Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

1234 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO.

5, MAY 2023

Resilient and Safe Platooning Control of Connected


Automated Vehicles Against Intermittent
Denial-of-Service Attacks
Xiaohua Ge, Senior Member, IEEE, Qing-Long Han, Fellow, IEEE, Qing Wu, Member, IEEE, and
Xian-Ming Zhang, Senior Member, IEEE

Abstract— Connected automated vehicles (CAVs) serve as a the automotive industry in recent years. CAVs can potentially
promising enabler for future intelligent transportation systems improve the traffic throughout and energy/fuel consumption,
because of their capabilities in improving traffic efficiency and
and reduce the risk of misoperation accidents by human
driving safety, and reducing fuel consumption and vehicle emis-
sions. A fundamental issue in CAVs is platooning control that drivers, thereby leading to great economical, environmental
empowers a convoy of CAVs to be cooperatively maneuvered and social benefits. This has been well demonstrated and testi-
with desired longitudinal spacings and identical velocities on fied by the numerous successfully implemented CAV pro-
roads. This paper addresses the issue of resilient and safe pla- grams and challenges, such as the PATH program [1], the
tooning control of CAVs subject to intermittent denial-of-service Energy-ITS program [2], the Demo 2000 program [3], the
(DoS) attacks that disrupt vehicle-to-vehicle communications.
First, a heterogeneous and uncertain vehicle longitudinal dynamic SARTRE program [4], and the Grand Cooperative Driving
model is presented to accommodate a variety of uncertainties, Challenges in 2011 [5] and in 2016 [6].
including diverse vehicle masses and engine inertial delays, CAVs have reinvigorated the relevant fields of vehicle
unknown and nonlinear resistance forces, and a dynamic platoon dynamics and control [7], AI-based learning [8]–[10], traffic
leader. Then, a resilient and safe distributed longitudinal platoon- flow [11], route planning [12], [13], wireless communication
ing control law is constructed with an aim to preserve simultane- [14], [15], multi-agent systems [16], cyber security [17], [18]
ous individual vehicle stability, attack resilience, platoon safety
and scalability. Furthermore, a numerically efficient offline design and so forth. Among the many fundamental issues in CAVs,
algorithm for determining the desired platoon control law is vehicle platooning control has attracted intensive attention.
developed, under which the platoon resilience against DoS attacks More specifically, vehicle platooning aims to guarantee a
can be maximized but the anticipated stability, safety and scala- string formation of a maneuvered vehicular platoon in terms
bility requirements remain preserved. Finally, extensive numeri- of some desired (fixed or time-varying) longitudinal spacings
cal experiments are provided to substantiate the efficacy of the and same velocities (and accelerations).
proposed platooning method.
In order to fulfil a platooning control objective, a suitable
Index Terms—Connected automated vehicles (CAVs), cooperative distributed cooperative longitudinal control strategy is deman-
adaptive cruise control, denial-of-service (DoS) attacks, resilient con-
trol, vehicle platooning, vehicle-to-vehicle communication. ded to coordinate the vehicular platoon to further form and
maintain the desired longitudinal gaps. This, however, largely
I. Introduction relies on the available sensory technologies equipped by each
ONNECTED automated vehicles (CAVs) have emerged vehicle and the levels of inter-vehicle connectivity and intra-
C as one promising intelligent transportation technology
and have gained significant popularity in both academia and
vehicle automation. For example, by virtue of only on-board
sensors (e.g., radar and scanning lidar), adaptive cruise con-
Manuscript received March 30, 2022; revised April 17, 2022; accepted
trol (ACC) has been widely adopted at the early stage to
April 30, 2022. This work was supported in part by Australian Research achieve a longitudinal vehicle-following control objective of
Council Discovery Early Career Researcher Award (DE210100273). automated vehicles. Nevertheless, an ACC-based platooning
Recommended by Associate Editor Zidong Wang. (Corresponding author:
Qing-Long Han.) strategy often results in relatively large spacings between
Citation: X. Ge, Q.-L. Han, Q. Wu, and X.-M. Zhang, “Resilient and safe adjacent vehicles, which degrades traffic throughout and
platooning control of connected automated vehicles against intermittent makes little contribution to fuel efficiency. Another limitation
denial-of-service attacks,” IEEE/CAA J. Autom. Sinica, vol. 10, no. 5, pp.
1234–1251, May 2023. of ACC-based vehicular platoons is the lack of a string stabil-
X. Ge, Q.-L. Han, and X.-M. Zhang are with the School of Science, ity guarantee, which is a system-level stability requirement for
Computing and Engineering Technologies, Swinburne University of preventing inter-vehicle collisions. On the other hand, owe to
Technology, Melbourne VIC 3122, Australia (e-mail: xge@swin.edu.au;
qhan@swin.edu.au; xianmingzhang@swin.edu.au). the mature IEEE 802.11p-based dedicated short range com-
Q. Wu is with the School of Engineering and Technology, Central munication technology and the emerging 5G cellular commu-
Queensland University, Rockhampton QLD 4701, Australia (e-mail: q.wu@ nication technology, cooperative adaptive cruise control
cqu.edu.au).
(CACC) has been prevalently exploited nowadays as it
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. enables each vehicle to possess certain wireless communica-
Digital Object Identifier 10.1109/JAS.2022.105845 tion capability. More precisely, CACC-enabled CAVs can not

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1235

only be driven under partial or full automation using informa- excessively high, especially when the vehicle platoon system
tion from on-board sensory devices, but also act coopera- involves a large number of such secure components.
tively and responsively via vehicle-to-everything (V2X) com- Detection-based methods aim to detect the occurrence of a
munications. Depending on platooning scenarios and applica- particular type of attacks such as false data injection attacks.
tions, the V2X communications typically include vehicle-to- From the physical protection viewpoint, CAVs may also
vehicle (V2V, i.e., communication among adjacent vehicles) deploy additional secure sensors for detecting abnormal activ-
communications, vehicle-to-infrastructure (V2I, e.g., road- ities. From the perspective of system estimation and control,
side units) communications and vehicle-to-pedestrian (V2P) however, the focus is placed on designing a suitable state esti-
communications. Compared with an ACC-based platooning mator or observer on each CAV, based on either Kalman fil-
control strategy, a CACC-based distributed cooperative longi- tering theory, set-evaluated theory, adaptive estimation theory,
tudinal platooning control strategy can guarantee the desired or neural networks, to estimate the corrupted vehicle dynami-
platoon performance under smaller spacings without compro- cal state in real-time and further set out a warning alarm when
mising safety due to the additional wireless V2V communica- attacks are detected. For example, in [21], a real-time sensor
tions beyond line-of-sight of onboard sensors. anomaly detection method, combining a convolutional neural
Albeit V2V communications allow CAVs to learn about the network, a Kalman filter, and a well-established χ2-anomaly
platoon dynamics and behaviors of their adjacent vehicles, the detector, is presented to detect and identify unusual CAV
introduction of a wireless network makes a vehicular platoon behaviors caused by either malicious cyber attacks or faulty
system more exposed to external malicious actors. In other vehicle sensors. In [22], an observer-based method, which
words, adversaries may viciously hack into, interfere with or leverages an adaptive extended Kalman filter and a previ-
interrupt the wireless network in order to disrupt the desired ously-trained one class support vector machine model, is
platooning control objective. For example, the broadcast data developed to improve the anomaly detection performance in
of interest between neighboring CAVs may be blocked from CAV sensors. In [23], the DoS attacks are modeled as a time
reaching their destinations under denial-of-service (DoS) delay in information processed by the network, and a set of
attacks, or modified by maliciously altering their contents and adaptive estimators and sliding mode observers are designed
inserting falsified data under false data injection attacks [19]. to estimate the delay. The DoS occurrences are then identi-
In the presence of malicious attacks, the traditional platoon- fied via comparing the delay and some predefined threshold at
ing control methods without incorporating security countermea- each time instant. In [24], a set-membership prediction and
sures thus need to be re-evaluated, and novel platooning con- estimation method is proposed to detect the occurrence of
trol strategies are highly desirable to ensure security and cyber attacks on both sensor measurements and control com-
safety for CAVs against attacks. mands of CAVs. Two recovery mechanisms are presented to
restore the attacked CAVs to normal operations.
A. Literature Review Resilience-based methods typically rely on some security-
The existing security methods for safeguarding CAV pla- inherited controllers or control laws that are robust and
toons can be arguably classified into three types: 1) resilient to the effects of attacks on the closed-loop system sta-
prevention-based methods, 2) detection-based methods, and 3) bility and performance. Here, resilience characterizes the sur-
resilience-based methods. vivability of the platoon control system to remain in operation
Prevention-based methods often leverage information assur- even in the presence of malicious attacks on CAVs. It is clear
ance techniques (e.g., advanced cryptographic algorithms, that detection-based methods only provide the first step
data authentication, access control) to prevent adversaries towards secure monitoring of vehicle platoon systems, while
from getting access to the communication network and perfor- resilience-based methods deliver a further control goal of pre-
ming malicious actions. For example, in [20], a secure IEEE serving the security requirement of vehicle platoon systems.
802.11p and VLC-based hybrid communication protocol, fea- Furthermore, most of the existing literature on resilient pla-
turing secret key establishment, message authentication and tooning control concerns two types of attacks on CAVs: DoS
jamming detection and reaction, is proposed for ensuring pla- attacks (or jamming attacks) and deception attacks. Specifi-
toon stability against data packet injection, channel overhea- cally, DoS represents a common cyber attack strategy that
ring, jamming, and platoon maneuver attacks. It should be makes the data transmissions unavailable or inaccessible at the
noted that the prevention-based methods do not exploit the receiver’s side, namely, disrupting the data availability. DoS
compatibility of data with the vehicle dynamics and thus may attacks on CAVs are typically accomplished by interfering
be ineffective against malicious attacks on vehicular system with the radio frequency of a vehicular communication net-
components. A comprehensive review of this type of security work or flooding the V2V network access with superfluous
methods for CAVs is beyond the scope of this study. On the requests in an attempt to overuse the network resources and
other hand, adopting secure system components like smart prevent legitimate vehicles from being connected. On the
sensors and actuators that can report unusual actions also other hand, deception attacks generally jeopardize the data
offers certain level of prevention against external attacks. integrity or trustworthiness. Typical deception strategies
Whereas, the cost of physically protecting the system compo- include false data injection, replaying, and spoofing. For
nents or employing advanced secure components can be example, in [18], both false data injection attacks against on-

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1236 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

