This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 1

ERTMS/ETCS Virtual Coupling: Proof of

Concept and Numerical Analysis
Carlo Di Meo, Marco Di Vaio , Francesco Flammini , Senior Member, IEEE, Roberto Nardone ,
Stefania Santini, Member, IEEE, and Valeria Vittorini

Abstract— Railway infrastructure operators need to push their step in this direction is to increase rail capacity by introducing
network capacity up to their limits in high-traffic corridors. moving block signalling. Some pilot ERTMS installations exist
Virtual coupling is considered among the most relevant innova- implementing moving block signalling although such experi-
tions to be studied within the European Horizon 2020 Shift2Rail
Joint Undertaking as it can drastically reduce headways and thus ences are very limited to date [3]. The Shift2Rail IP2 addresses
increase the line capacity by allowing to dynamically connect research in the field of moving block signalling and also
two or more trains in a single convoy. This paper provides a goes a step further by introducing the concept of Virtual
proof of concept of Virtual coupling by introducing a specific Coupling.
operating mode within the European rail traffic management Virtual Coupling could be a ground-breaking innovation in
system/European train control system (ERTMS/ETCS) standard
specification, and by defining a coupling control algorithm railways as it enables coupling/uncoupling of train convoys
accounting for time-varying delays affecting the communication by drastically reducing headways. That opens a completely
links. To that aim, we define one ploy to enrich the ERTMS/ETCS new scenario in which cooperative driving could be exploited
with Virtual coupling without changing its working principles based on Train-to-Train communication and distributed
and we borrow a numerical analysis methodology used to study control, in order to achieve a global coordination of virtually-
platooning in the automotive field. The numerical analysis is also
provided to support the proof of concept with quantitative results coupled convoys via wireless communication. The contribu-
in a case-study simulation scenario. tion of this paper is twofold: firstly, it provides technical hints
to introduce and evaluate Virtual Coupling in the context of
Index Terms— Railways, ERTMS/ETCS, automatic train
control, virtual coupling, simulation, numerical analysis. ERTMS; secondly, it provides a numerical analysis of rail
vehicle cooperation and platooning strategies borrowed from
I. I NTRODUCTION the automotive domain in order to perform a preliminary eval-
uation of feasibility and performance. In this study, we assume
T HE European Railway Traffic Management System/
European Train Control System (ERTMS/ETCS, or sim-
ply ERTMS hereinafter) [1] is an international standard aimed
to enrich ERTMS with Virtual Coupling instead of defining a
fully new signalling system, which is the preferred approach
at improving interoperability and performance of modern to ensure backward compatibility and minimize the impact on
railways. After several years of experience with ERTMS existing infrastructures since it guarantees the re-use of stan-
development, the railway industry is tackling the challenge dard operating modes and related safety mechanisms. Based
of improving the standard and designing the new generation on that assumption, we define and evaluate a coupling control
railway control systems. In order to foster coordination of strategy that is robust against communication impairments.
research and innovation efforts, the European Shift2Rail Joint The closed-loop exponential stability is analytically proved
Undertaking has been established under Horizon 2020 [2]. for this control strategy by exploiting a proper Lyapunov-
Shift2Rail has devoted an entire pillar of its programme Krasovskii functional that extends to a network of dynamical
(i.e. IP2, advanced Traffic Management and Control System) systems in presence of multiple delays the approach proposed
to enhance capacity and reliability through improved manage- in [4] for a single time-delayed system. A realistic case study
ment of signalling and supervision systems. One important is introduced to perform simulations and provide some insights
about the possibility of actuating platoons of trains. Although
Manuscript received May 31, 2018; revised December 21, 2018 and
April 16, 2019; accepted May 17, 2019. This work was partially supported the evaluation results reported in this paper are dependent
by the Department of Electrical Engineering and Information Technologies on the specific simulation scenario, the methodology we
(DIETI) of the University of Naples Federico II within the project MOdel- introduce is generic and applicable to diverse train speeds and
Driven AnaLysis of Critical Industrial Systems (MODAL). The Associate Edi-
tor for this paper was X. Cheng. (Corresponding author: Roberto Nardone.) configurations.
C. Di Meo, M. Di Vaio, R. Nardone, S. Santini, and V. Vittorini are This paper is organized as follows. Section II contains a
with the Department of Electrical Engineering and Information Technology brief overview of ERTMS concepts and terminology as well
(DIETI), University of Naples Federico II, 80125 Naples, Italy (e-mail:
[email protected]; [email protected]; [email protected]; a discussion about related works. Section III discusses how
[email protected]; [email protected]). to extend ERTMS with Virtual Coupling and some reliability-
F. Flammini is with the Department of Computer Science and related implications. Section IV introduces the coupling con-
Media Tech-nology, Linnaeus University, 351 95 Växjö, Sweden (e-mail:
[email protected]). trol strategy for cooperative driving and platooning in ERTMS.
Digital Object Identifier 10.1109/TITS.2019.2920290 Section V reports simulation results in the example scenario.
1524-9050 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 1. (a) ERTMS Level 2 - Fixed Block; (b) ERTMS Level 3 - Moving Block; (c) ERTMS Level 3 + Virtual Coupling.

Finally, Section VI provides some closing remarks and hints and braking curves. The speed profile is used to set train
about future developments. maximum speed depending on forward track conditions (speed
restrictions, distance from the preceding train, etc.).
II. BACKGROUND AND R ELATED W ORK At ERTMS Level 3, the MA is computed by considering the
A. ERTMS Levels actual distance between trains, while at Level 2 the occupation
status of the so-called track circuits must be detected (see
The ERTMS standard defines the functional architecture,
Fig. 1a and Fig. 1b). Hence, Level 3 allows the removing of
technologies and message format to ensure safe communi-
the track segmentation into fixed blocks. It is rather intuitive
cation between train onboard and trackside equipment. The
that Level 3 provides higher capacity compared to Level 2 due
specifications identify three levels of growing complexity that
to shorter headways, at the expense of additional train integrity
can be implemented singularly or jointly.
check mechanisms. The new paradigm of Virtual Coupling we
• ERTMS Level 1. At this level, communication is dis-
address in this paper has been associated by some analysts
continuous and based on signals transmitted by the
to a future ERTMS Level 4 [5], although there is currently
so-called Euro-balises that are devices installed between
no reference to that level in the official ERTMS documents.
track lines. Some additional mechanisms (i.e., infill) can
Virtual Coupling allows trains to join and increase line capac-
be used to make signalling semi-continuous, but ERTMS
ity due to extremely short headways (Fig. 1c). In this paper,
Level 1 cannot guarantee the capacity of higher levels
we suggest a simple way of implementing Virtual Coupling
due to technology limitations, and it is therefore out of
in the context of ERTMS by assigning the following train
the scope of this work.
a release speed at Veoa that is equal to the speed of the
• ERTMS Level 2. This level implements continuous radio
preceding train. Such a mechanism can be managed by the
signalling based on the EURORADIO safe protocol via
RBC based on the position reports received from the trains
the Global System for Mobile Communications - Railway
and including current train speeds. Together with the need for
(GSM-R). The exact positions of the trains and all the
the specification of a further ERTMS level and/or on-board
other necessary information are automatically transmitted
operating mode, the real implementation of Virtual Coupling
at periodical intervals and/or upon events. The movement
poses additional challenges in terms of reliability and safety
authority is transmitted to the train at any time together
that need to be investigated extensively before such a system
with all the information related to speed profiles associ-
can become operational.
ated to track characteristics, conditions, speed restrictions,
etc.
• ERTMS Level 3 implements moving block instead of B. Related Works
fixed block signalling. This allows for the definition of Virtual Coupling is a recent concept that is still far from
movement authorities regardless of fixed block sections being implemented and as such it represents an open research
according to the traffic conditions. As a consequence, this challenge in the railway domain. The transition towards
level is able to maximize track capacity. ERTMS Level 3 moving block signalling has been very slow
The trackside controller responsible for train separation is due to its benefits probably not justifying transition costs and
named Radio Block Center (RBC). The distance each train additional risks to be managed. A number of solutions are
can cover in safe conditions is named Movement Author- currently being developed at different levels and the expected
ity (MAs) and it is computed by the RBC based on train benefit of adding Virtual Coupling on top of ERTMS Level 3 is
position and track conditions. The end of the MA is named considered crucial for the future implementation of the stan-
end of authority (eoa) and the allowed speed at the end of dard [3]. A few recent reports from international railway orga-
authority is named Veoa. MAs allow the on-board European nizations and professional institutions introduce the idea of
Vital Computer (EVC) to compute the dynamic speed profile Virtual Coupling in general terms [2], [5]. The Shift2Rail Joint
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

