only the end-effector position
r = [rX , rY , rZ ]T ∈ R3 is considered, and thus the manipulator Parameter i k1 k2 α β̂(0) r̂(0)
can be viewed as a redundant manipulator. The simulation set- Value 50 1 1000 1 f (θ(0))
ting for the remote control follows the block diagram shown
in Fig. 2, with the FDIA modeled as (7) and (8) with β = 0.01, verges to β = 0.01. From Figs. 4(c) and 4(d), the end-effector
where the attack is performed during the whole simulation error e asymptotically converges to 0 with r converging to rd .
process. Accordingly, the parameter of Γ is set to ϵ = 10−4 .
Besides, in view of Fig. 4 (e) and Table II , the sent control
The desired end-effector position is set as rd = [−0.0127,
commands do not violate the joint velocity constraints during
0.0110, 0.3034]T . The setting of parameter values and initial
values of the FDIA detection system (12) of the proposed the whole remote control process, which, together with the
approach is shown in Table III . As seen from Fig. 4 (a), the model of the FDIA shown in (7) and (8), indicates that the
state variable r̂ of the FDIA detection system (12) converges control commands received by the remotely controlled manip-
to the end-effector coordinate r of the controlled manipulator. ulator do not violate the joint velocity constraints. In addition,
From Fig. 4 (b), it is found that the detection variable β̂ is Figs. 4(f) and 5 show that the joint angle limits are also not
always within set Γ, i.e., 10−4 ≤ β̂ ≤ 1, and asymptotically con- violated during the whole remote control process. The results

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 1016 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 4, APRIL 2023

0.4 100
r1 validate Theorem 1 and demonstrate the effectiveness of the
0.2 r2 10−1 β̂ proposed method for the remote kinematic control under
0 r3 FDIA.
10−2
r̂1 We have the following remark regarding the sharp change
−0.2 10−3
r̂2
−0.4 r̂3 in Fig. 4(b).
10−4
Remark 6: Fig. 4(b) shows the changes of the estimation β̂
−0.6
0 2 4 6 8 10 0 2 4 6 8 10 for the parameter of the FDIA model during the remote con-
t (s) t (s)
trol process. The sharp change only happens for β̂ , which is
(a) (b)
fine. In fact, the sharp change shows the fast convergence of
0.1 0.4
(m) the estimation for the parameter β in the FDIA model. In prac-
0
0.2 (m) rX tice, we only care about the control performance of the manip-
−0.1 eX rY
−0.2 eY 0 rZ ulator. As seen from the subfigures in Figs. 4 and 5, the track-
−0.3 eZ rdX
−0.2 ing error converges to zero, and the joint constraints are not
−0.4 rdY
−0.4 rdZ violated. It should be noted that the proposed method has a
−0.5
−0.6 −0.6
theoretical guarantee, and rigorous proofs are also given in
0 2 4 6 8 10 0 2 4 6 8 10 this paper. Thus, the proposed control law is reliable. To sup-
t (s) t (s) port this claim, we also conduct a simulation for the case
(c) (d)
where β = 0.001, with other parameters being fixed. As seen
400 50
300
(°/s) us1 (°) θ1 from Fig. 6, in this case, β̂ converges to 0.001 (i.e., the FDIA
200 us2 0 θ2 is accurately detected), and the tracking error still converges
100 us3 −50 θ3
θ4
to zero with the joint limits not being violated (i.e., the remote
0 us4
−100 us5 −100 θ5 control performance is maintained). The results show that, for
−200 θ6 the proposed method, the sharp change in the transient behav-
us6 −150
−300 ior of β̂ does not affect the stable remote control performance
−400 −200
0 2 4 6 8 10 0 2 4 6 8 10 of the manipulator under FDIA, and the proposed control law
t (s) t (s) is reliable.
(e) (f)
We also perform simulations to show the advantages of the
proposed method compared with a traditional control law [16]
Fig. 4. Data profiles during the remote kinematic control of the ABB IRB
( )
140 manipulator under FDIA described by (7) and (8) with β = 0.01 when the u s = PΩ αJ T (θ)(rd − f (θ)) . (25)
proposed remote control law (11) synthesized by the FDIA detection system
(12) is employed. (a) Profiles of end-effector coordinates of the manipulator r
Under the same settings for the parameters, the FDIA
and the state variable r̂ of the FDIA detection system (12); (b) Profile of
attack, and initial state, simulations are performed. The com-
detection variable β̂ of the FDIA detection system (12); (c) Profiles of end- parison of the end-effector error norm ∥e∥2 is shown in Fig. 7.
effector coordinate error e = r − rd ; (d) Profiles of the end-effector coordinate
As seen from Fig. 7 , although the control law (25) can drive
r of the remotely controlled manipulator and the desired end-effector coordi- the end-effector error norm ∥e∥2 to converge to zero under the
nate rd ; (e) Profiles of the remote control commands us sent by the remote FDIA, its convergence is much slower compared with the sce-
control center; (f) Profiles of joint angles θ of the controlled manipulator. nario where the proposed method under the same parameter
setting is adopted. This shows the advantage of the proposed
method in terms of its resilience to the FDIA described by (7)
4 and (8) and the capability to remedy for the effect of the attack
θ·1
θ·2
(rad/s) on the convergence of the end-effector error of the remotely
θ·3 controlled manipulator.
2
θ·4
θ·5 B. Time-Varying Reference Output rd
θ·6
0 For the case where rd is time-varying, simulation is also
conducted with β = 0.01 and the other settings are the same as
the above. As seen from Fig. 8, for the case with a time-vary-
−2 ing reference trajectory, the FDIA detection system (12) is
still able to detect the FDIA and the value of β is accurately
estimated. Meanwhile, the tracking error of the end-effector of
−4
0 2 4 6 8 10 the controlled manipulator is small, and the magnitudes of the
t (s) steady-state tracking errors are less than 3 × 10−3 m, i.e., 3
mm, which is acceptable for many industrial applications. In
Fig. 5. Data profiles of joint velocities during the remote kinematic control addition, it is verified that the joint limits are not violated. The
of the ABB IRB 140 manipulator under FDIA described by (7) and (8) with above results validate the effectiveness of the proposed
β = 0.01 when the proposed remote control law (11) synthesized by the FDIA method for the remote kinematic control of manipulators with
detection system (12) is employed. time-varying reference outputs under FDIA.