board range sensors and message delay attacks against data vation of this study.
transmissions in the dedicated short range communication While the platoon controller design of CACC-enabled
protocol are considered for platooning CAVs. The reachabil- CAVs has been widely investigated, the majority of the exist-
ity analysis method is then applied to evaluate the impact of ing design methods assume homogeneous vehicular platoons
two types of attacks on the safety performance of both ACC- in terms of same masses and engine inertial delays (a.k.a.,
based and CACC-based platoons. In [25], the impact of actuator lags, power-train constants) [18], [26], [27], [31]–[34]
stochastic jamming attacks on vehicular CACC systems is and/or linearized longitudinal vehicle dynamics [18], [25]–[30],
studied, where the jammer’s destructive signal is modeled as [32]–[34]. However, in practical platooning scenarios, auto-
an additive Gaussian noise. A sufficient condition on success- mated vehicles are essentially heterogeneous (e.g., different
ful packet delivery probability under which the mean string size and weight). The vehicle mass and power-train dynamics
stability of the platoon system remains stable is derived. In can significantly affect the vehicle driving performance such
[26], a secure adaptive synchronization-based control method as fuel cost. Furthermore, the dynamic behavior of a vehicle
is developed for cooperative driving of CAVs subject to com- longitudinal model is considerably complicated and nonlinear,
munication delays and cyber attacks. It is shown that the involving engine, torque converter, transmission, drive-train,
throttle and brake actuators, external disturbances (e.g., wind
designed platoon controllers can be used to resist several dif-
gust), and so on, which leads to inherently uncertain and non-
ferent types of attacks, including spoofing attacks on vehicu-
linear longitudinal vehicle dynamics. As a matter of fact,
lar accelerations, message falsification attacks on vehicular
cooperatively driving a platoon of vehicles has to overcome
positions and speeds, DoS attacks and burst transmission
the road friction, gravity and longitudinal aerodynamic drag
attacks on data transmissions over the network. In [27], a
forces. In this sense, the existing platooning control methods
switching distributed secure platoon control technique is assuming homogeneous and/or linearized longitudinal vehicle
devised for CAVs subject to DoS attacks. By modeling the dynamics may need to be re-assessed, which gives rise to the
DoS attacks as a series of switching behaviors at sampling second motivation of this study.
instants, the resultant platoon control system is modeled as a Vehicle platooning control is subject to requirements related
switched time-delay system. Similarly, a switching distri- to safety and scalability [35]. Specifically, a longitudinal vehi-
buted event-triggered controller is designed in [28] for vehi- cle control system should exhibit stable platoon behaviors,
cle platooning under a class of periodic jamming attacks, and including individual vehicle stability (achieving stable spac-
a CACC-ACC switching control method is exploited in [29], ing error and velocity/acceleration tracking error systems) and
[30] for CAVs subject to inter-vehicle communication inter- string stability (avoiding the amplification of disturbance
ruptions/losses. In [31], a neural adaptive secure distributed effects along the platoon), to prevent collisions. The defini-
platooning control approach is developed to cope with a class tions of string stability are relatively diverse and ambiguous in
of intermittent DoS attacks, wherein each platoon vehicle the literature. Interested readers are referred to the surveys
equips a logical data packet processor to verify the time- [36], [37] in detail. However, it is noteworthy that some
stamps of the received data packets and further identify the vehicular platoons achieve one form of string stability but
DoS attack occurrence. In [32], by modeling DoS attacks as may remain unstable in another sense [36]. This thus calls for
consecutive packet dropouts, a hybrid platoon controller, com- an intrinsically safety-guaranteed or collision-free platoo-
bining a continuous-time proportional-derivative regulator and ning control method. On the other hand, scalability represents
a zero-order-hold device, is proposed to guarantee required a key measure for evaluating applicability of a vehicle pla-
platoon string stability and performance while maximizing tooning control method in cases of variable platoon lengths
DoS resiliency. and dynamic maneuvers such as splitting and merging. Never-
theless, many existing platooning control methods are
B. Motivations
restricted to length-fixed vehicle platoons and/or specific V2V
Most resilient platooning control methods aforementioned communication topologies [18], [25], [28]–[30], [32] and [35].
for tolerating the effects of DoS attacks rely on certain This indicates that the platoon controllers therein might need
assumptions on the DoS attacks. For example, the DoS or to be re-designed in an attempt to ensure the desired stability
jamming attack signal in [25] is assumed to abide by a Gaus- and performance requirements for the new platoon due to
sian distribution noise of a priori statistical characteristics length or topology changes. This aspect has not been adequa-
(e.g., given mean and variance), the DoS attacks in [26] are tely addressed in the literature, but is of utmost importance in
assumed to cause a delay effect and then lumped together as a order to empower flexible and dynamic platoon maneuvers.
communication delay in data transmissions, the DoS attacks in How to develop a scalable collision-free platooning control
[27], [29] and [30] are required to be known (detected) in method for CAVs is the third motivation of this study.
advance such that the platoon controller can switch between
different control laws/strategies, and the DoS attacks in [28] C. Contributions
are assumed to occur periodically within some known (or Based on the observations above, in this paper, we investi-
active) time intervals. Hence, it is desirable to establish a gen- gate a distributed cooperative platooning control problem for
eral and realistic DoS attack model that requires a priori attack CAVs in the presence of intermittent DoS attacks and uncer-
knowledge as less as possible, which serves as the first moti- tain vehicle dynamics. The novelty of this study lies in the

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1237

development of a resilient and safe distributed longitudinal D. Outline of the Paper


platooning control method. To the best of the author’s knowl- The remainder of this paper is organized as follows. The
edge, this study is among the first few attempts to guarantee preliminaries and the problem formulation are presented in
simultaneously the platoon stability as well as resilience, Section II. The main results in terms of a design algorithm and
safety and scalability requirements of CAVs subject to DoS rigorous stability analysis are stated in Section III. The effec-
attacks. The main contributions of this paper are summarized tiveness of the proposed platooning control method is vali-
as follows. dated through several numerical experiments in Section IV.
1) A heterogeneous and uncertain longitudinal vehicle Section V concludes this paper.
dynamic model is established. The model accounts for the
simultaneous presence of heterogeneous vehicle parameters II. Preliminaries and Problem Formulation
such as unknown vehicle masses and engine inertial delays,
uncertain and nonlinear resistance forces, and a dynamic pla- A. Notations
toon leader. R denotes the set of real numbers. R≥0 denotes the set of
2) A general DoS attack model is presented, which incorpo- non-negative real numbers. Z≥0 denotes the set of non-nega-
rates the intermittent nature of realistic DoS actions but tive integers. The transpose of a matrix or vector Φ is denoted
requires less a priori knowledge of the DoS attacks. In particu- by ΦT . If a matrix is invertible, the superscript “ −1” repre-
lar, the model constrains only the attacker actions in terms of sents the matrix inverse. For a matrix Φ, its minimum and
the attack duration without requiring specific statistical and maximum singular values are denoted by σ(Φ) and σ(Φ) ,
accurate knowledge of DoS attacks. Furthermore, the dura- respectively. For a real, positive definite, and symmetric
tion bound of intermittent DoS attacks is not required to be matrix Φ, its minimum and maximum eigenvalues are denoted
precisely known. On one hand, a data packet analyser that can by λ(Φ) and λ(Φ), respectively. I (correspondingly, 0) repre-
be employed by each platoon vehicle is devised to identify the sents the identity (correspondingly, zero) vector or matrix of
occurrence of DoS attacks and further record the maximal an appropriate size. diag{Φ1 , Φ2 , . . . , Φm } is used to denote an
DoS duration over a certain time window. On the other hand, m-block diagonal matrix with Φi , ∀ i = 1, 2, . . . , m , being the i-
a theoretical condition for estimating the maximal allowable th diagonal element. ∥ · ∥ denotes the Euclidean norm of a vec-
DoS duration is also derived. The concerned attack model can tor. ⊗ denotes the Kronecker product. Matrices, if not explic-
be used to accommodate many different types of DoS attacks itly stated, are assumed to have compatible dimensions.
including the wide-studied periodic DoS attacks, aperiodic
B. Uncertain Heterogeneous Longitudinal Vehicle Dynamics
DoS attacks, and random DoS attacks.
3) A resilient and safe distributed longitudinal platooning Consider a group of N + 1 automated vehicles being maneu-
control law is developed. The proposed platoon controllers are vered by using a cooperative longitudinal control strategy for
capable to withstand the adverse effects of intermittent DoS a resilient and safe vehicle-following objective on a sloping
attacks and adaptively slow down each platooning vehicle road, as shown in Fig. 1 (a). The vehicle, indexed by 0, is
denoted as the leader and the other vehicles, indexed by
once it gets close to its direct predecessor to avoid absolute
i ∈ V = {1, 2, . . . , N}, are assigned as the followers. Each vehi-
collisions. Another prominent feature of the proposed platoon-
cle is modeled as a single point mass, and pi (t), vi (t) , ai (t), and
ing control method is its scalability. It will be shown that the
ȧi (t) ∈ R are used to denote the longitudinal position, velocity,
distributed longitudinal control law can be implemented by
acceleration, and jerk of vehicle i at time t ∈ R≥0, respectively.
each platoon vehicle in a distributed fashion, i.e., each vehicle
It is assumed that each platoon vehicle is equipped with 1) an
only requires information recorded by its on-board sensors on-board radar that can measure the gap distance between any
and information gathered from its underlying neighbors speci- two consecutive vehicles for collision detection, 2) a built-in
fied by the V2V communication topology. It will be also illus- GPS sensor for measuring the vehicular longitudinal position
trated that the platooning control method is flexible with information, 3) a wireless transceiver for sharing information
respect to V2V communication topologies (or information of interest with its proximal vehicles via V2V communica-
flow topologies), thus eliminating the limitation to the spe- tions, and 4) an engine control unit for calculating the desired
cific predecessor-following topology in [18], [25], [28]–[30], engine force/torque. Such an assumption is reasonable as the
[32], [35]. sensing, communication and actuation units requested above
4) A tractable offline design algorithm on the simultaneous are available in modern CAVs, and thus does not require spe-
existence of the desired local state observers and platoon con- cific changes in the existing vehicle configuration.
trollers is provided. Furthermore, the boundedness of the A vehicle longitudinal control system mainly consists of
resultant platoon tracking errors and spacing errors against engine, throttle and brake actuators, drive train, transmission,
various uncertainties and DoS attacks is formally proved. The and torque converter. Under a variety of resistance forces, as
design algorithm is scalable over various V2V communica- illustrated in Fig. 1(b), the longitudinal dynamics of each pla-
tion topologies in the sense that even in the event of vehicle toon vehicle i can be modeled by the following force balance
length variations the already designed platoon controller equation:
parameters may be re-used for the new platoon. It will be also
mi ai (t) = fie (t) − Fire (t), t ∈ R≥0 (1)
demonstrated that the design algorithm can be used to achieve
a comprehensive trade-off analysis between desired platoon where mi stands for the mass of vehicle i; fie (t)
is the desired
g
performance and attack severity. engine force acting on vehicle i ; and Fire (t) = fi (t) + fiw (t)+

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1238 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

pi (t),
vi (t), where
fi w(t) ai (t)
fi e(t) xi (t) = [pi (t), vi (t), ai (t)]T , ∆i (t) = ∆i (vi , ai , ȧi )
fi g(t)
mi g
0 1 0 
  [ ]T
DoS (b) θi(t) fi r(t) 0 0 1  , B̌ = 0, 0, 1 .
Ǎ =  
 τ0
(t, x̂0(t)) 0 0 − τ10
Radar (t, x̂1(t))
(t, x̂2(t)) Due to the digital nature of wireless V2V communications,
the vehicular data are required to be sampled in a discretized
manner and then shared over the wireless V2V channels in a
DoS
packet-based manner. Besides, virtually all control algorithms
Attacker are implemented on digital modules, which means that pla-
(a)
toon controllers operate only at discrete instants in time. For
this purpose, under the sampling period h > 0, the discrete-
Fig. 1. Platooning control and dynamics of CAVs: (a) The schematic of time version of the longitudinal dynamics of (6) is given as
platooning control in the presence of DoS attacks; (b) The main longitudinal
xi (k + 1) = Axi (k) + B(ui (k) + ∆i (k)), k ∈ Z≥0 (7)
forces acting on a CAV.
where xi (k) = [pi (k), vi (k), ai (k)]T denotesr the vehicular state at
fir (t) denotes the lumped uncertain resistance force on vehicle h
time step k, A = exp(Ǎh), and B = 0 exp(Ǎs)ds B̌. It is
g
i. Specifically, fi (t) = mi g sin(θi (t)) represents the gravity
assumed that a dynamic platoon leader is present, which can
component parallel to the road surface with g = 9.81 m/s2 be a substantial human-driven vehicle or a fictitious leader
denoting the acceleration of gravity and differentiable θi (t) acting as some reference for the platoon. Specifically, the pla-
representing the inclination angle of the road; fiw (t) is the air toon leader evolves according to
resistance force satisfying fiw (t) = (1/2)ρC D AF (vi (t) + vw (t))2
with ρ being the air density; AF being the maximal cross-sectio- x0 (k + 1) = Ax0 (k) + Bu0 (k), k ∈ Z≥0 (8)
nal/frontal area of the vehicle; C D being the aerodynamic drag where x0 (k) = [p0 (k), v0 (k), a0 (k)] denotes the leader state at
T

coefficient; and vw (t) being the uncertain headwind speed; and time step k and u0 (k) denotes an unknown input that serves as
fir (t) = µR mi g cos(θi (t)) denotes rolling resistance force with the reference acceleration commands for the platoon.
µR being the coefficient of rolling resistance.
C. A Vehicular Position-Based Local State Observer
By modeling the engine inertial lag as a first-order filter, the
engine force fie (t) can be given by Different from the existing state-based platoon control
approaches which require the full state information xi (k) of
τi f˙ie (t) + fie (t) = uei (t), t ∈ R≥0 (2) each vehicle for platoon control and maneuver, we are inter-
with being the actual engine throttle/brake input and τi
uei (t) esting in constructing and designing some local observers for
denoting the unknown inertial lag. the follower vehicles based on only the available position
Substituting fiw (t) into (1) yields that measurements. More specifically, for each follower vehicle i,
the real-time noisy position measurement is given as
1
fie (t) = mi g sin(θi (t)) + ρC D AF (vi (t) + vw (t))2
2 yi (k) = pi (k) + ϑi (k) = Cxi (k) + ϑi (k), ∀i ∈ V (9)
+ µR mi g cos(θi (t)) + mi ai (t). (3) where yi (k) ∈ R denotes the measurement output recorded by
Note that τi can be rewritten as τi = τ0 + τ̃i with τ0 denoting the built-in GPS unit of each platoon member, C = [1, 0, 0],
the nominal part and τ̃i denoting the uncertain part. Then, cal- and ϑi (k) ∈ R denotes the unknown measurement noise.
culating the time derivative of (1) yields that To acquire a real-time and accurate estimate of the unavail-
able full state xi (k) , each follower vehicle i runs the following
τ0 ȧi (t) + ai (t) = ui (t) + ∆i (vi , ai , ȧi ), t ∈ R≥0 (4) local state observer based on the noisy position measurement:
where ui (t) = uei (t)/mi denotes the control input and 


 x̂i (k + 1) = A x̂i (k) + B(ui (k) + ∆ˆ i (k))
τi ρC D AF 


∆i (vi , ai , ȧi ) = − (vi (t) + vw (t))(ai (t) + v̇w (t)) 
 + L(C x̂i (k) − yi (k)) (10)
mi 