DI MEO et al.: ERTMS/ETCS VIRTUAL COUPLING: PROOF OF CONCEPT AND NUMERICAL ANALYSIS 3

Undertaking Multi-Annual Action Plan [2] also states that

“virtual trains” represent “a total deviation from the traditional
railway operational concept” and proposes that any investiga-
tion “starts from the current interoperable signalling system”
(i.e., ERTMS). The report published by IRSE (Institution of
Railway Signal Engineers) [5] discusses the Virtual Coupling
concept and poses some questions about its safety and benefits.
A few recent works address the technologies required for the
implementation of Virtual Coupling (for example see [6]).
Virtual Coupling heavily depends on wireless communication,
hence works addressing innovative technologies for railway Fig. 2. Extended ERTMS Level 3 operating modes.
communications also mention Virtual Coupling and Train-
to-Train communication as promising applications (e.g., see
[7]–[10]). To the best of our knowledge, no studies have
been published yet which provide hints for implementation
of Virtual Coupling on top of ERTMS Level 3 leveraging
on existing ERTMS “language” (variables, packets, messages)
and evaluate feasible control strategies in realistic scenarios,
as we do in this paper. However, a new transportation paradigm
is emerging for connected ground vehicles, such as cars,
moving according to platoon-based driving patterns. To cope
with all the challenges and uncertainties arising from vehicle
cooperation, platooning has already been investigated from
different control perspectives in road transportation scenarios
(e.g., see [11]–[14] and reference therein for a survey of
the topic including communication constraints and related
hardware solutions). The analysis performed in this paper
also investigates the portability of some results achieved in Fig. 3. Theoretical performability model.
cooperative driving and car platooning to the railway domain.
Safety and reliability issues are only partially addressed by establishment of Train-To-Train communication, etc. It is
this study, as additional research is needed to consider all the not in the scope of this paper to define all the additional
relevant factors [5]. FSVC safety requirements that should derive from the hazard
analysis. If the infrastructure cannot guarantee all the required
III. V IRTUAL C OUPLING IN ERTMS safety conditions, then ERTMS must fall back to FS, provided
that its conditions are fulfilled, otherwise it must switch to PS
A. Extending the On-Board Operating Modes (see Fig. 2). No direct switching from PS to FSVC should be
We have mentioned that ERTMS Level 3 is suitable to possible. In FSVC, the EVC should always apply the most
achieve very high capacity on high traffic corridors; however, restrictive speed profile (MRSP) between the one provided by
its upgrade to Virtual Coupling while keeping backward the RBC and the one provided by the preceding train through
compatibility with existing infrastructures would ensure their train-to-train communication. Hence, one of the advantages of
full exploitation. In ERTMS, Operating Modes are defined for the approach based on the additional FSVC operating mode is
the on-board European Vital Computer (EVC) according to that the underlying safety and reliability conditions ensured by
the status of the track in order to guarantee adequate safety ERTMS remain unchanged. In FSVC, all the safety conditions
levels [1]. The Full Supervision operating mode is enabled related to trackside status, like speed restrictions, emergencies,
when all information data required for complete train supervi- route failures, etc., are still checked and supervised by the
sion is available. Partial Supervision (PS) is used when some RBC. Regarding MA and MRSP computation derived by train-
information is missing and includes the specific Shunting, to-train communication, it is expected that a critical factor
On-Sight and Staff Responsible modes. For the purposes of will be the capacity of the system to quickly and regularly
this discussion, we will only distinguish between FS and PS. update MA to trains so that they will not be obliged to slow
As already mentioned, in this study we assume to implement down and disconnect from the preceding trains. Therefore,
Virtual Coupling by adding a further ERTMS Level 3 Operat- if the communication infrastructure ensures enough throughput
ing Mode on top of Full Supervision. The new Operating Mode and reliability such that the T _N V C O N T AC T timer used
would be named “Full Supervision plus Virtual Coupling” for vitality checking can be set to sufficiently low values
(FSVC). In order to switch from FS to FSVC, the ERTMS (below 1s), then safety concerns will be automatically cleared.
infrastructure should check and guarantee that all the necessary
conditions are fulfilled, including sufficient MA assigned to the B. Performability Issues
preceding train, sufficient MA overlap between preceding and The transitions among the ERTMS operating modes dis-
following train to ensure same routing in stations, successful cussed above can be modeled by the Petri Net in Fig. 3.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

in FSVC most of the time, the graph shows a value close to

100 trains per hour, that is one train each 36 seconds as per
our calculations. Any other combinations of mode transition
frequencies are possible: for instance, if FSVC is kept very
reliably once reached a stationary condition, but the coupling
procedure is very slow, then the situation would be the one
of the extreme left corner of the graph: around 60 trains per
hour that is 1 train per minute, still 3 times better than FS.
A similar result (see extreme right corner of the graph) holds
if we have an efficient yet unstable transition to FSVC, with
trains easily getting from FS to FSVC but also leaving FSVC
often, e.g. due to communication faults.