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 ZHANG AND LI: KINEMATIC CONTROL OF SERIAL MANIPULATORS UNDER FALSE DATA INJECTION ATTACK 1017

0.4 100 0.7

(m)
0.2 r1 10−1 β̂ 0.6 This paper
r2
0 10−2 0.5 [16]
r3
−0.2 r̂1
10−3 0.4
r̂2
−0.4 r̂3 10−4 0.3
−0.6 10−5
0 2 4 6 8 10 0 2 4 6 8 10 0.2
t (s) t (s)
(a) (b) 0.1

0.1 0.4 0
(m) 0 20 40 60 80 100
0
0.2 (m) t (s)
−0.1 rX
eX 0 rY
−0.2 eY rZ Fig. 7. Profiles of end-effector error norm ∥e∥2 of the remotely controlled
−0.3 eZ −0.2 rdX
−0.4 rdY manipulator under FDIA described by (7) and (8) with β = 0.01 when the con-
−0.4 rdZ trol law (11) proposed in this paper is adopted and when the control law (25)
−0.5
−0.6 −0.6 proposed in [16] is adopted under the same parameter setting for the same
0 2 4 6 8 10 0 2 4 6 8 10 desired end-effector coordinate rd .
t (s) t (s)
(c) (d)
0.6 100
400 50
(°/s) (°) 0.4 r1
us1 θ1 r2 10−1 β̂
0 0.2
200 us2 θ2 r3
θ3 0 r̂1 10 −2