ˆ
ρC D AF ∆i (k + 1) = ∆ˆ i (k) + F(C x̂i (k) − yi (k))
− (vi (t) + vw (t))2 − g sin(θi (t)) for any k ∈ Z≥0, where x̂i (k) = [ p̂i (k), v̂i (k), âi (k)]T is the vehic-
2mi
ular state estimate; ∆ˆ i (k) denotes an estimate of the unknown
− τi g cos(θi (t))θ̇i (t) − µR g cos(θi (t))
input signal ∆i (k); ui (k) denotes the desired control input sig-
+ τi µR g sin(θi (t))θ̇i (t) − τ̃i ȧi (t) (5) nal to be specified later; L and F are the observer gain matri-
represents the lumped uncertainties resulting from various ces to be designed.
resistance forces and uncertain dynamics. The above longitu-
D. V2V Communication
dinal dynamics can be rewritten in a compact form of
In order achieve the cooperative longitudinal platooning
ẋi (t) = Ǎxi (t) + B̌(ui (t) + ∆i (t)), t ∈ R≥0 (6) control objective, each platoon member i employs a wireless

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1239

transceiver, allowing its local vehicle information to be shared Denote the relative distance between vehicle i and its prede-
with its underlying neighbors in accordance with certain V2V cessor i − 1 at time t by
communication topology over the network. Ri (k) = pi−1 (k) − pi (k), ∀ i ∈ V (11)
By regarding each follower vehicle as a node in a graph, the
V2V communication topology of the N-vehicle platoon can be which can be obtained in real time via an on-board radar
modeled by a weighted graph G = (V, E, A) , where V = {1, mounted in front of each vehicle. We next introduce the fol-
2, . . . , N} is a node set, E ⊆ V × V is an edge set of paired lowing collision avoidance function at vehicle i.
nodes with ( j, i) representing a directional edge from node j to Definition 1: The collision avoidance function uoa i (Ri (k)) is
node i , and A = [ai j ] ∈ RN×N is a weighted adjacency matrix a piecewise continuous function of Ri (k) satisfying the follow-
ing properties:
with ai j ≥ 0 being the adjacency element of the edge ( j, i) . It is
assumed that self-loops are not contained in the graph, i.e., i (Ri (k)) → ∞ if Ri (k) → 0;
1) uoa
aii = 0 . The graph G is directed if A is a non-symmetric i (Ri (k)) = 0 if Ri (k) > S 0; and
2) uoa
matrix with ai j , a ji and undirected if A is a symmetric i (Ri (k)) is bounded if Ri (k) ≤ S 0;
3) uoa
matrix with ai j = a ji for all i, j ∈ V . The incoming neighbors where S 0 denotes the prescribed safe distance threshold.
Let Ωi (pi ) = {pi (k)| Ri (k) ≤ S 0 } be a compact set that speci-
of node i is denoted by Niin = { j ∈ V : ( j, i) ∈ E} and the outgo-
fies the collision-risky zone for each vehicle i moving towards
ing neighbors of node i is denoted by Niout = { j ∈ V :(i, j) ∈ E}.
its predecessor i − 1. Then it is clear from Definition 1 that the
The Laplacian matrix associated with G is defined as
∑ ∑ ∑ collision avoidance function uoa i (Ri (k)) will not be activated if
L = D − A with D = diag{ Nj=1 a1 j , Nj=1 a2 j , . . . , Nj=1 aN j }.
vehicle i is far away from its predecessor, namely, pi (k) <
Let the leader vehicle be an additional node 0. Whether or Ωi (pi ). However, as vehicle i enters the collision-risky zone,
not a follower (node i) can receive information from the leader the collision avoidance function uoa i (Ri (k)) comes into play
(node 0) is specified by a nonnegative element ai0 ≥ 0 for any and pushes vehicle i away when vehicle i is getting close to its
j ∈ V ∪ {0}. Denote A0 = diag{a10 , a20 , . . . , aN0 }. For simplic- predecessor from S 0.
ity but without loss of generality, it is assumed that the matrix
H = L + A0 is nonsingular. Then, there exists an invertible F. Model of Intermittent DoS Attacks
matrix U such that H can be factorized as U −1 HU = Λ, For each platoon vehicle i ∈ V , the vehicular state estimate
where Λ is the diagonal matrix whose diagonal elements are x̂i (k) is encapsulated into a data packet with its time-stamp k,
the corresponding eigenvalues of H , i.e., Λii = λi for any namely, (k, x̂i (k)) . Furthermore, the encapsulated vehicular
i ∈ V . Such an assumption is mild and can accommodate sev- data packets are broadcast and transmitted to their neighbors
eral typical V2V communication topologies [38], including over wireless V2V communication channels in a concurrent
predecessor following (PF) topology, leader-predecessor-fol- manner for accomplishing a cooperative longitudinal control
lowing (LPF) topology, bidirectional (BD) topology, leader- objective. However, due to the existence of malicious adver-
bidirectional (LBD) topology, two predecessor-following saries, the data packets sent at time step k from each vehicle i
(TPF) topology, and leader-two-predecessor-following (LTPF). may not arrive at its neighboring vehicles immediately and
Assumption 1: For each follower vehicle i ∈ V , the local successfully at the same time step k. For example, one typical
state estimate x̂i (k) is shared with its underlying outgoing attacking scenario is that a malicious vehicle driving aside the
neighbors specified by Niout . platoon is to deliberately block desired information exchanges
Remark 1: It is much less demanding to exchange the vehic- between vehicles. Another possible attacking scenario is that a
ular state estimate x̂i (k) than the unknown full state xi (k) for signal jammer mounted on a flying drone over the platoon is
each vehicle i ∈ V . This is because the real-time state compo- equipped with some signal broadcasting device in such a way
nents, including position pi (k) , velocity vi (k) and acceleration as to jam the V2V communication channels whenever the
ai (k), are often obtained via on-board sensors. Some state vehicles transmit their information. A direct consequence of
components such as the real-time velocity and acceleration are such malicious attacks is that the real-time V2V communica-
difficult to measure accurately in practice. In contrast, the tions are intermittently interrupted, resulting in the vehicular
state estimate x̂i (k) can be precisely provided by the local state data packet losses over the network. Our emphasis is then
observer of each CAV. placed on investigating the effects of such malicious DoS
attacks. To capture the intermittent behavior of DoS attacks,
E. Platoon Safety we denote by Sm ≜ {tm |m ∈ N≥0 ; tm ∈ N≥0 ; t0 = 0} the set of
When the vehicle longitudinal dynamics are linear and non-decreasing packet arriving time steps of the transmitted
deterministic, string stability is often analyzed using the fre- data packets at their destinations and Im ≜ {tm , tm + 1, . . . , tm+1 −
quency domain approach for assessing platoon safety. How- 1} the data packet arriving intervals between the last arriving
ever, considering the uncertain heterogeneous longitudinal time step tm and the current time step tm+1 for any i ∈ V . Fur-
model (1) or (6) and the generic wireless communication thermore, define Dm ≜ {tm+1 − tm |tm+1 − tm ≥ 1; m ∈ N≥0 } , wh-
topology G as well as the intermittent DoS attacks, such a ere each element in the set Dm is referred to as the DoS attack
string stability notion is difficult to be adopted here to evalu- duration. When the V2V communication channels are free of
ate the platoon performance. Therefore, we propose an effec- DoS attacks, one has that tm+1 − tm = 1.
tive collision avoidance mechanism to guarantee safe behav- Given the time-stamped feature of the transmitted vehicular
iors of platooning vehicles. data packets, a data packet analyser (DPA) is developed in

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1240 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

To neighboring From neighboring


vehicles Wireless V2V communication vehicles

Transmitter WiFi Receiver


Noise ϑi(k)
(k, x^i(k)) {(tm, x^j(tm)), ∀j∈Ni}
pi(k) + yi(k) x^i(k) τM
GPS + Local state observer Data packet
(x^i(k), Δ^ i(k)) analyser {tm}

Ri(k) uioa(Ri(k)) ηi(k)


Radar Built-in collision Δ^ i(k) Attackid(k)
avoidance mechanism K

Engine control ui(k) High-level +



unit controller +
Kηi(k)

Fig. 2. The longitudinal control system architecture on each platoon vehicle i, where its main system components include sensor units (radar, GPS), a wire-
less transceiver (transmitter, receiver), data processing units (a local state estimator, a high-lever controller, a collision avoidance mechanism, a data packet
analyser), and an engine control unit.

Algorithm 1 for each platoon vehicle i ∈ V to manage the designed, ∆ˆ i (k) denotes the estimate compensation for the
transmissions of the data packets and also detect the intermit- lumped uncertainty, uoa i (Ri (k)) denotes the repulsive force
tent DoS attacks. More specifically, the main function of the input acting on each vehicle i in such a way as to avoid poten-
DPA includes: tial collision with its predecessor, and ηi (k) denotes the neigh-
1) Detecting attack occurrence at each time step and out- borhood tracking error defined as
putting an indication signal Attackid (k); ∑ ( )
2) Storing the data packet arriving instants {tm }; ηi (k) = ai j x̂i (tm ) − di − ( x̂ j (tm ) − d j ) (13)
3) Recording the maximal DoS attack duration τ M ; and j∈Niin ∪{0}
4) Outputting the aggregated data ηi (k) for each vehicle i to where di = [−i × d0 , 0, 0]T denotes the prescribed spacing vec-
form the feedback control input Kηi (k) in (12). tor associated with follower i with d0 representing the longitu-
Algorithm 1 Data Packet Analyser (DPA) i, ∀ i ∈ V dinal distance between follower i and the leader 0, and
x̂0 (tm ) = x0 (k), meaning that the leader’s state is secure during
Input: Monitoring start time k0 and window size T S network communication.
Last recorded data packet arrival instant tm
Output: τ M , {tm }, {Attackid (k)}, and {ηi (k)} H. Estimation Error and Tracking Error Dynamics
Data: (k, x̂i (k)) and {(k, x̂ j (k)), j ∈ Niin } For each follower i ∈ V , define the estimation errors as
1: for k = k0 : k0 + T S do
2: Broadcast (k, x̂i (k)) to its out-neighbors Niout over G x̃i (k) = x̂i (k) − xi (k), ∆˜ i (k) = ∆ˆ i (k) − ∆i (k)
3: if data packets {(k, x̂ j (k)), j ∈ Niin } arrived then and the local estimation tracking error as
4: Set Attackid (k) = “off ” δi (k) = x̂i (k) − x0 (k) − di .
5: if k − tm − 1 > τ M then Then, the tracking error ηi (k) in (13) can be rewritten as
6: Update the bound τ M ← k − tm − 1 ∑ ( )
7: end if ηi (k) = ai j δi (tm ) − δ j (tm ) . (14)
8: Update the packet arrival instant tm ← k j∈Niin ∪{0}
9: Compute the fusion data ηi (k) in (13)
For notational convenience, define the stacked vectors
10: Update the packet arrival instant index m = m + 1
11: else //* DoS attack occurs *// νi (k) = [ϑiT (k), ∆Ti (k + 1) − ∆Ti (k)]T
12: Set Attackid (k) = “on”
13: Retrieve the last packet arrival instant tm ξi (k) = [ x̃iT (k), ∆˜ Ti (k)]T , d = [d1T , d2T , . . . , dNT ]T
14: Compute the fusion data ηi (k) in (13) ν(k) = [ν1T (k), ν2T (k), . . . , νTN (k)]T
15: end if
16: end for ξ(k) = [ξ1T (k), ξ2T (k), . . . , ξNT (k)]T
δ(k) = [δT1 (k), δT2 (k), . . . , δTN (k)]T
G. A Resilient and Safe Platoon Tracking Control Law
x̃(k) = [ x̃1T (k), x̃2T (k), . . . , x̃TN (k)]T
As illustrated in Fig. 2 , we propose the following dis-
tributed cooperative longitudinal control law on follower vehi- ũ0 (k) = [u0 (k), u0 (k), . . . , u0 (k)]T
cle i ∈ V :
ũoa (k) = [uoa
1 (R1 (k)), u2 (R2 (k)), . . . , uN (RN (k))] .
oa oa T

ui (k) = Kηi (k) − ∆ˆ i (k) + uoa


i (Ri (k)), k ∈ Im (12) Combining (7) and (10), one obtains the following estima-
where K denotes a feedback control gain matrix to be tion error dynamics:

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1241

ξ(k + 1) = (I ⊗ Ā)ξ(k) + (I ⊗ Ē)ν(k) (15) Ē T = [LT , F T ] = Ě P̄−1. Furthermore, implementing the obser-
[A B] ver (10) under Ē , one can further derive the estimates x̂i (k)
where Ā =  + ÊC̄ and Ē = [−Ê, [0, −1]T ] with  = 0 1 , and ∆ˆ i (k) for each follower i ∈ V at time step k ∈ Z≥0.
[ ]
Ê = FL , and C̄ = [C, 0]. Step 2: Solve out the feedback control gain matrix K . For
In view of (8)−(10), (12) and (14), the platoon tracking error scalars κ1 > 0, κ2 > 0 , and a positive integer τ M , solve the fol-
dynamics are given as lowing linear matrix inequality:
[ ]
δ(k + 1) = Ãδ(k) + B̃δ(tm ) + ϖ(k), k ∈ Im (16) Ξ̌m Ṽ
< 0, m = 1, 2 (20)
ṼT −κ1 I
where à = I ⊗ A, B̃ = H ⊗ BK , and ϖ(k) = (I ⊗ LC) x̃(k)−
(I ⊗ B)ũ0 (k) + (I ⊗ B)ũoa (k) + (I ⊗ A)d − (I ⊗ L)v(k). to determine matrices Qq > 0, q = 1, 2, 3, 4 , and V, K , where
Ξ̌m is symmetric 4-by-4 block matrix whose nonzero entries
I. The Problem to be Addressed
are given by
The longitudinal platooning controller design aims to deter-
mine a desired distributed coordination control law ui (k), Ξ̌1,1 = Q2 − Q3 − α2 Q4 + (A − I)V + VT (A − I)T
i ∈ V , such that the N followers eventually reach the same Ξ̌1,2 = Q1 − V + κ2 VT (A − I)T
speed and acceleration as the leader 0; i.e.,
Ξ̌1,3 = α2 Q4 + ςm BK, Ξ̌1,4 = Q3
evi (k) = vi (k) − v0 (k) → 0, eai (k) = ai (k) − a0 (k) → 0
as k → ∞, while simultaneously maintaining a desired longitu- Ξ̌2,2 = Q1 + τ2M Q3 + Q4 − κ2 V − κ2 VT
dinal spacing d0 between any two adjacent vehicles; viz., the Ξ̌2,3 = ςm κ2 BK, Ξ̌3,3 = −α2 Q4
position tracking error edi (k) satisfies
Ξ̌4,4 = −Q2 − Q3 , V̌ = diag{V, V, V, V}
edi (k) = pi (k) − p0 (k) − di0 → 0 as k → ∞
with α = 2 sin(π/(2(2τ M + 1))), ς1 = min{λi } and ς2 = max{λi }.
where di0 = −i × d0 with d0 being the specified constant spac- If (20) is feasible, then one has that K = KV−1. Furthermore,
ing. Note that edi (k) → 0 implies that the spacing error under a specified V2V communication topology, form the
eis (k) = pi−1 (k) − pi (k) − d0 → 0. (17) neighborhood tracking error in (13) and determine the input
term Kηi (k) of ui (k) for each follower i ∈ V .
However, given the simultaneous presence of unknown Step 3: Design the collision avoidance term uoa i (Ri (k)). For
nonlinearity ∆i (k), leader control input u0 (k), collision avoid-
any i ∈ V , based on the real-time relative distance Ri (k), the
ance function uoai (k), and measurement noise ϑi (k) , it is gener- potential of the relative distance is chosen as
ally difficult to preserve the accurate asymptotic stability of
−βi ρi (k)
the platoon tracking errors edi (k), evi (k), eai (k), or equivalently, Fioa (Ri (k)) = Ωi (k) (21)
ei (k) = [edi (k), evi (k), eai (k)]T = xi (k) − x0 (k) − di (18) where βi ≥ 0 is a prescribed scalar, ρi (k) = 1 − sgn(Ri (k) − S 0 )
and Ωi (k) = 1 − γi Φ2i (k)/(1 + Φ2i (k)) with γi = 1 + S 0−4 and
to zero for all i ∈ V as k → ∞. Instead, we aim to address the
Φi (k) = R2i (k) − S 02 .
following platooning control problem.
Then, the repulsive force input for collision avoidance at
Problem 1: For the V2V-enabled vehicular platoon of the
vehicle i is defined as the negative gradient of the potential
uncertain longitudinal dynamics (7), the objective is to design
Fioa (Ri (k)) along pi (k) , i.e.,
an admissible DoS-resilient and collision-free distributed
i (Ri (k)) = − ∇ pi F i (Ri (k))
uoa
cooperative longitudinal control law ui (k) of the form (12) for oa

any k ∈ Iim such that the resulting estimation error ξi (k) in (15)
4γi βi ρi (k)Φi (k)Ri (k) −βi ρi (k)−1
and the platoon tracking error ei (k) in (18) are uniformly ulti- = Ωi (k). (22)
(1 + Φ2i (k))2
mately bounded against system uncertainties and attacks.

III. Main Results Remark 2: From (22), if Ri (k) → 0, then Ωi → 0 and thus
i (Ri (k)) → ∞. If Ri (k) > S 0, then ρi (k) = 0 and thus
uoa
A. A Three-Step Design Algorithm i (Ri (k)) = 0. On the other hand, if Ri (k) ≤ S 0, one has that
uoa
−βi ρi (k)−1
The following algorithm outlines the main design steps of i (Ri (k)) ≤ 4γi βi ρi (k)Φi (k)Ri (k)Ωi
uoa (k). Furthermore,
−βi ρi (k)−1
the observer-based platooning control laws in (10) and (12). βi ρi (k)Ωi (k) is bounded, one thus only needs to guar-
1
Step 1: Determine the local observer gain matrices L and F antee that the function f (Φi ) = 4γi Φi (Φi + S 02 ) 2 is bounded.
in (10), and derive the estimates x̂i (k) and ∆ˆ i (k). For a scalar Indeed, the maximal value of f (Φi ) can be readily found via
κ0 > 0 , find feasible matrices P̄ > 0 and Ě such that the extreme value theorem, which equals 8αi S 03 /3 2 at Φi =
3

 
 −P̄ ÂT P̄ + C̄ T Ě ÂT P̄ + C̄ T Ě  −(2/3)S 02 . As a result, uoa i (Ri (k)) in (22) is bounded. In other
 T  < 0 words, uoa (R (k)) in (22) is a valid collision function as speci-
P̄ Â + Ě C̄ −P̄
T (19)
0  i i
fied in Definition 1.
P̄ Â + Ě C̄
T T 0 −κ0 I
Remark 3: From Steps 1 and 2, it is clear that the observer
holds, where  and C̄ are given in (15). Then, one obtains that and controller gain matrices L, F and K can be both deter-

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1242 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

mined in an offline manner via solving the linear matrix nonzero entries given by
inequalities (19) and (20). Regarding the positive integer τ M
Ξ1,1 = Q̄2 − Q̄3 − α2 Q̄4 , Ξ1,2 = Q̄1 , Ξ1,3 = α2 Q̄4
in Step 2, the following two aspects are noted. First, given the
offline feature of the design algorithm above, the system Ξ1,4 = Q̄3 , Ξ2,2 = Q̄1 + τ2M Q̄3 + Q̄4
designers may use τ M as a tuning parameter for evaluating the
Ξ3,3 = −α2 Q̄4 , Ξ4,4 = −Q̄2 − Q̄3
resilience of the proposed design algorithm against intermit-
tent DoS attacks. This enables a tradeoff analysis between pla- V̌ = [I ⊗ V, I ⊗ κ2 V, 0, 0], Ǎ = [(Ã − I), −I, B̃, 0].
toon performance and attack severity, which will be demon- Furthermore, the spacing errors eis (k), i ∈ V , in (17) on each
strated through a numerical case study in Section IV-B. In follower vehicle can be regulated into some small neighbor-
particular, the maximal attack resilience can be determined by hood around zero.
finding a solution to an optimization problem τ∗M = max(τ M ) Proof: See Appendix. ■
subject to (20). In this case, τ∗M represents an estimate of the Remark 5: It is noted from (23) that the observer gain matri-
attack duration bound τ M in the DPA and can be interpreted as ces L, F can not be directly solved out due to the coupling
the maximum allowable DoS duration bound under the pro- with the Lyapunov matrix P̄ . The equivalence between the
posed platooning control law. Second, one may turn the pro- Riccati equation (23) and the linear matrix inequality (19) can
posed controller design step into an online procedure to estab- be verified via the Schur complement [39]. On the other hand,
lish a direct relationship between the real-time network condi- due to the nonlinear terms of V T BK in (24), it is difficult to
tion under DoS attacks and the adjustable control perfor- solve out the control gain K . Furthermore, (24) involves the
mance. For example, the proposed DPA can be employed to matrix H and the Kronecker product ⊗ . This means that the
monitor the interrupted network condition and record the control design criterion specified by (24) is not scalable to a
attack duration τ M over some sliding window. Such a parame- vehicular platoon of a variable or large size. To tackle the two
ter τ M is then used to achieve the resilient controller design in issues above, pre-multiplying and post-multiplying (24) by
Step 2. diag{U −1 ⊗ I, U −1 ⊗ I,U −1 ⊗ I, U −1 ⊗ I} and its inverse, respec-
Remark 4: A salient feature of the proposed design algo- tively, and recalling U −1 HU = Λ, the left-hand side of the
rithm above lies in its scalability to variable platoon lengths resultant inequality is an N -block diagonal matrix. Applying
under certain V2V communication topologies. In other words, the Schur complement [39], one further has that
even when the number of platooning vehicles varies, the [ ]
Ξ̃ + Ṽ T Ai + ATi Ṽ I
implementing controllers may remain functional. For exam- Θ̃ = < 0, ∀ i ∈ V (25)
I −κ1 I
ple, under a PF topology, ς1 = ς2 = 1 in (20) holds for any
number of vehicles (i.e., N ); under LPF and TPF topologies, where Ξ̃ is symmetric 4-by-4 block matrix with its nonzero
ς1 = 1 and ς2 = 2 hold for any N ; under an LTPF topology, entries given by
ς1 = 1 and ς2 = 3 hold for any N. Although ς1 and ς2 are dif-
ferent under BD and LBD topologies when the vehicle length Ξ̃1,1 = Q2 − Q3 − α2 Q4 , Ξ̃1,2 = Q1
varies, one may predict the potential maximum platoon length Ξ̃1,3 = α2 Q4 , Ξ̃1,4 = Q3 , Ξ̃2,2 = Q1 + τ2M Q3 + Q4
and derive a worst-case platoon controller design. Such a scal-
able feature is promising for flexible platoon maneuvers, such Ξ̃3,3 = −α2 Q4 , Ξ̃4,4 = −Q2 − Q3
as platoon splitting and merging as well as lane changing. Ṽ = [V, κ2 V, 0, 0], A = [(A − I), −I, λi BK, 0].
B. Formal Stability Analysis It is clear from the second diagonal element of Θ̃ that
The following theorem states a formal stability analysis cri- κ2 (V + V T ) > Q1 + τ2M Q3 + Q4 > 0, which indicates that V is
terion for the estimation error dynamics (15) and platoon nonsingular. Then, pre-multiplying and post-multiplying (25)
tracking error dynamics (16). by diag{V −T , V −T , V −T , V −T } and its transpose, respectively,
Theorem 1: For the vehicular platoon (7) and (8), under the and denoting V = V −1 , Qq = VT Qq V , q = 1, 2, 3, 4 , and K =
position-based local observer (10) and the intermittent pla- KV yield that
toon tracking control law (12), all variables of the estimation [ ]
Ξ̌i V̌
error dynamics (15) and the tracking error dynamics (16) are < 0, ∀ i ∈ V (26)
uniformly ultimately bounded if 1) there exists a matrix P̄ > 0 V̌T −κ1 I
such that the following discrete algebraic Riccati equation: where Ξ̌i shares the same structure with Ξ̌m in (20) except
replacing ςm in Ξ̌m by λi . It is noteworthy that the inequalities
ĀT P̄Ā − P̄ + κ0−1 ĀT P̄P̄Ā = −Ū (23) (26) are linear with respect to V, Qq , q = 1, 2, 3, 4 , and K for
holds, where k0 > 0 is a scalar and Ū > 0 is a matrix, and 2) fixed λi . As a result, it suffices to verify two inequalities in
there exist matrices Q̄q = I ⊗ Qq > 0, q = 1, 2, 3, 4 , and V such terms of the largest and smallest eigenvalues ς1 and ς2 , which
that the following linear matrix inequality: straightforwardly leads to (20).
Θ = Ξ + V̌ T Ǎ + ǍT V̌ + κ1−1 I < 0 (24) IV. Numerical Experiments
holds, where κ1 , κ2 are positive scalars, τ M is a positive inte- In this section, the effectiveness and merits of the proposed
ger, Ξ = [Ξm,n ] is symmetric 4-by-4 block matrix with its platooning control approach are validated via a convoy of