C. The Need for Train-to-Train Communication

If there were no limitations on the performance and reli-
ability of communication channel between train and track-
Fig. 4. Results of FSVC performability evaluation.
side, then ERTMS Level 3 could support Virtual Coupling
to a certain extent by simply setting the Veoa variable
equal to the speed of the preceding train, as aforementioned.
The delays associated to the stochastic transitions representing
However, as trains get closer (see Fig. 1 (c)), the infrastructure
mode changes have a negative-exponential distribution. That
would need to send MA and any emergency messages at
assumption is realistic since those transitions represent time-
ultra-high speeds in order to ensure safe latencies due to
to-fail (i.e., moving from higher to lower supervision) and
extremely short headways. Consequently, engineers should
time-to-repair (i.e., moving from lower to higher supervision)
reduce the T _N V C O N T AC T vitality timer in proportion to
whose stochastic variables are demonstrated in the reliability
the headway reduction. Today, that timer is set to few seconds
theory to be well-modeled by negative exponential distribu-
to get a good trade-off between safety and performance.
tions. The transitions T_FSVC and T_FS have deterministic
In absence of Train-to-Train communication, the timer should
firing rates as they represent the frequency of trains stably
be set to fractions of second, scaling down one or two orders
running in FSVC and FS, respectively. The places P_T and
of magnitude. That is clearly not compatible with current
P_R and the transition T have been added to evaluate the
communication technology used in ERTMS.
train frequency as a function of sojourn times in FSVC, FS
Hence, in the rest of this paper we assume that Train-to-
and PS. Specifically, the throughput of transition T provides
Train communication is provided by means of Wi-Fi connec-
an upper bound to the number of trains the system is able to
tion or next-generation networks (e.g., 5G).
manage in the time unit (i.e., theoretical system capacity) as
Together with train mass and braking capability, actual
a function of model parameters. As an example, if we assume
Virtual Coupling headways will be dependent on train
the trains running at 300 km/h on the Milano-Bologna high-
positioning system precision (e.g. odometer error or satellite
speed ERTMS track in Italy, with realistic MAs at full speed
positioning resolution in future generation ERTMS [15]),
of 15 km, then the maximum frequency would be 15/300 =
which has an influence on the entity of safe margins
0.05 h (i.e., 3 minutes minimum dispatching time).
(i.e. Minimum/Maximum safe Rear/Front End), and on the
At FSVC, the distance between trains will be given by
Train-to-Train connection performance (including bandwidth
the train minimum/maximum safe rear/front ends that are the
and stability) together with the availability of appropriate
train extremes augmented by the odometric errors. We assume
sensors/radars.
an inter-distance of 3 km at 300 km/h, which correspond to
a frequency of 3/300 = 0.01 h (i.e., 36 seconds minimum
IV. C OUPLING S TRATEGY
dispatching time).
Performability model evaluation results are shown in Fig. 4. A. Communication Topology and Preliminaries
Without FSVC, the T frequency would be 20 trains per hour As justified in the previous Section, we assume that trains
that is one train each 3 minutes. Such theoretical minimum are equipped with Train-to-Train communication interfaces to
is the closest corner in the graph (i.e., lowest point), where share information on their current state (position, velocity, and
the rate of the transition towards FS is higher and the rate acceleration). We also assume that the RBC acts as virtual
of the transition towards FSVC is lower: that corresponds leader by dispatching the reference behavior to the other trains
to maximum time spent in FS due to improper implemen- under its supervision.
tation of VC or faulty train-to-train communication channel. The communication network established among N trains
On the opposite, with a low rate of faulty conditions in can be modeled by a direct graph G N = (V N , E N , A N ), where
FSVC, we would be closer to the farther corner in the graph V N = {1, . . . , N} is the set of nodes (trains), and E N ⊆ V N ×
(i.e., highest point), where the rate of the transition towards V N the set of edge (communication links among trains). The
FSVC is very high while the rate of the transition towards topology of the graph is associated to the adjacency matrix
FS is much lower. In such a condition, where the trains stay A N = [ai j ] N×N encoding train communication relationship,
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

DI MEO et al.: ERTMS/ETCS VIRTUAL COUPLING: PROOF OF CONCEPT AND NUMERICAL ANALYSIS 5

where ai j = 1 if (i, j ) ∈ E N , and ai j = 0 otherwise. Note that, the information received from neighbors and infrastructure.
aii = 0 since the self-loops (i, i ) are not allowed. Moreover, The control protocol embeds the error w.r.t. the reference
definingthe degree matrix as  = diag{1 , . . . ,  N }, with behavior that trains are forced to reach and keep during
i = j ∈V N ai j , the Laplacian matrix of G N is defined a the coupling, as well as the time-varying communication
L N =  − A N . Furthermore, a path of length η in a graph delays, as:
is an ordered sequence of η + 1 vertices such that any pair of 1 N
consecutive vertices in the sequence is an edge of the graph. ui = − ki j αi j [ri (t)−r j (t −τi j (t))−h i j v 0 − di0
st
]
i j =0
If there exists a path from node i to node j , we say that j is − bi ρi0 [v i (t) − v 0 ] − γi ρi0 ai (t)
reachable from i . 1 N
We consider N trains plus an additional agent representing + ki j αi j τi j (t)v 0 (3)
i j =0
the RBC labeled with the index zero, i.e. node 0. Then,
the resulting network topology is described by a directed graph where ki j , bi and γi are the control parameters that act as
G N+1 . Finally, we assume that node 0 is globally reachable, stiffness and damping coefficients to be tuned for regulating
i.e. always there is a path in G N+1 from every node i in G N+1 the mutual behavior among neighbor trains. Since the strategy
to node 0. is designed to mitigate communication impairments, τi j (t) and
τi0 (t) are the time-varying delays affecting the communica-
tion with the i -th vehicle when information is transmitted
B. Control Design from the vehicle j and the leader, respectively. In general
The control design starts with a simplified model of the τi j (t) = τ j i (t)); h i j is the constant time headway (i.e., the time
system. Considering a group of N trains (i = 1, . . . , N) partic- necessary for vehicle i -th to travel the distance separating it
ipating to the coupling, we denote the kinematics parameters from its predecessor); distj is the distance between vehicles i
of the i -th train moving along the track as position ri [m], and j at standstill. Notoriously [19], delays are assumed to
velocity v i [m/s], acceleration ai [m/s2 ] and jerk je i [m/s3 ] be bounded and detectable. Specifically, we suppose that each
measured w.r.t. a railway reference frame. The dynamics can train transmits information with a timestamp (i.e., the time
be expressed in the state space as the following third order instant when the information is sent) so that the communica-
model [16]: tion delay over a link can be estimated by each train when
⎧ information is received [20]–[22]. Before testing the approach
⎪
⎪ ṙi (t) = v i (t),
⎨ in realistic scenarios (Section V) we need a proof of stability
v̇ i (t) = ai (t), (1) for the coupling protocol.
⎪
⎪ 1 1
⎩ȧi (t) = je i (t) = − ai (t) + u i (t).
Ti Ti C. Closed-Loop Stability Analysis
where Ti > 0 [s] is the time constant of the drive train In this section the exponential asymptotic stability of the
depending on the train specific features, while u i is the desired consensus equilibrium exhibited by trains coupled by the con-
acceleration to be set to the i -th train to be coupled within trol action (IV) is proven. Note that the use of an exponential
a platoon. Since we are dealing with automated longitudinal criterion not only allows to prove the stability of - or around -
control, model (1) considers only the actuation lag (i.e., the the equilibrium solution, but also allows demonstrating the
actual acceleration of the train is assumed to track the desired ability in reaching the required virtual coupling regime with
acceleration with a time constant Ti ). an exponentially bounded dynamic behavior. Indeed, while the
According to the Virtual Coupling paradigm, the reference asymptotic stability of the equilibrium is significant when the
behavior ṙ0 (t) = v 0 , v̇ 0 (t) = a0 = 0 (being r0 [m], v 0 [m/s], coupled trains are moving in steady-state conditions, the expo-
and a0 [m/s2 ], reference position, velocity, acceleration) is nential stability for delayed systems provides a key ingredient
transmitted to the trains by the RBC. Hence, the Virtual for addressing transient maneuvers in presence of communi-
Coupling goal can be expressed as: cation delays. That is especially important in the crucial task
of coupling engagement (see e.g. [23], [24] for analogous
ri (t) → r0 (t) + di0 ; v i (t) → v 0 ; ai (t) → a0 ; (2) automotive applications).
where di0 = h i0 v 0 + di0 st is the desired distance of the train We define the errors w.r.t. the desired equilibrium (2) as:
st
i from the virtual leader to be set according to the spacing r̄i = ri (t) − r0 (t) − h i0 v 0 − di0 ,
policy, being h i0 the time headway (i.e., the time necessary v̄ i = v i (t) − v 0 ,
st
for the train i -th to travel the distance from the leader) and di0 āi = ai (t). (4)
is the desired distance between vehicles i and the leader 0 at By re-writing the coupling control action u i in terms of
standstill [17]. Note that, with no loss of generality, the spacing the state errors r̄i , v̄ i and āi and by expressing headway
policy can be referred to the leading reference since train
constants h i j and standstill distances distj w.r.t. the leading
separation can be expressed as the spacing w.r.t. the reference
vehicle, namely h i j = h i0 − h j 0 and distj = di0
st − d st , after
position r0 (t) [18] provided by the infrastructure. j0
some algebraic manipulation, the closed-loop dynamics for the
To achieve the Virtual Coupling goal in (2), we propose
generic i-th train within the formation can be rewritten as:
the following decentralized control strategy composed of two
terms: a local action depending on the state variables of r̄˙i (t) = v̄ i (t),
the vehicle (measured on-board) and an action depending on v̄˙i (t) = āi (t),
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