us3 −50 −0.2 r̂2

0 θ4 10−3
us4 r̂3
−100 θ5 −0.4
us5 θ6 10−4
−200 −0.6
us6 −150 −0.8
0 2 4 6 8 10 0 2 4 6 8 10
−400 −200
0 2 4 6 8 10 0 2 4 6 8 10 t (s) t (s)
t (s) t (s) (a) (b)
(e) (f) 0.6
0.06 (m) eX (m)
0.4
0.04 eY
Fig. 6. Data profiles during the remote kinematic control of the ABB IRB eZ 0.2 rX
0.02 rY
140 manipulator under FDIA described by (7) and (8) with β = 0.001 when 0 0 rZ
−0.02 × 10−3 −0.2 rdX
the proposed remote control law (11) synthesized by the FDIA detection sys- −0.04 2 rdY
−0.4 rdZ
tem (12) is employed. (a) Profiles of end-effector coordinates of the manipu- −0.06 0
−0.08 −0.6
−2
lator r and the state variable r̂ of the FDIA detection system (12); (b) Profile 6 8 10
−0.10 −0.8
of detection variable β̂ of the FDIA detection system (12); (c) Profiles of end- 0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
effector coordinate error e = r − rd ; (d) Profiles of the end-effector coordinate (c) (d)
r of the remotely controlled manipulator and the desired end-effector coordi- 400 0
nate rd ; (e) Profiles of the remote control commands us sent by the remote 300 (°/s) (°)
us1 θ1
control center; (f) Profiles of joint angles θ of the controlled manipulator. 200 us2 −50 θ2
100 us3 θ3
0 us4 −100 θ4
C. Robustness to External Disturbance or Network Transmis- us5 θ5
−100 θ6
us6
sion Error −200 −150
−300
In practice, given that the external disturbance or network −400 −200
transmission error is relatively small, they can be viewed as 0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
part of the FDIA. Let ud (t) denote the sum of external distur- (e) (f)
bance and network transmission error. The actual control
command received by the controlled manipulator becomes Fig. 8. Data profiles during the remote kinematic control of the ABB IRB
140 manipulator under FDIA described by (7) and (8) with β = 0.01 when the
ur (t) = us (t) + ua (t) + ud (t) (26)
proposed remote control law (11) synthesized by the FDIA detection system
which is equivalent to (12) is employed for time-varying reference trajectory rd (t). (a) Profiles of
end-effector coordinates of the manipulator r and the state variable r̂ of the
ur (t) = us (t) + uA (t) (27)
FDIA detection system (12); (b) Profile of detection variable β̂ of the FDIA
where uA (t) = ua (t) + ud (t) can be viewed as the total effects detection system (12); (c) Profiles of end-effector coordinate error e = r − rd ;
caused by external disturbance, network transmission error, (d) Profiles of the end-effector coordinate r of the remotely controlled manip-
and FDIA. Obviously, (27) bears similar structure to the FDIA ulator and the desired end-effector coordinate rd ; (e) Profiles of the remote
model (7). Thus, it can be expected that the proposed method control commands us sent by the remote control center; (f) Profiles of joint
still works given that the elements in the term ud (t) are small. angles θ of the controlled manipulator.

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 1018 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 4, APRIL 2023