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1243

seven CAVs moving on a real-world 9.65 km section of road- four speed references (80, 120, 90, 40 ) km/h during platoon
way measured by [40]. The road profile, including road length driving on the road segment shown in Fig. 3(a). Specifically,
and road elevation, is shown in Fig. 3(a). the platoon maneuver includes two acceleration stages (from
initial 40 to 80 and then from 80 to 120) and two deceleration
350 stages (from 120 to 90 and then from 90 to 40). For this pur-
340 120
Elevation (m)

90 40 pose, the platoon leader is commanded according to the speci-


330
320 80 fied control input and speed reference as given in Fig. 3(b). It
310
300 is noteworthy that the varying road slopes and leader com-
290 mand changes above pose a significant challenge for preserv-
0 1 2 3 4 5 6 7 8 9 10
Distance (km) ing the platoon stability and performance.
(a) Road profile
In what follows, we further examine two typical V2V com-
0.5 munication topologies for each platoon scenario above.
120
v0(t) (km/h)

100 0
1) Predecessor-Follower Topology (PF): Each follower

u0(t) (m/s2)
80 i ∈ V only obtains information (state estimate x̂i−1 (tm )) from
60 −0.5
its direct predecessor i − 1. Under this communication topol-
40
−1.0 ogy, the parameter τ M , if not elsewhere clarified, is set as 58,
0 50 100 150
200 250 300 350 400
Time (s) namely, the maximal DoS attack duration is 0.58 s. During the
(b) Commanded leader control input and leader speed collision avoidance operation, the prescribed scalars βi are set
as {βi }6i=1 = {0.10 , 0.08, 0.07, 0.08, 0.09 , 0.12}.
Fig. 3. The profile of a real-world 9.65 km section of roadway: (a) road ele-
2) Bidirectional Topology (BD): Each follower i ∈ V \ {N}
vation [40] and (b) the commanded leader control input u0 (t) and speed v0 (t)
gathers information (state estimates x̂i−1 (tm ) and x̂i+1 (tm ))
for achieving and maintaining the specified speed references (80 km/h, 120
from both its direct predecessor i − 1 and successor i + 1, while
km/h, 90 km/h, 40 km/h).
the tail follower N receives information only from its prede-
A. Platooning Scenarios and Parameter Setups cessor N − 1. Under this communication topology, if not else-
where clarified, τ M = 11, namely, the maximal DoS attack
We consider the following two platooning scenarios.
duration is set as 0.11 s. During the collision avoidance opera-
(S1) A platoon of seven passenger vehicles: the heterog-
tion, the prescribed scalars βi are set as {βi }6i=1 = {0.20 , 0.15,
enous vehicle masses are chosen as {mi }6i=0 = {2.350 , 2.260 ,
0.18, 0.18, 0.10 , 0.10}.
2.393, 2.400 , 2.380 , 2.385, 2.295} ton.
(S2) A platoon of seven heavy-duty vehicles: the heterog- B. Trade-Off Analysis Between Platoon Performance and Attack
enous vehicle masses are chosen as {mi }6i=0 = {38.5 , 37.5, 40.5, Severity
41.1, 38.8, 39.5, 38.0} ton. In order to facilitate comparison and performance analysis,
The other platoon parameters in both scenarios are outlined we introduce a platoon performance metric Rave (t), called the
as follows. The scalar parameters in the design algorithm are real-time average spacing (average inter-vehicle distance), at
taken as κ0 = 1000, κ1 = 0.5 and κ2 = 3.5. The safe distance the continuous time t , which is given by Rave (t) ≜ (1/N)×
threshold is set as S 0 = 5 m. To characterize the air drag force ∑N ∑N ( )
i=1 Ri (t) = (1/N) i=1 pi−1 (t) − pi (t) .
fiw (t), we assume a time-varying wind velocity vw (t) = We first perform a trade-off analysis between resultant pla-
Aw sin(2π fw t) m/s (positive for headwind and negative for tail- toon performance (characterized by Rave (t)) and different lev-
wind) on each follower vehicle i ∈ V during platoon maneu- els of attack severity (specified by τ M ), under the above two
vering in both (S1) and (S2) scenarios, where Aw = 10 and the platooning scenarios (S1) and (S2) as well as the two topolo-
excitation frequency fw = 0.01 Hz. Furthermore, ρ = 1.2 gies (PF) and (BD). More specifically, we focus on the
kg/m3, AF = 2.23 m2, Cd = 0.59 and µR = 0.01 in the (S1) sce- resilience of the controlled vehicular platoons. In other words,
nario; ρ = 1.29 kg/m3, AF = 10.26 m2, Cd = 0.6 and an emphasis will be placed on examining how different val-
µR = 0.007 in the (S2) scenario; the road inclination angle ues of the DoS duration bound τ M affect the platoon perfor-
θi (t) and its derivative in both scenarios are obtained based on mance and what is the possible maximal allowable DoS dura-
the road profile shown in Fig. 3(a). The platoon starts from the tion bound under each simulated platooning scenario. For this
origin with a constant cruise speed (p0 (0), v0 (0), a0 (0)) = reason, the proposed collision avoidance function uoai (Ri (k)) in
(0 m, 40 km/h, 0 m/s 2 ) and zero initial tracking errors. The (22) is not applied in this case study.
inertial delays for each platoon vehicle in either platoon sce- Applying the three-step design algorithm in Section III-A,
nario above are taken as {τi }6i=1 = {0.770, 0.815, 0.820 , 0.805, the local observer gain matrices L, F and controller gain
0.810 , 0.780} s with τ0 = 0.800 s. The desired spacing bet- matrix K can be designed in each case. For the sake of con-
ween any two successive vehicles is set as d0 = 10 m. The cise, the matrices are not outlined here. Implementing the
sampling period is chosen as h = 10 ms, and the simulation designed local observer (10) and control law (12), we have the
time is taken as 418 s. The position measurements on each simulation results in Fig. 4 under several numerical tests. It is
vehicle suffer from some random noises ϑi (k) causing noteworthy that the DoS attacks subject to the constrained
[−0.1, 0.1] m fluctuations on the GPS position readings. duration bound τ M are simulated randomly in each case. For
To simulate a dynamic maneuvering environment, we set example, τ M = 10 means that the simulated DoS attacks occur

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1244 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

15 15

N Σi=1 Ri(t)
10 10

1 N
Safe zone τM = 1 (0.01 s) Safe zone τM = 1 (0.01 s)

Rave(t) = −
5 τM = 10 (0.1 s) 5 τM = 2 (0.02 s)
τM = 20 (0.2 s) τM = 4 (0.04 s)
τM = 30 (0.3 s) τM = 6 (0.06 s)
Risky zone τM = 40 (0.4 s) Risky zone τM = 8 (0.08 s)
0 τM = 50 (0.5 s)
0 Collided zone τM = 10 (0.1 s)
Collided zone τM = 60 (0.6 s) τM = 12 (0.12 s)

0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(a) Average interval-vehicle distance (S1-PF) (b) Average interval-vehicle distance (S1-BD)

15 15
N Σi=1 Ri(t)

10 10
1 N

Safe zone τM = 1 (0.01 s) Safe zone τM = 1 (0.01 s)


Rave(t) = −

5 τM = 10 (0.1 s) 5 τM = 2 (0.02 s)
τM = 20 (0.2 s) τM = 4 (0.04 s)
τM = 30 (0.3 s) τM = 6 (0.06 s)
Risky zone τM = 40 (0.4 s) Risky zone τM = 8 (0.08 s)
τM = 50 (0.5 s) τM = 10 (0.1 s)
0 Collided zone τM = 60 (0.6 s) 0 τM = 12 (0.12 s)
Collided zone
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(c) Average interval-vehicle distance (S2-PF) (d) Average interval-vehicle distance (S2-BD)

Fig. 4. The resultant average inter-vehicle distances Rave (t) under different scenarios and V2V topologies: (a) Platooning scenario (S1) and topology (PF); (b)
Platooning scenario (S1) and topology (BD); (c) Platooning scenario (S2) and topology (PF); (d) Platooning scenario (S2) and topology (BD).

intermittently and randomly with an upper duration bound of 60 1400


0.1 s. In a particular case, τ M = 1 means that there is no DoS 50 τM = 58 (0.58 s) 1200 τM = 11 (0.11 s)
40 1000
attack and the V2V communication channels are deemed to be
800
ideal. It can be clearly observed from Fig. 4 that the intermit- 30
600
tent DoS attacks exert a major and adverse impact on the 20 400
10 200
desired platoon performance. Specifically, as the attack sever-
0 0
ity increases (i.e., τ M goes larger), the platoon performance 0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.02 0.04 0.06 0.08 0.10
{tm+1 − tm}(s) {tm+1 − tm}(s)
becomes degraded (i.e., Rave (t) becomes larger). As shown in
(a) Distribution of attack durations (PF) (b) Distribution of attack durations (BD)
Fig. 4(a), when the DoS duration bound is 0.6 s, the real-time
average spacing Rave (t) becomes negative at around t = 365 s, Fig. 5. The distribution of the simulated intermittent DoS attack durations
which indicates serious collision of the simulated platoon in {tm+1 − tm }: (a) τ M = 58 under PF; (b) τ M = 11 under BD.
the (S1-PF) case. Similar collision situations can be spotted in
Figs. 4(c) and 4 (d). Although Rave (t) is not negative under (22). Under the simulated DoS attacks in Fig. 5 , the relative
τ M = 12 during the simulated time window in Fig. 4 (b), it is positions to the leader, i.e., p0 (t) − pi (t), and the real-time
found through the individual spacings Ri (t) that some vehicles inter-vehicle distances, i.e., Ri (t), in different cases are
already collided with their predecessors. depicted in Fig. 6 . It is clear that without adopting the pro-
In summary, the DoS attacks that intermittently prevent the posed collision avoidance mechanism, the platoon fails to
V2V data communications can significantly worsen the pla- maintain its driving safety under the intermittent DoS attacks.
toon performance and even jeopardize the platoon stability. Next, we consider the collision avoidance function uoa i (Ri (k))
On the other hand, even in the absence of the proposed colli- in (22) added into the desired platooning control law ui (k) as
sion avoidance mechanism, the derived design algorithm is in (12) and repeat the numerical experiment above. The results
promising for conducting a trade-off analysis between satis- are shown in Figs. 7 and 8. Specifically, Fig. 7 illustrates the
factory platoon performance and tolerable attack severity. relative positions to the leader and the real-time inter-vehicle
distances in different cases. It is obvious that the real-time
C. Resilient and Safe Platooning With Guaranteed Collision spacings between any two consecutive vehicles in either case
Avoidance are significantly enlarged in comparison with those in Fig. 6.
In this subsection, we investigate the security and safety Simple calculations of the inter-vehicle distances show that
performance of the controlled vehicle platoons under the pro- the minimal spacing of the simulated vehicular platoon is:
posed collision avoidance mechanism. For illustration pur- min∀i,t {Ri (t)} = 1.5715 m in the case (S1-PF); min∀i,t {Ri (t)} =
poses, we set the DoS duration bound as τ M = 58 under PF 4.3700 m in the case (S1-BD); min∀i,t {Ri (t)} = 1.6117 m in the
and τ M = 11 under BD. The distribution of the simulated inter- case (S2-PF); and min∀i,t {Ri (t)} = 4.3727 m in the case (S2-
mittent DoS attack durations is illustrated in Fig. 5. BD). It is noteworthy that the minimal safe inter-vehicle dis-
First, we perform the numerical experiment without emplo- tance in either case above can be further enlarged by suitably
ying the proposed collision avoidance function uoa i (Ri (k)) in configuring the collision avoidance function and its parame-