1 1 N
ā˙ i (t) = − [ ki j αi j (r̄i (t) − r̄ j (t − τi j (t))) inequality holds
Ti i j =1
1 
m 
m
+bi ρi0 v̄ i (t) + (1 + γi ρi0 )āi (t) + ki0 αi0 r̄i (t)]. [2δ P + P A
0 + P A0 + Qp + ψ R p ] < 0, (10)
i p=1 p=1
(5)
with matrix A0 defined as in (7), then the closed-loop delayed
In order to describe train network dynamics in the pres- system (5) is exponentially stable with a decay rate δ [25].
ence of time-varying delays associated to the communica- Proof 1: Consider the following Lyapunov-Krasovskii
tion links in a more compact form, we define the position, functional:
speed and acceleration error vectors as r̄ = [r̄1 , . . . , r̄ N ] , V (x t ) = V1 (x̄(t)) + V2 (x t ) + V3 (x t ), (11)
v̄ = [v̄ 1 , . . . , v̄ N ] , and ā = [ā1 , . . . , ā N ] ; the error state
vector as x̄(t) = [r̄  v̄  ā  ] and we furthermore define where x t = x̄(t + θ ), θ ∈ [−τ , 0] and
 
τ p (t), p = 1, 2, . . . , m (with m ≤ N(N − 1))) as an element V1 (x(t)) = x̄  (t)P x̄(t) ,
of the sequence {τi j (t) : i, j = 1, 2, . . . , N, i = j }. Note  
that, m is equal to its maximum, N(N − 1), only when m  t
2δ(θ−t ) 
the communication topology connecting coupled trains is V2 (x t ) = e x̄ (θ )Q p x̄(θ )dθ ,
p=1 t −τ (t )
p
represented by a directed complete graph and all time-delays  
are different. The closed-loop train network can be hence m  t
1 2δ(θ−t ) 
represented as the following set of time-delayed differential V3 (x t ) = e x̄ (θ )R p x̄(θ )dθ .
p=1 1−μ t −τ p (t )
equations:
(12)

m
˙ = A0 x̄(t) +
x̄(t) A p x̄(t − τ p (t)), (6) with μ = max p (d p ) ( p = 1, . . . , m). Following the approach
p=1 in [4], it can be verified that functional (11) satisfies Lyapunov-
Krasovskii stability condition [25].
where
⎡ ⎤ α1 x̄(t) 2
≤ V (x t ) ≤ α2 x t 2
(13)
0 N×N I N×N 0 N×N
⎢ ⎥ with
A0 = ⎣0 N×N 0 N×N I N×N ⎦ , (7)
−T K −T B −T α1 := λmin (P)
⎡ ⎤ α2 := λmax (P)
0 N×N 0 N×N 0 N×N  m 
⎢ ⎥ m
A p = ⎣0 N×N 0 N×N 0 N×N ⎦ , (8) +τ λmax (Q p ) + ψ λmax (R p ) (14)
p=1 p=1
T Kp 0 N×N 0 N×N −1
where ψ = (1 − μ) .
with 0 N×N the null matrix of dimension N × N , T = diag Now, we differentiate each of the terms in (11).
{ T11 , . . . , T1N }, T = T ,  = diag{1 + γ1 ρ10 , . . . , 1 + γ N ρ N0 }, We have:
B = diag{b1ρ10 , . . . , b N ρ N0 }, K = diag{k11 , . . . , k N N } V̇1 (x̄(t)) = x̄˙  (t)P x̄ (t) + x̄  (t)P x̄(t)
˙ (15)
(being kii = 1i Nj=0 ki j αi j ), and K p = [k̄ pi j ] ∈ R N×N which, after some algebraic manipulation, can be rewritten
( p = 1, . . . , m) as: using (6) as
⎧α k
⎪
⎪
ij ij V̇1 (x̄(t)) = x̄  (t)[ A
0 P + P A 0 ] x̄(t)
⎨  , j = i, τ p (·) = τi j (·)
i
k̄ pi j 0, (9) 
m
⎪ j = i, τ p (·) = τi j (·) +2 x̄  (t)P A p x̄(t − τ p (t)).
⎪
⎩ (16)
0, j = i. p=1
Differentiating V2 (x t ) yields:
We prove the exponential stability of the closed-loop net- m
work in face of communication delays by leveraging on a V̇2 (x t ) = −2δV2 (x t ) + x̄  (t) Q p x̄(t)
p=1
proper Lyapunov-Krasovskii functional. The approach pro- 
m
vides information about the exponential convergence of the − (1 − τ̇ p (t))x̄  (t − τ p (t))
error system (i.e. about the creation of the train platoons p=1

with a proper timing, or waiting time). The stability condi- × Q p x̄(t − τ p (t))e−2δτ p (t ) (17)
tions for the delayed network is expressed as Linear Matrix
Inequality (LMI) criterion that allows the optimization of the which, exploiting the bound on the delay functions [26], can
convergence rate. be recast as:
m
Theorem 1: Consider the delayed closed-loop error sys- V̇2 (x t ) ≤ −2δV2 (x t ) + x̄  (t) Q p x̄(t)
tem (5). Assume delays τ p (t) ( p = 1, . . . , m) to be bounded, p=1
m  
i.e. τ p (t) ∈ [0, τ ], τ̇ p (t) ∈ (−∞, d p ] (∀t ∀ p) and d p ≤ 1. − ψ −1 x̄  (t − τ p (t))Q p x̄(t − τ p (t))e−2δτ .
If constants δ > 0, ψ > 0, and matrices P = P  > 0, p=1
Q p > 0, R p > 0 ( p = 1, . . . , m) exist so that the following (18)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