To verify this point, we have conducted additional simulation VI. Conclusions

with each element of ud (t) being 0.01 sin(t), and the results are In this paper, we have proposed a novel resilient remote
shown in Fig. 9 . As seen from the figure, under FDIA and kinematic control method for manipulators under FDIA,
small transmission errors, the proposed method still has good which consists of a kinematic control law and a FDIA detec-
performance. tion system. Theoretical analysis has shown that the proposed
Before ending this section, we provide a remark regarding approach can guarantee the exact detection of the FDIA and
the effect of the FDIA considered in this paper. the convergence of the end-effector configuration error of the
Remark 7: While it is difficult to quantitatively evaluate the remotely controlled manipulator in the presence of the FDIA.
An illustrative example has also been presented, which shows
control efficiency decrease caused by the FDIA, the effect of
the efficacy and advantages of the proposed method. It should
the FDIA on the control performance can be significant. As
be pointed out that, to our knowledge, this is the first work
seen from Fig. 7 , due to the FDIA, the convergence of the
that can simultaneously detect and address false data injection
end-effector error is very slow, taking more than 80 seconds. attacks for the control of manipulators. For the case where a
This indicates that the FDIA can cause great damage in the false-date injection is made on high-level control command,
efficiency of the manufacturing. The illustrative example also e.g., the Cartesian pose and velocity, it would be difficult for
shows the concealment of the FDIA. Specifically, under the the remote control center to validate the existence of the attack
FDIA, the task can still be completed by the manipulator, without using public-key cryptography tools. However, it is
although with very low efficiency. widely recognized that the usage of public-key cryptography
tools would cost more time [40], which is not favorable for
real-time control. In the future, we will try to tackle this issue.
0.2 100
(m) β̂
0
eX 10−2 References
−0.2 eY [1] D. Ye and T. Zhang, “Summation detector for false data-injection attack
eZ
10−4 in cyber-physical systems,” IEEE Trans. Cybern., vol. 50, no. 6, pp. 2338–
−0.4 2345, Jun. 2020.
−0.6 10−6 [2] K. Kashima and D. Inoue, “Replay attack detection in control systems
0 5 10 15 20 0 5 10 15 20 with quantized signals,” in Proc. European Control Conf., Linz,
t (s) t (s) Austria, Jul. 2015, pp. 782–787.
(a) (b) [3] Y. Zhao, Z. Chen, C. Zhou, Y. C. Tian, and Y. Qin, “Passivity-based
6 robust control against quantified false data injection attacks in cyber-
(°/s) 0.4 (m) physical systems,” IEEE/CAA J. Autom. Sinica, vol. 8, no. 8, pp. 1440–
4 θ·1 1450, Aug. 2021.
θ·2 0.2 r1
2 θ·3 r2 [4] Y. Dong, N. Gupta, and N. Chopra, “ False data injection attacks in
0 r3
θ·4
bilateral teleoperation systems,” IEEE Trans. Control Syst. Techonol.,
0 r̂1
θ·5 −0.2
r̂2
vol. 28, no. 3, pp. 1168–1176, May 2020.
−2 θ·
6 −0.4 r̂3 [5] G. Liang, J. Zhao, F. Luo, S. R. Weller, and Z. Y. Dong, “A review of
false data injection attacks against modern power systems,” IEEE
−4 −0.6 Trans. Smart Grid, vol. 8, no. 4, pp. 1630–1638, Jul. 2017.
0 5 10 15 20 0 5 10 15 20
t (s) t (s) [6] Z. H. Pang, G. P. Liu, D. Zhou, F. Hou, and D. Sun, “Two-channel false
(c) (d) data injection attacks against output tracking control of networked
systems,” IEEE Trans. Ind. Electron., vol. 63, no. 5, pp. 3242–3251,
50 (°) May 2016.
400 (°/s)
us1 0 θ1 [7] A. Y. Lu and G. H. Yang, “ False data injection attacks against state
200 us2 θ2 estimation in the presence of sensor failures,” Inf. Sci., vol. 508,
us3 −50 θ3 pp. 92–104, Jan. 2020.
0 us4 θ4
us5 −100 θ5 [8] Z. Gu, J. H. Park, D. Yue, Z. G. Wu, and X. Xie, “Event-triggered
us6 θ6 security output feedback control for networked interconnected systems
−200 −150
subject to cyber-attacks,” IEEE Trans. Syst., Man, Cybern., Syst.,
−400 −200 vol. 51, no. 10, pp. 6197–6206, Oct. 2021.
0 5 10 15 20 0 5 10 15 20
[9] H. Modares, B. Kiumarsi, F. L. Lewis, F. Ferrese, and A. Davoudi,
t (s) t (s)
“Resilient and robust synchronization of multiagent systems under
(e) (f) attacks on sensors and actuators,” IEEE Trans. Cybern., vol. 50, no. 3,
pp. 1240–1250, Mar. 2020.
Fig. 9. Data profiles during the remote kinematic control of the ABB IRB [10] W. He, Z. Mo, Q. L. Han, and F. Qian, “ Secure impulsive
140 manipulator under FDIA and transmission error described by (26) with synchronization in Lipschitz-type multi-agent systems subject to
each element of ud (t) being 0.01 sin(t) when the proposed remote control law deception attacks,” IEEE/CAA J. Autom. Sinica, vol. 7, no. 5, pp. 1326–
1334, Sept. 2020.
synthesized by the proposed FDIA detection system is employed. (a) Profiles
[11] W. Fu, J. Qin, Y. Shi, W. X. Zheng, and Y. Kang, “Resilient consensus
of end-effector coordinate error e = r − rd ; (b) Profile of detection variable β̂ of discrete-time complex cyber-physical networks under deception
of the FDIA detection system; (c) Profiles of joint velocity vector of the attacks,” IEEE Trans. Ind. Inform., vol. 16, no. 7, pp. 4868–4877, Jul.
2020.
manipulator; (d) Profiles of end-effector coordinates of the manipulator r and
[12] Y. Yuan, H. Yuan, L. Guo, H. Yang, and S. Sun, “Resilient control of
the state variable r̂ of the FDIA detection system; (e) Profiles of the remote
networked control system under DoS attacks: A unified game
control commands us sent by the remote control center; (f) Profiles of joint approach,” IEEE Trans. Ind. Inform., vol. 12, no. 5, pp. 1786–1794, Oct.
angles θ of the controlled manipulator. 2016.