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1245

Veh 1 Veh 2 Veh 3 Veh 4 Veh 5 Veh 6

80
20
Safe zone
60

pi−1(t) − pi(t)
p0(t) − pi(t)

40 10
20
Risky zone
0
0 Collided zone Collision occurred
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(a) Relative positions to leader (m) (b) Real-time interval-vehicle distances (m)

80
20
60 Safe zone

pi−1(t) − pi(t)
p0(t) − pi(t)

40
10
20
Risky zone
0
0 Collided zone Collision occurred
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(c) Relative positions to leader (m) (d) Real-time interval-vehicle distances (m)

80
20
Safe zone
60

pi−1(t) − pi(t)
p0(t) − pi(t)

40 10
20
Risky zone
0
0 Collided zone Collision occurred
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(e) Relative positions to leader (m) (f) Real-time interval-vehicle distances (m)
80
20
60 Safe zone

pi−1(t) − pi(t)
p0(t) − pi(t)

40
10
20
Risky zone
0
0 Collided zone Collision occurred
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(g) Relative positions to leader (m) (h) Real-time interval-vehicle distances (m)

Fig. 6. The resultant relative positions to the leader (i.e., p0 (t) − pi (t) ) and the real-time inter-vehicle distances (i.e., pi−1 (t) − pi (t)) without the proposed colli-
sion avoidance mechanism in different cases: The first row (S1-PF); The second row (S1-BD); The third row (S2-PF); The last row (S2-BD).

ters. We only demonstrate here that the proposed collision the estimation error stability and tracking error stability. On
avoidance function has a great potential to increase the inter- the other hand, the string stability of a longitudinal platoon
vehicle distances in all cases in such a way as to avoid possi- control system generally depends on the desired platoon con-
ble collisions. On the other hand, Fig. 8 depicts the trajecto- trol law and spacing policy as well as the V2V communi-
ries of the collision avoidance function uoa i (Ri (k)). It can be cation topology configured. The notion of string stability
seen that only when vehicle i enters its potential collision-risk adopted in the existing literature is relatively diverse, among
zone specified by Ωi (pi ) = {pi | Ri ≤ 5 m}, the collision avoid- which the L2-string stability [35], [41]–[43] and the L∞-string
ance function uoai (Ri (k)) is activated to push vehicle i away stability [36], [43] are widely explored. In this study, we aim
from its direct predecessor i − 1 to circumvent collision.
to develop a scalable distributed platoon control approach that
D. Platoon Stability and Controller Performance accounts for generic communication topologies. Furthermore,
In this subsection, we take a close look at the stability of the the proposed platoon control system exhibits the simultane-
controlled vehicle platoon and the performance of the pro- ous presence of a variety of uncertainties (including unknown
posed resilient platoon controllers. Two stability notions heterogenous engine inertial delay, unknown resistance forces,
including individual vehicle stability and string stability are unknown measurement noise and leader input, intermittent
used here to verify the proposed design approach. DoS attacks). These two observations make it difficult to for-
Specifically, the individual vehicle stability corresponds to mally prove the uniform string stability of the concerned pla-

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1246 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

80
20
Safe zone
60

pi−1(t) − pi(t)
p0(t) − pi(t)
40 10
20
Risky zone
0 Collision avoided
0 Collided zone
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(a) Relative positions to leader (m) (b) Real-time interval-vehicle distances (m)

80
20
60 Safe zone

pi−1(t) − pi(t)
p0(t) − pi(t)

40
10
20
Risky zone
0 Collision avoided
0 Collided zone
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(c) Relative positions to leader (m) (d) Real-time interval-vehicle distances (m)

80
20
Safe zone
60

pi−1(t) − pi(t)
p0(t) − pi(t)

40 10
20
Risky zone
0 Collision avoided
0 Collided zone
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(e) Relative positions to leader (m) (f) Real-time interval-vehicle distances (m)
80
20
60 Safe zone

pi−1(t) − pi(t)
p0(t) − pi(t)

40
10
20
Risky zone
0
0 Collided zone Collision avoided
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(g) Relative positions to leader (m) (h) Real-time interval-vehicle distances (m)

Fig. 7. The resultant relative positions to the leader (i.e., p0 (t) − pi (t) ) and the real-time inter-vehicle distances (i.e., pi−1 (t) − pi (t)) with the proposed colli-
sion avoidance mechanism in different cases: The first row (S1-PF); The second row (S1-BD); The third row (S2-PF); The last row (S2-BD).

toon as in the literature [35], [36] and [41]–[43]. ing the lumped uncertainties ∆i (t) in vehicle longitudinal dyna-
To provide an intuitional demonstration of the string stabil- mics, from which one can obviously see that the uncertainties
ity of the controlled platoon, we only revisit the heavy-duty including the imposed disturbance on each vehicle can be well
vehicle platoon scenario (S2) under the BD topology afore- estimated and the resulting estimation errors are regulated into
mentioned while imposing an additional external disturbance some small bounds.
εi (t) = 0.8 sin(2π(1/10)(t − 200)) m/s2, ∀ t ∈ [200, 230] s on Fig. 10(a) shows the resulting string stability in terms of the
each platoon follower vehicle i . This can be done by adding L2-norms and L∞-norms of the spacing errors eis (t) = pi−1 (t)−
εi (t) into ∆i (vi , ai , ȧi ) in (8). It is clear that the imposed distur- pi (t) − d0 on each platoon vehicle. It can be seen that the strict
bance εi (t) causes each platoon vehicle’s acceleration to fluc- L2-string stability (monotonically decreasing L2-norms) is
tuate between −0.8 m/s2 and 0.8 m/s2 and has an adverse ensured. Even though the strict L∞-string stability is not
effect on the platoon control system stability. Boxplots are noticed, the L∞-norms of the spacing errors do not get ampli-
used here to depict the distribution of platoon tracking errors fied from platoon head to tail under the proposed platooning
in terms of position errors p0 (t) − pi (t) − i · d0 , speed errors approach.
v0 (t) − vi (t) and acceleration errors a0 (t) − ai (t), as shown in Fig. 10(b) demonstrates the distribution of each vehicle’s
Figs. 9(a)−9(c). It can be observed that the platoon tracking accelerations ai (t) during the simulated platoon driving sce-
errors can be driven into small neighborhoods around zero. nario. Specifically, the driving comfort of a heavy-duty pla-
Figs. 9(d)−9(i) shows the individual estimation (approxima- toon vehicle can be specified by the maximum deceleration of
tion) performance of the designed local state observers regard- amin = −3 m/s2 and the maximum acceleration of amax =

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1247

0 0

−0.5 0 −1 0
uioa(t), i ∈ v −1.0 −2

uioa(t), i ∈ v
−1 −2
−1.5 −3
−2
−2.0 −4 −4

−2.5 −3 −5 −6
290 300 310 365 370 375 380
−3.0 −6
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(a) Collision avoidance function in (S1-PF) case (b) Collision avoidance function in (S1-BD) case

0 0

−0.5 0 −1 0
−1.0 −2 −1

uioa(t), i ∈ v
uioa(t), i ∈ v

−1
−1.5 −3 −2

−2.0 −2 −4 −3

−2.5 −3 −5 −4
370 380 390 292 294 296 298 300 302
−3.0 −6
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
Time (s) Time (s)
(c) Collision avoidance function in (S2-PF) case (d) Collision avoidance function in (S1-BD) case

Fig. 8. The collision avoidance functions uoa


i (Ri (t)) in different cases.

2
p0(t) − pi(t) − i·d0

10 1

a0(t) − ai(t)
v0(t) − vi(t)

0
0 0
−10
−20
−2 −1
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
Vehicle index i, i = 1, 2, …, 6 Vehicle index i, i = 1, 2, …, 6 Vehicle index i, i = 1, 2, …, 6
(a) Position tracking errors (b) Speed tracking errors (c) Acceleration tracking errors

1.0 1.0 1.0


Superposition of disturbance ε1(t) Superposition of disturbance ε2(t) Superposition of disturbance ε3(t)
0.5 0.5 0.5
0 0 0
−0.5 −0.5 −0.5
−1.0 Δ1(t) −1.0 Δ2(t) −1.0 Δ3(t)
Δ^ 1(t) Δ^ 2(t) Δ^ 3(t)
−1.5 −1.5 −1.5
0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
Time (s) Time (s) Time (s)
(d) Estimation of uncertainty (e) Estimation of uncertainty (f) Estimation of uncertainty

1.0 1.0 1.0


Superposition of disturbance ε4(t) Superposition of disturbance ε5(t) Superposition of disturbance ε6(t)
0.5 −0.5 0.5
0 0 0
−0.5 −0.5 −0.5
−1.0 Δ4(t) −1.0 Δ5(t) −1.0 Δ6(t)
Δ^ 4(t) Δ^ 5(t) Δ^ 6(t)
−1.5 −1.5 −1.5
0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
Time (s) Time (s) Time (s)
(g) Estimation of uncertainty (h) Estimation of uncertainty (i) Estimation of uncertainty

Fig. 9. Individual vehicle stability under platooning scenario (S2) and BD topology: (a)−(c) the distribution of the platoon position errors p0 (t) − pi (t) − i · d0,
speed errors v0 (t) − vi (t) and acceleration errors a0 (t) − ai (t); (d)−(i) the lumped uncertainties ∆i (t) (including the imposed disturbance εi (t) ) on vehicle i and its
estimate ∆ˆ i (t), i = 1, 2, . . . , 6.

2 m/s2 [44]. It is clearly shown that the accelerations of all the which meets certain driving comfort even in the presence of
platoon vehicles are guaranteed to lie in the range of uncertainties and malicious DoS attacks.
(−2.02, 0.65) m/s2 under the proposed platooning approach, Figs. 10(c) and 10(d) further illustrates the resultant engine

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1248 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

6
500 0.5

0
400 5

ai(t) (m/s2)
||esi(t)||L∞
||esi(t)||L2
−0.5
300 4 −1.0

200 −1.5
3
−2.0
1 2 3 4 5 6 1 2 3 4 5 6
Vehicle index i, i = 1, 2, …, 6 Vehicle index i, i = 1, 2, …, 6
(a) L2 and L∞ string stability (b) Driving comfort
50 40
0
30 −10
0 20 −20

f (t), i∈v
f ie(t), i∈v

120 125
10
0
0

re
−50

i
−10
−10
−20
118 120 122 124 −20
−100
0 100 200 300 400 0 50 100 150 200 250 300 350 400
Time (s) Time (s)
(c) Engine forces (kN) (d) Resistance forces (kN)

Fig. 10. String stability and controller performance under platooning scenario (S2) and BD topology: (a) The string stability in terms of the L2-norms and
L∞-norms of the spacing errors eis (t) = pi−1 (t) − pi (t) − d0 on each platoon vehicle i under the BD topology; (b) The distribution of each vehicle’s accelerations
g
ai (t) under the BD topology; (c) The resultant engine forces fie (t) on each platoon vehicle i; (d) The lumped resistance forces Fi (t) = fi (t) + fiw (t) + fir (t) on
each platoon vehicle i, i = 1, 2, . . . , 6.