DI MEO et al.: ERTMS/ETCS VIRTUAL COUPLING: PROOF OF CONCEPT AND NUMERICAL ANALYSIS 7

Analogously, by differentiating V3 (x t ) and by exploiting the From hypothesis (10), we know that matrix (22) is negative
bounds on the delay functions, we obtain: definite. Hence, from (21) we have:
m
V̇3 (x t ) ≤ −2δV3 (x t ) + ψ x̄  (t) R p x̄(t) V̇ (x t ) + 2δV (x t ) ≤ 0 ⇒ V (x t ) ≤ V (φ)e−2δt for t ≥ 0.
p=1
m   (24)
− x̄  (t − τ p (t))R p x̄(t − τ p (t))e−2δτ . (19)
p=1 By combining (13) and (24), we obtain [4]:
Therefore, by using (16), (18), (19), we obtain: α1 x̄(t; t0 , φ) 2
≤ V (x t ) ≤ V (φ)e−2δt ≤ α2 e−2δt φ 2c .
m (25)
V̇ (xt ) ≤ x̄  (t)[A  
0 P + P A 0 ] x̄(t) + x̄ (t) Q p x̄(t)
p=1
m Hence,
+2x̄  (t)P A p x̄(t − τ p (t)) − 2δV2 (xt ) 
p=1 α2 −δt
m   x̄(t; t0 , φ) ≤ e φ c for t ≥ 0. (26)
− ψ −1 x̄  (t − τ p (t))Q p x̄(t − τ p (t))e−2δτ α1
p=1
m Therefore, the statement is proven.
−2δV3 (xt ) + ψ x̄  (t) R p x̄(t) Finally, we remark that the LMI (10) that provides the
p=1
m   stability region for the control gains is feasible and can be
− x̄  (t − τ p (t))R p x̄(t − τ p (t))e−2δτ . solved numerically, e.g. by a classical interior-point algorithm.
p=1
Furthermore, for any given control gain set, by leveraging on
Summing and subtracting 2δV1 (x(t)) from the right, after
the LMI-based tuning procedure for exponential stability it
some algebraic manipulation, we get:
is possible to evaluate the achievable decay rate and hence
V̇ (xt ) + 2δV (xt ) to design the controller to ensure the convergence rate to the
closed-loop dynamics.
≤ x̄  (t)[A 
0 P + P A 0 ] x̄(t)
m

+2x̄ (t)P A p x̄(t − τ p (t)) + 2δ x̄  (t)P x̄(t) V. N UMERICAL A NALYSIS
p=1
m m In this section, we apply the proposed Virtual Coupling
+x̄  (t) Q p x̄(t) + ψ x̄  (t) R p x̄(t) algorithm to a realistic case study. The goal is to provide a
p=1 p=1
m   first insight on the possibility of reaching and keeping the
− ψ −1 x̄  (t − τ p (t))Q p x̄(t − τ p (t))e−2δτ reference behavior set by the virtual leader so that all coupled
p=1
m   trains can safely move together with a desired velocity and
− x̄  (t − τ p (t))R p x̄(t − τ p (t))e−2δτ . tight spacing. The reference scenario is depicted in Fig. 5.
p=1
We consider a single RBC (labeled as 0) and three trains with
By defining a new augmented delayed state vector as:
the same mechanical features (numbered from 1 to 3) under
ξ(t) = [x̄(t), x̄ (t − τ1 (t)), . . . , x̄(t − τm (t))] (20) its supervision.
Although the on-board control design leverages a simplified
it is possible to recast inequality (20) as: linear model of the system to be controlled (namely the one
in (1)) to limit control complexity for its real-time imple-
V̇ (ξ ) + 2δV (ξ ) ≤ ξ  (t)ξ(t), (21) mentation, we use a more accurate model of train dynamics.
The objective is to test the robustness of the control approach
where
⎡ ⎤ w.r.t. model uncertainties or unknown dynamics (e.g., friction,
1 2P A1 ··· · · · 2P Am aerodynamic drag, saturation in the maximum acceleration and
⎢ 03N×3N 2 03N×3N · · · 03N×3N ⎥ traction force, etc.).
⎢ ⎥
⎢ .. .. ⎥ The dynamic model describing the motion of the i -th train,
⎢ .. ⎥
⎢ . 03N×3N 3 . . ⎥ assuming train position varying along the route, is given by
=⎢ ⎥ (22)
⎢ . . .. .. ⎥ the following nonlinear system [27]:
⎢ .. .. . . 03N×3N ⎥
⎢ ⎥
⎣ ⎦ m i ai (t) = Fi − R(v i (t)) − Fg (ri (t)) − Fc (ri (t)), (27)
.. ..
03N×3N ··· . . m+1
where ri , v i and ai are the train position, velocity and
with acceleration, respectively; m i is the train mass; Fi is the
m m bounded tractive or braking force that the traction control
1 = 2δ P + P A
0 + P A0 + Qp + ψ Rp, have to impose to the i -th train for tracking the desired
p=1 p=1
acceleration profile computed by the coupling protocol u i (t)
2 = −e−2δτ (ψ Q 1 + R1 ), in (3); R(v i ) is the resistive force depending on rolling and
.. aero-dynamic resistances, expressed as quadratic function of
. the velocity according to some empirical parameters known
m+1 = −e−2δτ (ψ Q m + Rm ), (23) in literature as the Davis parameters, i.e., R(v i ) = A +
B|v i | + Cv i2 ; the gradient and curve forces, Fg (ri ) and Fc (ri ),
being 03N×3N the null matrix of dimension 3N × 3N. are described respectively as Fg (ri ) = m i gsin(α(ri )) and
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

TABLE I
N UMERICAL A NALYSIS PARAMETERS [28]–[32]

Fig. 5. Virtual Coupling of three trains: (a) Example scenario. (b) Commu-
nication topology.