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.
 ZHANG AND LI: KINEMATIC CONTROL OF SERIAL MANIPULATORS UNDER FALSE DATA INJECTION ATTACK 1019

[13] K. Guo, H. Su, and C. Yang, “ A small opening workspace control enabled network parameter modifications in power systems: Attack and
strategy for redundant manipulator based on RCM method,” IEEE detection,” IEEE Trans. Ind. Inform., DOI: 10.1109/TII.2022.3172688.
Trans. Control Syst. Technol., DOI: 10.1109/TCST.2022.3145645. [32] Q. Zhang, K. Liu, Y. Xia, and A. Ma, “ Optimal stealthy deception
[14] C. Yang, Y. Jiang, W. He, J. Na, Z. Li, and B. Xu, “Adaptive parameter attack against cyber-physical systems,” IEEE Trans. Cybern., vol. 50,
estimation and control design for robot manipulators with finite-time no. 9, pp. 3963–3972, Sept. 2020.
convergence,” IEEE Trans. Ind. Electron., vol. 65, no. 10, pp. 8112– [33] Y. Zhang, S. Li, and X. Jiang, “ Near-optimal control without solving
8123, Oct. 2018. HJB equations and its applications,” IEEE Trans. Ind. Electron., vol. 65,
[15] D. Li, W. Zhang, W. He, C. Li, and S. S. Ge, “ Two-layer distributed no. 9, pp. 7173–7184, Sept. 2018.
formation-containment control of multiple Euler-Lagrange systems by [34] Y. Lin, Z. Chen, and B. Yao, “Decoupled torque control of series elastic
output feedback,” IEEE Trans. Cybern., vol. 49, no. 2, pp. 675–687, actuator with adaptive robust compensation of time-varying load-side
2019. dynamics,” IEEE Trans. Ind. Electron., vol. 67, no. 7, pp. 5604–5614,
[16] S. Li, Y. Zhang, and L. Jin, “ Kinematic control of redundant Jul. 2020.
manipulators using neural networks,” IEEE Trans. Neural Netw. Learn. [35] L. Chen, R. Cui, C. Yang, and W. Yan, “ Adaptive neural network
Syst., vol. 28, no. 10, pp. 2243–2254, Oct. 2017.
control of underactuated surface vessels with guaranteed transient
[17] Y. Zhang, S. Li, J. Zou, and A. H. Khan, “A passivity-based approach performance: Theory and experimental results,” IEEE Trans. Ind.
for kinematic control of manipulators with constraints,” IEEE Trans. Electron., vol. 67, no. 5, pp. 4024–4035, May 2020.
Ind. Inform., vol. 16, no. 5, pp. 3029–3038, May 2020.
[36] J. Dattorro, Convex Optimization And Euclidean Distance Geometry.
[18] L. Jin, S. Li, J. Yu, and J. He, “Robot manipulator control using neural California, USA: Meboo Publishing, 2016.
networks: A survey,” Neurocomputing, vol. 285, pp. 23–34, Apr. 2018.
[37] H. K. Khalil, Nonlinear Systems, 3rd ed. Englewood Cliffs, NJ, USA:
[19] D. Guo, F. Xu, and L. Yan, “ New pseudoinverse-based path-planning Prentice-Hall, 2001.
scheme with PID characteristic for redundant robot manipulators in the
[38] Y. Zhang, S. Li, and X. Liu, “ Adaptive near-optimal control of
presence of noise,” IEEE Trans. Control Syst. Technol., vol. 26, no. 6,
uncertain systems with application to underactuated surface vessels,”
pp. 2008–2019, Nov. 2018.
IEEE Trans. Control Syst. Techonol., vol. 26, no. 4, pp. 1204–1218, Jul.
[20] L. Jin, S. Li, H. M. La, and X. Luo, “ Manipulability optimization of 2018.
redundant manipulators using dynamic neural networks,” IEEE Trans.
[39] Y. Zhang, S. Li, and X. Zhou, “Recurrent-neural-network-based
Ind. Electron., vol. 64, no. 6, pp. 4710–4720, Jun. 2017.
velocity-level redundancy resolution for manipulators subject to a joint
[21] Y. Zhang, “ Tri-projection neural network for redundant manipulators,” acceleration limit,” IEEE Trans. Ind. Electron., vol. 66, no. 5, pp. 3573–
IEEE Trans. Circuits Syst. II Exp. Briefs, DOI: 10.1109/TCSII.2022. 3582, May 2019.
3189664.
[40] O. Goldreich, Foundations of Cryptography: Volume 2, Basic
[22] F. Maggi, A. M. Zanchettin, and S. Zanero, “An experimental security Applications. New York, NY, USA: Cambridge Univ. Press, 2009.