forces fie (t) and the lumped resistance forces Fi (t) on each develop effective platooning controllers that further guarantee
platoon vehicle i. Generally, the platoon vehicles require more satisfactory operational performance, such as fuel efficiency
traction forces to maintain speed uphill. In particular, during a and improved riding comfort, serves as an interesting and sig-
sharp deceleration stage, more braking forces are necessary nificant topic. As another potential extension, a resilient and
for the platoon vehicles to maintain speed transition and per- safe platooning control design method that can distinguish and
form safe driving, as shown in Fig. 10(c) over the time inter- further deal with simultaneous network-induced V2V commu-
val [360, 380] s. nication failures and DoS-incurred communication interrup-
Summarizing the above results, it can be concluded that the tions is preferable. Furthermore, it would be promising to pro-
proposed resilient and safe platooning control approach is pose resilient scalable platooning control solutions that deal
promising for preserving both DoS-resilience and collision- with various and concurrent network-induced constraints (e.g.,
free safety under the simultaneous model uncertainties and communication delays, finite bit rate, data quantization and
malicious DoS attacks in various maneuvering scenarios.
disorder, channel fading, random topologies [45], [46]).
V. Conclusions
Appendix
In this paper, the resilient and safe distributed cooperative Proof of Theorem 1
longitudinal platooning control problem is investigated for
The proof is divided into two parts.
CAVs in the presence of intermittent DoS attacks. An uncer-
tain nonlinear heterogenous longitudinal vehicle dynamical 1) Proof of the Boundedness of Estimation Errors: Choose
model is proposed, which incorporates unknown heteroge- the Lyapunov functional candidate as V1 (k) = ξT (k)(I⊗ P̄)ξ(k) .
nous masses and engine time lags, realistic resistance forces, The forward difference of V1 (k) along (15) is calculated as
and dynamic leader. To solve the platooning problem, a novel DV1 (k) = V1 (k + 1) − V1 (k)
platooning control law is developed to guarantee the simulta- ( )
neous individual vehicle stability, attack resilience, and driv- = ξT (k) I ⊗ (ĀT P̄Ā − P̄) ξ(k)
( )
ing safety. Furthermore, theoretical stability analysis and con- + νT (k) I ⊗ (Ē T P̄Ē) ν(k)
trol design conditions are derived such that the resultant spac- ( )
ing errors and platoon tracking errors are uniformly ulti- + 2ξT (k) I ⊗ (ĀT P̄Ē) ν(k).
mately bounded. Numerical experiments are finally presented By Young’s inequality, its last term is bounded as
to demonstrate the efficacy and advantages of the proposed ( )
2ξT (k) I⊗ (ĀT P̄Ē) ν(k)
platooning control method.
( )
The main focus of this study is on preserving the desired ≤κ0−1 ξT (k) I ⊗ (ĀT P̄P̄Ā) ξ(k)
platoon control performance, including individual vehicle sta- ( )
bility, driving safety, scalability, and attack resilience. How to + κ0 νT (k) I ⊗ (Ē T Ē) ν(k)

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1249

where κ0 is some positive scalar. To ensure the boundedness where ϵ3 = σ(Θ), and ϵ4 = κ1 σ(V̌ V̌ T )ϖ2M . Similarly, the inequ-
1
of all variables in the resulting estimation error dynamics (15), ality above is negative as long as ∥δ(k)∥ > (ϵ4 /ϵ3 ) 2. Therefore,
without loss of generality, the unknown measurement noise ϑi the platoon tracking error system (16) is uniformly ultimately
and the variation of the lumped nonlinearity ∆i (k) are sup- bounded.
posed to be continuous and bounded; i.e., ∥ν(k)∥ ≤ ν , where Furthermore, it is clear that
ν > 0 is not necessarily known. Then, one has that
∥e s (k)∥2 ≤ ∥e(k)∥2 = ∥δ(k)− x̃(k)∥2 ≤ ∥ x̃(k)∥2 + ∥δ(k)∥2
DV1 (k) ≤ −ξ (k)Uξ(k) + ϵ2 ν ≤ −ϵ1 ∥ξ(k)∥ + ϵ2 ν
T 2 2 2
which implies that the spacing error eis (k) for any i ∈ V is also
where ϵ1 = σ(U) and ϵ2 = σ(Ē T P̄Ē) + κ0 σ(Ē T Ē). Note that
1 bounded. Clearly, the bound of the spacing error eis (k) is
DV1 (k) is negative when ∥ξ(k)∥ > ν(ϵ2 /ϵ1 ) 2 , and thus the esti-
closely related to the magnitude-bounds ν, ϖ M as well as the
mation error system (15) is uniformly ultimately bounded
designed parameters κ0 , κ1 , P̄, Ū, V, Θ . By suitably adjusting
[47], which further implies that x̃i (k) and ∆˜ i (k) are bounded.
these parameters, the spacing error can be driven into some
2) Proof of the Boundedness of Platoon Tracking Errors
and Spacing Errors: We next choose the following Lyapunov small neighborhood around zero. ■
functional candidate V2 (k) = Vτ2 (k) + Vm2 (k), k ∈ Im , where
References

k−1
Vτ2 (k) = δ (k)Q̄1 δ(k) +
T
δ (s)Q̄2 δ(s)
T [1] X.-Y. Lu and S. Shladover, “ Truck CACC system design and DSRC
messages,” FHWA Explor. Adv. Res. Program Cooperat. Agreement,
s=k−τ M Task 2.1-Design CACC Control Syst. Trucks, Pleasanton, USA, Res.
Rep. DTFH61-13-H00012, 2018.

0 ∑
k−1
[2] S. Tsugawa, “ Results and issues of an automated truck platoon within
+ τM δ̆T (r)Q̄3 δ̆(r) the energy ITS project,” in Proc. IEEE Intell. Vehicles Symp., Dearborn,
s=−τ M +1 r=k+s−1 USA, Jun. 2014, pp. 642–647.

k−1 [3] S. Kato, S. Tsugawa, K. Tokuda, T. Matsui, and H. Fujii, “Vehicle
control algorithms for cooperative driving with automated vehicles and
Vm
2 (k) = δ̆T (s)Q̄4 δ̆(s) − α2 λ(k) intervehicle communications,” IEEE Trans. Intell. Transp. Syst., vol. 3,
s=−τ M no. 3, pp. 155–161, Sept. 2002.

with δ̆(r) = δ(r + 1) − δ(r), λ(k) = k−1 s=tm ρ (s) Q̄4 ρ(s) if
T k∈ [4] E. Chan, “SARTRE automated platooning vehicles,” in Towards Innov.
Freight Logistics. Hoboken, USA: Wiley, 2016, pp. 137–150.
Im \ {tm } and λ(k) = 0 if k = tm , ρ(s) = δ(s) − δ(tm ). By noting
[5] J. Ploeg, S. Shladover, H. Nijmeijer, and N. van deWouw, “Introduction
that δ̆(s) = ρ(s + 1) − ρ(s), it can be readily derived that to the special issue on the 2011 grand cooperative driving challenge,”
Vm2 (k) ≥ 0 [48]. Therefore, V2 (k) > 0 for any k ∈ Im .
IEEE Trans. Intell. Transp. Syst., vol. 13, no. 3, pp. 989–993, Sept. 2012.
Calculating the forward difference of V2 (k) along (16) [6] C. Englund, L. Chen, J. Ploeg, E. Semsar-Kazerooni, A. Voronov, H.
Bengtsson, and J. Didoff, “ The grand cooperative driving challenge
yields that 2016: Boosting the introduction of cooperative automated vehicles,”
IEEE Wireless Commun., vol. 23, no. 4, pp. 146–152, Aug. 2016.
DV2 (k) ≤ 2δT (k)Q̄1 δ̆(k) + δ̆T (k)Q̄1 δ̆(k) + δT (k)Q̄2 δ(k)
[7] T. Ersal, I. Kolmanovsky, N. Masoud, N. Ozay, J. Scruggs, R.
− δT (k − τ M )Q̄2 δ(k − τ M ) Vasudevan, and G. Orosz, “ Connected and automated road vehicles:
State of the art and future challenges,” Veh. Syst. Dyn., vol. 58, no. 5,
+ δ̆T (k)(τ2M Q̄3 + Q̄4 )δ̆(k) pp. 672–704, May 2020.
[8] F.-Y. Wang, N.-N. Zheng, D. Cao, C. Martinez, L. Li, and T. Liu,
− (δ(k) − δ(k − τ M ))T Q̄3 (δ(k) − δ(k − τ M )) “Parallel driving in CPSS: A unified approach for transport automation
and vehicle intelligence,” IEEE/CAA J. Autom. Sinica, vol. 4, no. 4,
− α2 (δ(k) − δ(tm ))T Q̄4 (δ(k) − δ(tm )) pp. 577–587, Oct. 2017.
where the discrete-time Jensen inequality [48] is applied. On [9] J. Zhang and K. Letaief, “ Mobile edge intelligence and computing for
the Internet of vehicles,” Proc. IEEE, vol. 108, no. 2, pp. 246–261, Feb.
the other hand, from the closed-loop system (16), it is true that 2020.
there exists a matrix V̌ such that [10] Y. Ma, Z. Wang, H. Yang, and L. Yang, “ Artificial intelligence
[ ] applications in the development of autonomous vehicles: A survey,”
2ϕT (k)V̌ T Ǎϕ(k) + ϖ(k) = 0 IEEE/CAA J. Autom. Sinica, vol. 7, no. 2, pp. 315–329, Mar. 2020.
where [11] H. Yu, R. Jiang, Z. He, Z. Zheng, L. Li, R. Liu, and X. Chen,
“Automated vehicle-involved traffic flow studies: A survey of
ϕ(k) = [δT (k), δ̆T (k), δT (tm ), δT (k − τ M )]T assumptions, models, speculations, and perspectives,” Transp. Res. Part
C: Emerg. Technol., vol. 127, p. 103101, Jun. 2021.
V̌ = [I ⊗ V, I ⊗ κ2 V, 0, 0], Ǎ = [(Ã − I), −I B̃, 0]. [12] C. Zu, C. Yang, J. Wang, W. Gao, D. Cao, and F.-Y. Wang,
“Simulation and field testing of multiple vehicles collision avoidance
By virtue of Young’s inequality, one has that algorithms,” IEEE/CAA J. Autom. Sinica, vol. 7, no. 4, pp. 1045–1063,
Jul. 2020.
2ϕT (k)V̌ T ϖ(k) ≤ κ1−1 ϕT (k)Iϕ(k) + κ1 ϖT (k)V̌ V̌ T ϖ(k).
[13] M. Fanti, A. Mangini, A. Favenza, and G. Difilippo, “ An eco-route
Furthermore, the leader control input u0 (k) is supposed to be planner for heavy duty vehicles,” IEEE/CAA J. Autom. Sinica, vol. 8,
no. 1, pp. 37–51, Jan. 2021.
continuous and bounded. On the other hand, uoa (Ri (k)) is also
bounded. Then, the augmented unknown input ϖ(k) satisfies [14] W. Xu, H. Zhou, N. Cheng, F. Lyu, W. Shi, J. Chen, and X. Shen,
“Internet of vehicles in big data era,” IEEE/CAA J. Autom. Sinica,
that ∥ϖ(k)∥ ≤ ϖ M . From (24), the forward difference of V2 (k) vol. 5, no. 1, pp. 19–35, Jan. 2018.
then can be calculated as [15] H. Zhou, W. Xu, J. Chen, and W. Wang, “Evolutionary V2X technolo-
gies toward the Internet of vehicles: Challenges and opportunities,”
DV2 (k) ≤ −ϕT (k)Θϕ(k) + ϵ4 ≤ −ϵ3 ∥δ(k)∥2 + ϵ4 Proc. IEEE, vol. 108, no. 2, pp. 308–323, Feb. 2020.