sends the reference speed v 0 = 210 [km/h] to the three trains,

Fc (ri ) = m i gk(ri )/ρ(ri ), being g the gravity acceleration, which in turn share their current position, speed and accelera-
while k, ρ, α are empirical parameters whose values depend on tion via Train-to-Train communication according to a Leader-
route layout, i.e. curvature radius and track gauge. Parameter Predecessor-Follower (LPF) topology as in Fig. 5b. Results
α can be positive (i.e. ascent) or negative (i.e., descend). in Fig. 6 show that after a transient period the three trains are
Train and control parameters are reported in Table I. The effectively coupled and equidistant, according to the prescribed
mechanical characteristics (i.e., maximum speed, acceleration, spacing policy, and afterward they all run at the steady-state
deceleration, length and mass) are referred to the ETR class reference speed set by the RBC (see Fig. 6b). As the control
of locomotives commonly used on ERTMS lines in Italy (e.g., gains have been tuned following the LMI- based exponential
see [28]–[31] and references therein). Other parameters (i.e., stability criterion (see parameters in Table I), the trajectories
reference coupling-speed, headway and distance at standstill) are upper bounded by exponential envelopes, whose decay rate
depend on the specific application and have to be considered can be estimated according to (26), as shown in Fig. 6a, where
as an initial test data since no similar studies have been the time history of position errors is reported. Results show
performed yet for railway applications (see e.g. [32] for that according to the theoretical derivation all trains are able to
automotive applications). Nevertheless, the scenario considers reach and keep the desired formation in 1500 [s] with bounded
a challenging coupled motion requiring very tight bumper-to- transient dynamics.
bumper gaps. The numerical simulations have been performed
in MATLAB , c while the Yalmip  c Toolbox [33] has been
B. Coordinate Emergency Braking Maneuver
used to tune control gains by solving the LMIs in (10). The
multiple time-varying delays along the communication links Results reported in Fig. 7 represent a more challenging
have been emulated as stochastic variables with a uniform situation when trains are moving as coupled with short gaps
discrete distribution, τ p (t) ∈ [0, τ ] ( p = 1, . . . m) where the and an unpredictable transient maneuver is needed, e.g. emer-
maximum value is set above the average end-to-end communi- gency braking. In Virtual Coupling such manouvers must be
cation delay typical of IEEE 802.11p vehicular networks [34], executed in a coordinated fashion so to keep trains coupled
i.e., τ = 15 · 10−2 [s]. while the inter-train distance dynamically changes. Therefore,
the spacing policy must be automatically adapted to velocity
variations to achieve cooperative coordinate deceleration. The
A. Virtual Coupling Engagement example scenario considers three coupled trains moving on a
In order to test and confirm the effectiveness of the theo- flat straight track needing to decelerate due to a temporary
retical derivation, we consider an example scenario where the speed restriction. The RBC sends at time instant t = 100 [s]
coupling command is issued by the RBC when all trains are on the related information and train start decelerating from their
a flat track and traveling with similar velocities. We consider steady-state reference velocity of 210 [Km/h] toward a very
that train T1 has an initial speed of 208 [Km/h], the initial low speed of 40 [Km/h]. Results show that the emergency
speed of the following train T2 is 200 [Km/h] and the initial maneuver is safely and automatically executed and all trains
speed of T3 is 205 [Km/h]. The initial distance between train correctly follow the trapezoidal breaking profile set by the
couples is assumed to be 2000 meters. At time instant 0, RBC (see Fig.7a) preserving the bumper-to-bumper spacing
the RBC issues the coupling command to the trains. Hence, as confirmed by time histories of distances and relative speed
the two communication links depicted in red in Fig. 5a are error in Fig. 7b and Fig. 7c. It is worth noting that during
established. From that instant of time, the RBC continuously braking inter-train distance dynamically changes according to
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

DI MEO et al.: ERTMS/ETCS VIRTUAL COUPLING: PROOF OF CONCEPT AND NUMERICAL ANALYSIS 9

Fig. 6. Couplings creation and maintenance. (a) Time history of the position errors (green dashed line:r2 (t) − r1 (t) − h 21 v 0 − d st ; blue dotted line:
r3 (t) − r1 (t) − h 31 v 0 − d st ), exponential envelops are shown in red). (b) Time history of the speed error with respect to the reference. (c) Time history of
the train accelerations.

Fig. 7. Emergency Braking. (a) Time history of speed profile. (b) Time history of the bumper-to-bumper distance between the trains. (c) Time history of the
speed error. (d) Time history of the train accelerations.

the spacing policy that safely shapes the gaps as the velocity since even in perturbed coupling conditions all trains safely
changes (Fig. 7b). reach the coupling goals, with an appropriate settling time
below 2000 [s] according to the control tuning for the expo-
C. Sensitivity to Perturbed Coupling Conditions nential convergence. Results also show that, as expected, trains
In order to investigate the robustness of the control algo- automatically implement diverse transient acceleration profiles
rithm w.r.t. uncertainties in initial train conditions, differ- (e.g. middle train T2 strongly accelerates due to its low initial
ent train velocities have also been considered. To that aim, velocity) in order to reach the prescribed coupling distances
we refer to the coupling engagement maneuver with initial and the required velocity regime.
speeds of trains T1 and T3 still quite similar, i.e. 208 [Km/h]
for T1 and 205 [Km/h] for T3 , but with train T2 being D. Evaluation of Virtual Coupling in a Real Scenario
much slower, with initial speed of 150 [Km/h]. All trains are In this section we show how to evaluate the Virtual Coupling
initially running at a bumper-to-bumper distance of 2000 [m]. performance in realistic operating conditions by means of a
At time 0, the RBC sends the coupling command with a case study addressing the dynamic response of trains traveling
reference velocity of v 0 = 210 [Km/h], with trains and the on the Italian ERTMS track connecting Milano to Bologna.
infrastructure communicating via the LPF topology. Results in The model considers detailed information accounting for the
Figs. 8a and 8b confirm the robustness of the control algorithm traction effort, dynamic constraints (maximum acceleration
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 8. Spread in the initial conditions. (a) Time history of the bumper-to-bumper distance between the trains. (b) Time history of the speed error. (c) Time
history of the train accelerations.

Fig. 9. Virtual coupling while traveling from Milano to Bologna (Italy). (a) Time history of the bumper-to-bumper distance between the trains. (b) Time
history of the speed error. (c) Time history of the train accelerations.