analysis of an industrial robot controller,” in Proc. IEEE Symp. Security
Privacy, Davide, Quarta, 2017, pp. 268–285.
[23] M. El-hajj, A. Fadlallah, M. Chamoun, and A. Serhrouchni, “A survey Yinyan Zhang received the B.E. degree in automa-
of internet of things (IoT) authentication schemes,” Sensors , vol. 19, tion from Sun Yat-sen University in 2016, and the
no. 5, p. 1141, 2019. Ph.D. degree in computing from The Hong Kong
Polytechnic University, Hong Kong, China, in 2019.
[24] Y. Zhang, S. Chen, S. Li, and Z. Zhang, “ Adaptive projection neural He is currently with the College of Cyber Secu-
network for kinematic control of redundant manipulators with unknown rity, Jinan University. Before joining Jinan Univer-
physical parameters,” IEEE Trans. Ind. Electron., vol. 65, no. 6, sity in 2019, he was a Postdoctoral Fellow with the
pp. 4909–4920, Jun. 2018. Department of Computing, The Hong Kong Poly-
[25] M. W. Spong, S. Hutchinson, and M. Vidyasagar, Robot Modeling and technic University, Hong Kong, China. His current
Control, vol. 3. New York, NY, USA: Wiley, 2006. research interests include security and control of
[26] S. Li, J. He, Y. Li, and M. U. Rafique, “ Distributed recurrent neural multi-agent systems, intelligent control, and dynamic neural networks.
networks for cooperative control of manipulators: A game-theoretic Dr. Zhang serves as an Associate Editor for the Neural Processing Let-
perspective,” IEEE Trans. Neuarl Netw. Learn. Syst., vol. 28, no. 2, ters, and an Editorial Board Member for the Automatic Control and Com-
pp. 415–426, Feb. 2017. puter Sciences.
[27] Y. Zhang, J. Wang, and Y. Xia, “A dual neural network for redundancy
resolution of kinematically redundant manipulators subject to joint Shuai Li received the B.E. degree in precision
limits and joint velocity limits,” IEEE Trans. Neural Netw., vol. 14, mechanical engineering from Hefei University of
no. 3, pp. 658–667, May 2003. Technology in 2005, the M.E. degree in automatic
[28] S. Yoo, “ Neural-network-based adaptive resilient dynamic surface control engineering from University of Science and
control against unknown deception attacks of uncertain nonlinear time- Technology of China in 2008, and the Ph.D. degree
delay cyberphysical systems,” IEEE Trans. Neural Netw. Learn. Syst., in electrical and computer engineering from Stevens
vol. 31, no. 10, pp. 4341–4353, Oct. 2020. Institute of Technology, USA, in 2014.
[29] S. Song, J. H. Park, B. Zhang, and X. Song, “ Event-based adaptive He is currently an Associate Professor (Reader)
fuzzy fixed-time secure control for nonlinear CPSs against unknown at Swansea University, UK, leading the Robotic Lab,
false data injection and backlash-like hysteresis,” IEEE Trans. Fuzyy conducting research on robot manipulation and
Syst., vol. 30, no. 6, pp. 1939–1951, Jun. 2022. impedance control, multi-robot coordination, distributed control, intelligent
optimization and control, and legged robots.
[30] Q. Zhang, K. Liu, D. Han, G. Su, and Y. Xia, “ Design of stealthy Dr. Li is the founding Editor-in-Chief of the International Journal of
deception attacks with partial system knowledge,” IEEE Trans. Autom. Robotics and Control, an Associate Editor of the IEEE Transactions on Cir-
Control, DOI: 10.1109/TAC.2022.3146079. cuits and Systems II: Express Briefs, and the General Co-Chair of the 2018
[31] C. Liu, W. He, R. Deng, Y. Tian, and W. Du, “ False data injection International Conference on Advanced Robotics and Intelligent Control.

Authorized licensed use limited to: King Fahd University of Petroleum and Minerals. Downloaded on May 22,2024 at 05:56:59 UTC from IEEE Xplore. Restrictions apply.