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
1250 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 5, MAY 2023

[16] X. Ge, Q.-L. Han, L. Ding, Y.-L. Wang, and X.-M. Zhang, “Dynamic Rev. Control, vol. 44, pp. 152–172, Oct. 2017.
event-triggered distributed coordination control and its applications: A [37] S. Feng, Y. Zhang, S. Li, Z. Cao, H. Liu, and L. Li, “String stability for
survey of trends and techniques,” IEEE Trans. Syst., Man, Cybern., vehicular platoon control: Definitions and analysis methods,” Annu.
Syst., vol. 50, no. 9, pp. 3112–3125, Sept. 2020. Rev. Control, vol. 47, pp. 81–97, Jun. 2019.
[17] S. Parkinson, P. Ward, K. Wilson, and J. Miller, “Cyber threats facing [38] X. Ge, Q.-L. Han, J. Wang, and X.-M. Zhang, “ Scalable and resilient
autonomous and connected vehicles: Future challenges,” IEEE Trans. platooning control of cooperative automated vehicles,” IEEE Trans.
Intell. Transp. Syst., vol. 18, no. 11, pp. 2898–2915, Nov. 2017. Veh. Technol., vol. 71, no. 4, pp. 3595–3608, Apr. 2022.
[18] M. Sun, A. Al-Hashimi, M. Li, and R. Gerdes, “Impacts of constrained [39] K. Gu, V. Kharitonov, and J. Chen, Stability of Time-Delay Systems.
sensing and communication based attacks on vehicular platoons,” IEEE Boston, USA: Birkhäuser, 2003.
Trans. Vech. Technol., vol. 69, no. 5, pp. 4773–4787, May 2020.
[40] E. Wood, E. Burton, A. Duran, and J. Gonder, “ Appending high-
[19] S. Xiao, X. Ge, Q.-L. Han, and Y. Zhang, “ Secure and collision-free resolution elevation data to GPS speed traces for vehicle energy
multi-platoon control of automated vehicles under data falsification modeling and simulation,” National Renewable Energy Laboratory,
attacks,” Automatica, vol. 145, p. 110531, Nov. 2022. Golden, USA, Technical Report NREL/TP-5400-61109, 2014,
[20] S. Ucar, S. Ergen, and O. Ozkasap, “ IEEE 802.11p and visible light [41] D. Swaroop and K. Hedrick, “ String stability of interconnected
hybrid communication based secure autonomous platoon,” IEEE Trans. systems,” IEEE Trans. Autom. Control, vol. 41, no. 3, pp. 349–357, Mar.
Vech. Technol., vol. 67, no. 9, pp. 8667–8681, Sept. 2018. 1996.
[21] F. van Wyk, Y. Wang, A. Khojandi, and N. Masoud, “Real-time sensor [42] J. Monteil, M. Bouroche, and D. Leith, “ L2 and L p stability analysis of
anomaly detection and identification in automated vehicles,” IEEE heterogeneous traffic with application to parameter optimization for the
Trans. Intell. Transp. Syst., vol. 21, no. 3, pp. 1264–1276, Mar. 2020. control of automated vehicles” IEEE Trans. Control Syst. Technol.,
[22] Y. Wang, N. Masoud, and A. Khojandi, “ Real-time sensor anomaly vol. 27, no. 3, pp. 934–949, May 2019.
detection and recovery in connected automated vehicle sensors,” IEEE [43] E. van Nunen, J. Reinders, E. Semsar-Kazerooni, and N. van de Wouw,
Trans. Intell. Transp. Syst., vol. 22, no. 3, pp. 1411–1421, Mar. 2021. “String stable model predictive cooperative adaptive cruise control for
[23] Z. Biron, S. Dey, and Pisu, “ Real-time detection and estimation of heterogeneous platoons,” IEEE Trans. Intell. Veh., vol. 4, no. 2, pp. 186–
denial of service attack in connected vehicle systems,” IEEE Trans. 196, Jun. 2019.
Intell. Transp. Syst., vol. 19, no. 12, pp. 3893–3902, Dec. 2018. [44] Q. Deng, “A general simulation framework for modeling and analysis of
[24] E. Mousavinejad, F. Yang, Q.-L. Han, X. Ge, and L. Vlacic, “Distribu- heavy-duty vehicle platooning,” IEEE Trans. Intell. Transp. Syst.,
ted cyber attacks detection and recovery mechanism for vehicle vol. 17, no. 11, pp. 3252–3262, Nov. 2016.
platooning,” IEEE Trans. Intell. Transp. Syst., vol. 21, no. 9, pp. 3821– [45] X. Ge, Q.-L. Han, X.-M. Zhang, L. Ding, and F. Yang, “Distributed
3834, Sept. 2020. event-triggered estimation over sensor networks: A survey,” IEEE
[25] A. Alipour-Fanid, M. Dabaghchian, and K. Zeng, “Impact of jamming Trans. Cybern., vol. 50, no. 3, pp. 1306–1320, Mar. 2020.
attacks on vehicular cooperative adaptive cruise control systems,” IEEE [46] H. Geng, H. Liu, L. Ma, and X. Yi, “Multi-sensor filtering fusion meets
Trans. Vech. Technol., vol. 69, no. 11, pp. 12679–12693, Nov. 2020. censored measurements under a constrained network environment:
[26] A. Petrillo, A. Pescapé, and S. Santini, “ A secure adaptive control for Advances, challenges and prospects,” Int. J. Syst. Sci., vol. 52, no. 16,
cooperative driving of autonomous connected vehicles in the presence pp. 3410–3436, Dec. 2021.
of heterogeneous communication delays and cyberattacks,” IEEE Trans. [47] J. Sarangapani, Neural Network Control of Nonlinear Discrete-Time
Cybern., vol. 51, no. 3, pp. 1134–1149, Mar. 2021. Systems. Boca Raton, USA: CRC Press, 2006.
[27] D. Zhang, Y.-P. Shen, S.-Q. Zhou, X.-W. Dong, and L. Yu, [48] A. Seuret and E. Fridman, “ Wirtinger-like Lyapunov-Krasovskii
“Distributed secure platoon control of connected vehicles subject to functionals for discrete-time delay systems,” IMA J. Math. Control Inf.,
DoS attack: Theory and application,” IEEE Trans. Syst., Man, Cybern., vol. 35, no. 3, pp. 861–876, Sept. 2018.
Syst., vol. 51, no. 11, pp. 7269–7278, Nov. 2021.
[28] Z. Li, B. Hu, and Z. Yang, “ Co-design of distributed event-triggered
controller for string stability of vehicle platooning under periodic Xiaohua Ge (Senior Member, IEEE) received the
jamming attacks,” IEEE Trans. Vech. Technol., vol. 70, no. 12, pp. 13115– Ph.D. degree in computer engineering from Central
13128, Dec. 2021. Queensland University, Australia in 2014. From
2015 to 2017, he was a Research Fellow with the
[29] Y. Harfouch, S. Yuan, and S. Baldi, “ An adaptive switched control Griffith School of Engineering, Griffith University,
approach to heterogeneous platooning with intervehicle communication Australia. He is currently a Senior Lecturer with the
losses,” IEEE Trans. Control Netw. Syst., vol. 5, no. 3, pp. 1434–1444, School of Software and Electrical Engineering,
Sept. 2018. Swinburne University of Technology, Australia. His
[30] G. Guo, J. Kang, H. Lei, and D. Li, “ Finite-time stabilization of a research interests include networked, secure, and
collection of connected vehicles subject to communication intelligent control and estimation theories, and their
interruptions,” IEEE Trans. Intell. Transp. Syst., vol. 23, no. 8, pp. 10627– applications in electric vehicles, intelligent robotic vehicles, connected auto-
10635, Aug. 2022. mated vehicles, intelligent transportation systems and so on.
[31] S. Xiao, X. Ge, Q.-L. Han, and Y. Zhang, “Secure distributed adaptive Dr. Ge is an Associate Editor of IEEE Transaction on Systems, Man, and
platooning control of automated vehicles over vehicular ad-hoc Cybernetics: Systems, IEEE Transactions on Intelligent Vehicles, IEEE
networks under denial-of-service attacks,” IEEE Trans. Cybern., vol. 52, Transactions on Circuits and Systems —II: Express Briefs, and IEEE/CAA
no. 11, pp. 12003–12015, Nov. 2022. Journal of Automatica Sinica.
[32] R. Merco, F. Ferrante, and P. Pisu, “ A hybrid controller for DOS-
resilient string-stable vehicle platoons,” IEEE Trans. Intell. Transp. Qing-Long Han (Fellow, IEEE) received the B.Sc.
Syst., vol. 22, no. 3, pp. 1697–1707, Mar. 2021. degree in mathematics from Shandong Normal Uni-
[33] X. Ge, S. Xiao, Q.-L. Han, X.-M. Zhang, and D. Ding, “Dynamic event- versity, in 1983, and the M.Sc. and Ph.D. degrees in
triggered scheduling and platooning control co-design for automated control engineering from East China University of
vehicles over vehicular ad-hoc networks,” IEEE/CAA J. Autom. Sinica, Science and Technology, in 1992 and 1997, respec-
vol. 9, no. 1, pp. 31–46, Jan. 2022. tively. He is Pro Vice-Chancellor (Research Quality)
[34] X. Ge, Q.-L. Han, J. Wang, and X.-M. Zhang, “ A scalable adaptive and a Distinguished Professor at Swinburne Univer-
approach to multi-vehicle formation control with obstacle avoidance,” sity of Technology, Australia. He held various aca-
IEEE/CAA J. Autom. Sinica, vol. 9, no. 6, pp. 990–1004, Jun. 2022. demic and management positions at Griffith Univer-
sity and Central Queensland University, Australia.
[35] J. Ploeg, D. Shukla, N. van de Wouw, and H. Nijmeijer, “Controller His research interests include networked control systems, multi-agent sys-
synthesis for string stability of vehicle platoons,” IEEE Trans. Intell. tems, time-delay systems, smart grids, unmanned surface vehicles, and neural
Transp. Syst., vol. 15, no. 2, pp. 854–865, Feb. 2014. networks.
[36] S. Stüdli, M. Seron, and R. Middleton, “ From vehicular platoons to Professor Han was awarded The 2021 Norbert Wiener Award (the High-
general networked systems: String stability and related concepts,” Annu. est Award in systems science and engineering, and cybernetics) and The 2021

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.
GE et al.: RESILIENT AND SAFE PLATOONING CONTROL OF CAVS AGAINST INTERMITTENT DOS ATTACKS 1251

M. A. Sargent Medal (the Highest Award of the Electrical College Board of multiobjective optimization, dynamics control, and their applications in road
Engineers Australia). He was the recipient of The 2021 IEEE/CAA Journal of and railway vehicles.
Automatica Sinica Norbert Wiener Review Award, The 2020 IEEE Systems, Dr. Wu is the recipient of the Discovery Early Career Award funded by
Man, and Cybernetics (SMC) Society Andrew P. Sage Best Transactions the Australian Research Council. He is on editorial boards of Journal of Traf-
Paper Award, The 2020 IEEE Transactions on Industrial Informatics Out- fic and Transport Engineering, and Railway Engineering Science and a Guest
standing Paper Award, and The 2019 IEEE SMC Society Andrew P. Sage Editor for several Special Issues.
Best Transactions Paper Award.
Professor Han is a Member of the Academia Europaea (The Academy of
Europe). He is a Fellow of The International Federation of Automatic Con- Xian-Ming Zhang (Senior Member, IEEE) received
trol (IFAC) and a Fellow of The Institution of Engineers Australia (IEAust). the M.Sc. degree in applied mathematics and the
He is a Highly Cited Researcher in both Engineering and Computer Science Ph.D. degree in control theory and control engineer-
(Clarivate Analytics). He has served as an AdCom Member of IEEE Indus- ing from Central South University, in 1992 and 2006,
trial Electronics Society (IES), a Member of IEEE IES Fellows Committee, a respectively. From 2007 to 2014, he was a Post-Doc-
Member of IEEE IES Publications Committee, and Chair of IEEE IES Tech- toral Research Fellow and a Lecturer with the School
nical Committee on Networked Control Systems. Currently, he is Editor-in- of Engineering and Technology, Central Queensland
Chief of IEEE/CAA Journal of Automatica Sinica, Co-Editor-in-Chief of University, Australia. From 2014 to 2016, he was a
IEEE Transactions on Industrial Informatics, and Co-Editor of Australian Lecturer with the Griffith School of Engineering,
Journal of Electrical and Electronic Engineering. Griffith University, Australia. In 2016, he joined the
Swinburne University of Technology, Australia, where he is currently an
Associate Professor with the School of Science, Computing and Engineering
Qing Wu (Member, IEEE) received the B.Eng. Technologies. His current research interests include H-infinity filtering,
degree in vehicle engineering (mechanical) from event-triggered control systems, networked control systems, neural networks,
Southwest Jiaotong University, in 2010, the M.Eng.
multi-agent systems, offshore platforms and time-delay systems.
degree in vehicle engineering (mechanical) from
Associate Professor Zhang was a recipient of second National Natural Sci-
State Key Laboratory of Traction Power, in 2012,
and the Ph.D. degree in engineering (multiobjective ence Award in China in 2013, jointly with Prof. M. Wu and Prof. Y. He. He
optimization and parallel computing) from Central was also the recipient of The 2021 IEEE/CAA Journal of Automatica Sinica
Queensland University, Australia in 2016. From Nobert Wiener Review Award, The 2020 IEEE Transactions on Industrial
2012 to 2013, he was a Research Assistant with the Informatics Outstanding Paper Award, The 2019 IEEE Systems, Man, and
State Key Laboratory of Traction Power, Southwest Cybernetics Society Andrew P. Sage Best Transactions Paper Award, and The
Jiaotong University. From 2016 to 2018, he was a Senior Research Officer 2016 IET Control Theory and Applications Premium Award. He is an Asso-
with the School of Engineering and Technology, Central Queensland Univer- ciate Editor of the IEEE Transactions on Cybernetics, Journal of the Franklin
sity, Australia. He is currently a Research Fellow with the same school. His Institute, International Journal of Control, Automation, and Systems, Neuro-
research interests include mechanical system dynamics, parallel computing, computing, and Neural Processing Letters.

Authorized licensed use limited to: CQUniversity. Downloaded on September 03,2023 at 05:53:29 UTC from IEEE Xplore. Restrictions apply.

You might also like