and deceleration), train resistances, and also includes the vehicle control layer or guidance layer, sensing the environ-
nonlinear effects due to gradient and curve forces along the ment and controlling the vehicle (traction, brake, etc.) [36];
route. The goal is to evaluate the ability in reaching and vehicle management layer, coordinating the vehicles while
keeping the coupled behavior with three trains in face of they are performing maneuvers (such as platooning) based on
nonlinear time-varying disturbances due to the environmental inter-vehicle communication; traffic control layer, addressing
conditions. advisory instructions to optimize the traffic flow for each
Since the autonomous departure and arrival are not to segment (see e.g. [14], [37]–[39]).
executed in Virtual Coupling, in the scenario we assume
that the trains are already traveling when the RBC sends VI. C ONCLUSION
the coupling command and train T1 is located at a distance We have shown in this paper how railway Virtual Coupling
of 6 [km] from the start, while trains T2 and T3 are at a distance can be implemented by extending the standard set of ERTMS
of 4 [km] and 2 [km], respectively, with initial velocities Operating Modes through a proof-of-concept and example
v 1 = 208 [km/h], v 2 = 200 [km/h] and v 3 = 205 [km/h]. numerical analysis. In Full Supervision, when the operational
Results in Fig. 9 illustrate the coupling engagement maneouver conditions allow Virtual Coupling, the RBC issues to the trains
from the initial instant (time 0) when the RBC issues the the coupling command acting as a Virtual Leader. Virtual
coupling command. Results confirm that the required dynamic Coupling enables extremely short headways and hence higher
behavior is correctly set to the trains (see Figs. 9a and 9b). track capacity. As demonstrated by the results reported in this
As shown in Fig. 9c, the traction force varies according to paper, appropriate mathematical models and control algorithms
the control input u i to counteract the presence of gradient borrowed from the automotive field can be employed to prove
and curve forces as well as of forces depending on rolling the effectiveness of cooperative control strategies in reference
and aerodynamic resistances. Acceleration dynamics of all coupling scenarios and even in degraded scenarios like emer-
trains are always within admissible bounds despite the changes gency braking (e.g. line sections out of service). Regardless
in track elevation, with a variation that is always below of the specific results, the numerical analysis and simulation
the threshold required for guaranteeing ride comfort and for tools provided in this paper allow engineers to evaluate the
preventing motion sickness [35]. feasibility and performance of Virtual Coupling in diverse
Regarding implementation related aspects, the target archi- scenarios and operating conditions. Although this paper pro-
tecture for railways is expected to be similar to the one used in vides a preliminary analysis of railway Virtual Coupling,
collaborative driving systems in the automotive field. In order the challenges posed by the safe coordination of coupled train
to tackle implementation issues coming from time-critical convoys open up several future directions as a precondition
driving applications, connected autonomous vehicles leverage for real implementation. To that aim, further research efforts
on a stratified architecture including the following layers: are needed to assess the implications of Virtual Coupling in
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

DI MEO et al.: ERTMS/ETCS VIRTUAL COUPLING: PROOF OF CONCEPT AND NUMERICAL ANALYSIS 11

ERTMS operational scenarios. In particular, the safety-related [22] H. Venkataraman, A. D’Ussel, T. Corre, C. H. Muntean, and
impact must be investigated further considering new potential G. M. Muntean, “Performance analysis of real-time multimedia trans-
mission in 802.11p based multihop hybrid vehicular networks,” in
hazards. Performance evaluation must also be extended to Proc. 6th Int. Wireless Commun. Mobile Comput. Conf., Jun. 2010,
evaluate the capacity increase in common traffic scenarios, pp. 1151–1155.
including mixed traffic. Therefore, we plan to study addi- [23] C. Nowakowski, D. Thompson, S. E. Shladover, A. Kailas, and X.-Y. Lu,
“Operational concepts for truck maneuvers with cooperative adaptive
tional scenarios accounting for topology variations, complex cruise control,” Transp. Res. Rec., vol. 2559, no. 1, pp. 57–64, 2016.
maneuvers, as well as to address solutions for fast and reliable [24] L. Stranden, “A framework for complex scenarios for vehicle pla-
Train-to-Train communication. toons,” in Proc. IEEE 18th Int. Conf. Intell. Transp. Syst., Sep. 2015,
pp. 2186–2193.
[25] K. Gu, V. L. Kharitonov, and J. Chen, Stability of Time-Delay Systems.
R EFERENCES Boston, MA, USA: Springer, 2003.
[1] ERTMS/ETCS System Requirements Specification, UNISIG Standard. [26] E. Fridman and Y. Orlov, “Exponential stability of linear distributed
SUBSET-026, 2002. parameter systems with time-varying delays,” Automatica, vol. 45, no. 1,
[2] Shift2Rail Joint Undertaking. (2015). Multi-Annual Action pp. 194–201, Jan. 2009.
Plan. [Online]. Available: https://www.shift2rail.org/wp-content/ [27] A. Albrecht, P. Howlett, P. Pudney, X. Vu, and P. Zhou, “The key
uploads/2013/07/MAAP-final_final.pdf principles of optimal train control—Part 1: Formulation of the model,
[3] N. Furness, H. van Houten, L. Arenas, and M. Bartholomeus, “ERTMS strategies of optimal type, evolutionary lines, location of optimal switch-
Level 3: The game-changer,” IRSE News, vol. 232, pp. 2–9, Apr. 2017. ing points,” Transp. Res. Part B, Methodol., vol. 94, pp. 482–508,
[4] L. Hien, Q. P. Ha, and V. N. Phat, “Stability and stabilization of switched Dec. 2016.
linear dynamic systems with time delay and uncertainties,” Appl. Math. [28] ETR460-463 Class of Locomotives. Technical Sheet. Accessed:
Comput., vol. 210, no. 1, pp. 223–231, Apr. 2009. Apr. 16, 2019. [Online]. Available: http://www.leferrovie.it/leferrovie/
[5] I. Mitchell et al., “ERTMS level 4, train convoys or virtual coupling,” wiki/doku.php?id=schede_tecniche:start
Int. Tech. Commitee, vol. 219, pp. 1–3, Feb. 2016. [29] International Union of Railway. High-Speed Locomotives.
[6] J. Goikoetxea, “Roadmap towards the wireless virtual coupling of Technical Sheet. Accessed: Apr. 16, 2019. [Online]. Available:
trains,” in Proc. Int. Workshop Commun. Technol. Vehicles, May 2016, https://uic.org/IMG/pdf/high_speed_brochure.pdf
pp. 3–9. [30] Italian State Railways. High-Speed Locomotives. Technical sheet.
[7] B. Ai et al., “Challenges toward wireless communications for high-speed Accessed: Apr. 16, 2019. [Online]. Available: https://www.fsitaliane.it/
railway,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 5, pp. 2143–2158, content/dam/fsitaliane/Documents/media-ed-eventi/comunicati-stampa-
Oct. 2014. e-news/anno-2017/ottobre/2017_10_10_scheda_tecnica_pop.pdf
[8] J. Moreno, J. M. Riera, L. de Haro, and C. Rodriguez, “A survey on [31] A. Steimel, Electric Traction-Motive Power and Energy Supply: Basics
future railway radio communications services: Challenges and opportu- and Practical Experience. Munich, Germany: Oldenbourg Industriever-
nities,” IEEE Commun. Mag., vol. 53, no. 10, pp. 62–68, Oct. 2015. lag, 2008.
[9] P. Unterhuber, A. Lehner, and F. de Ponte Müller, “Measurement and [32] S. E. Li, Y. Zheng, K. Li, L.-Y. Wang, and H. Zhang, “Platoon control
analysis of ITS-G5 in railway environments,” in Proc. Int. Workshop of connected vehicles from a networked control perspective: Literature
Commun. Technol. Vehicles, May 2016, pp. 62–73. review, component modeling, and controller synthesis,” IEEE Trans. Veh.
[10] P. Fraga-Lamas, T. M. Fernández-Caramés, and L. Castedo, “Towards Technol., to be published.
the Internet of smart trains: A review on industrial IoT-connected [33] J. Lofberg, “YALMIP: A toolbox for modeling and optimization in
railways,” Sensors, vol. 17, no. 6, p. 1457, Jun. 2017. MATLAB,” in Proc. IEEE Int. Conf. Robot. Automat., Sep. 2004,
[11] D. Jia, K. Lu, J. Wang, X. Zhang, and X. Shen, “A survey on platoon- pp. 284–289.
based vehicular cyber-physical systems,” IEEE Commun. Surveys Tuts., [34] W. Alasmary and W. Zhuang, “Mobility impact in IEEE 802.11p
vol. 18, no. 1, pp. 263–284, 1st Quart., 2016. infrastructureless vehicular networks,” Ad Hoc Netw., vol. 10, no. 2,
[12] S. Santini, A. Salvi, A. S. Valente, A. Pescapé, M. Segata, and pp. 222–230, 2012.
R. L. Cigno, “A consensus-based approach for platooning with interve- [35] J. Förstberg, “Ride comfort and motion sickness in tilting trains. Human
hicular communications and its validation in realistic scenarios,” IEEE responses to motion environments in train and simulator experiments,”
Trans. Veh. Technol., vol. 66, no. 3, pp. 1985–1999, Mar. 2017. Ph.D. dissertation, Royal Inst. Technol., Stockholm, Sweden, 2000.
[13] M. di Bernardo, A. Salvi, and S. Santini, “Distributed consensus strategy [36] Y.-D. Song, Q. Song, and W.-C. Cai, “Fault-tolerant adaptive control of
for platooning of vehicles in the presence of time-varying heterogeneous high-speed trains under traction/braking failures: A virtual parameter-
communication delays,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 1, based approach,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 2,
pp. 102–112, Feb. 2015. pp. 737–748, Apr. 2014.
[14] R. Hult et al., “Design and experimental validation of a cooperative [37] S. Tsugawa, S. Kato, T. Matsui, H. Naganawa, and H. Fujii, “An archi-
driving control architecture for the grand cooperative driving challenge tecture for cooperative driving of automated vehicles,” in Proc. IEEE
2016,” IEEE Trans. Intell. Transp. Syst., vol. 19, no. 4, pp. 1290–1301, Intell. Transp. Syst., Oct. 2000, pp. 422–427.
Apr. 2018. [38] S. Hallé and B. Chaib-Draa, “Collaborative driving system using team-
[15] F. Rispoli. (2013). GNSS for Train Control Systems. [Online]. Available: work for platoon formations,” in Proc. Appl. Agent Technol. Traffic
https://www.gsa.europa.eu/sites/default/files/APPAP_Workshop_ Transp., Mar. 2005, pp. 133–151.
Ansaldo_0.pdf [39] A. Morales and H. Nijmeijer, “Merging strategy for vehicles by applying
[16] R. Rajamani, Vehicle Dynamics and Control. Boston, MA, USA: cooperative tracking control,” IEEE Trans. Intell. Transp. Syst., vol. 17,
Springer, 2011. no. 12, pp. 3423–3433, Dec. 2016.
[17] G. J. L. Naus, R. P. A. Vugts, J. Ploeg, M. J. G. van de Molengraft, and
M. Steinbuch, “String-stable CACC design and experimental validation:
A frequency-domain approach,” IEEE Trans. Veh. Technol., vol. 59,
no. 9, pp. 4268–4279, Nov. 2010.
[18] M. di Bernardo, P. Falcone, A. Salvi, and S. Santini, “Design, analysis,
and experimental validation of a distributed protocol for platooning
in the presence of time-varying heterogeneous delays,” IEEE Trans. Carlo Di Meo received the M.Sc. degree in control
Control Syst. Technol., vol. 24, no. 2, pp. 413–427, Mar. 2016. systems engineering from the University of Naples
[19] G. Chen and F. L. Lewis, “Leader-following control for multiple inertial Federico II, Naples, Italy, in 2017. He is currently
agents,” Int. J. Robust Nonlinear Control, vol. 21, no. 8, pp. 925–942, a Junior Research Fellow with the Department of
2011. Electrical Engineering and Information Technology,
[20] F. Librino, M. E. Renda, and P. Santi, “Multihop beaconing forwarding University of Naples Federico II. His research inter-
strategies in congested IEEE 802.11p vehicular networks,” IEEE Trans. ests include intelligent transportation systems and
Veh. Technol., vol. 65, no. 9, pp. 7515–7528, Sep. 2016. communication-based train control.
[21] J. Mittag, F. Thomas, J. Härri, and H. Hartenstein, “A comparison of
single-and multi-hop beaconing in VANETs,” in Proc. 6th ACM Int.
Workshop VehiculAr InterNETworking, Sep. 2009, pp. 69–78.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Marco Di Vaio received the M.Sc. degree in control Stefania Santini received the M.Sc. degree in elec-
systems engineering from the University of Naples tronic engineering and the Ph.D. degree in automatic
Federico II, Naples, Italy, in 2017. He was a Visiting control from the University of Naples Federico II,
Student with the Chalmers University of Technol- Naples, Italy, in 1996 and 1999, respectively. She
ogy, Göteborg, Sweden, in 2016. He is currently a is currently an Associate Professor in automatic
Research Fellow with the Department of Electrical control with the University of Napoli Federico II.
Engineering and Information Technology, University Her research interests include analysis and control
of Naples Federico II. His research interest includes of nonlinear, networked, and delayed systems with
the distributed control applied to autonomous driving applications to computational biology, energy, auto-
and cooperative systems. motive engineering, and transportation technologies.
She is involved in many projects with industry,
including SMEs operating in the automotive field.
Francesco Flammini received the M.D. and Ph.D.
degrees in computer engineering from the Uni-
versity of Naples Federico II, in 2003 and 2006,
respectively. He is currently a Senior Lecturer
with Linnaeus University, Sweden. He has a
long industry experience working on international
research projects addressing ERTMS/ETCS and
critical infrastructures, including the EU funded
projects, named PilotSHIELD and NewSHIELD,
specializing on the dependability of embedded sys-
tems in railway applications.

Roberto Nardone received the M.D. and Ph.D. Valeria Vittorini received the M.D. degree in math-
degrees in computer engineering from the University ematics and the Ph.D. degree in computer engi-
of Naples Federico II, Naples, Italy, in 2009 and neering from the University of Naples Federico II,
2013, respectively. He is currently a Post-doctoral in 1991 and 1995, respectively. She has been an
Fellow with the University of Naples Federico II. Associate Professor with the University of Naples
His research interests include quantitative evaluation Federico II, since 2005. She teaches computer
of non-functional properties, with a particular focus programming, formal modeling, and workflow and
on dependability and performability assessment and process automation. Her current research interests
threat propagation analysis, by means of model- include dependability and performance evaluation of
based and model-driven techniques. He is involved computer systems, validation and verification of crit-
in research projects with academic and industrial ical systems, and critical infrastructures protection
partners. with a special focus on railway transportation systems.