ArticlePDF Available

Abstract and Figures

Digital twins found their genesis in the halls of NASA and the methods of product lifecycle management. Rapidly evolving trends around the proliferation of sensors, the internet of things, industry 4.0, and cyber-physical systems have spurred the growth of digital twins. This paper reviews digital twins and their use in connected and automated vehicles (CAVs). Strictly speaking, digital twins must have communication between a physical system and its model, as opposed to similar methodologies that achieve indirect communication through iteration, or that substitute different parts of a system simulation with bits of hardware or software for testing. In practice, new methodologies for testing CAVs are sufficiently complex and difficult to apply simple labels. This is seen in our review of vehicular digital twins. Several gaps and challenges are apparent for the continued advancement of digital twin applications. We note some developing areas as traffic management centers, digital maps, onboard diagnostics, and logistics. Digital twins foster model reuse and encourage the use of multiple models at different scales of resolution. The role of digital twins will continue to grow as models become more tightly integrated to the physical systems they represent. This will drive such models towards uniqueness (matching a particular vehicle or road), adaptability (evolving with changing conditions and subject to wear and tear), and interpretability (conveying useful information to a human user). A maturing connected infrastructure and the development of smart cities will cause the number of new digital twin services to explode in a myriad of unforeseen ways.
Content may be subject to copyright.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
The Role of Digital
Twins in Connected and
Automated Vehicles
Dig ital Obj ect Iden tifie r 10.1109/MIT S.2021.3129524
Dat e of curre nt versi on: 5 January 2022
Chris Schwarz
Is with the National Advanced Driving Simulator, University of Iowa,
Iowa City, Iowa, 52242, USA.
Email: chris-schwarz@
Ziran Wang
Is with Toyota Motor North America R&D, InfoTech Labs, Mountain View,
California, 94043, USA.
Abstract—Digital twins found their genesis in the halls of NASA and the methods of product lifecycle
management. Rapidly evolving trends around the proliferation of sensors, the Internet of Things, Industry 4.0,
and cyber-physical systems have spurred the growth of digital twins. This paper reviews digital t wins and
their use in connected and automated vehicles (CAVs). Strictly speaking, digital twins must have communi-
cation between a physical system and its model, as opposed to similar methodologies that achieve indirect
communication through iteration, or that substitute different parts of a system simulation with bits of
hardware or sof tware for testi ng. In practice, new methodologies for testing CAVs are sufficiently complex
and difficult to apply simple labels. This is seen in our review of vehicular digital twins. Several gaps and
challenges are apparent for the continued advancement of digital twin applications. We note some devel-
oping areas as traffic management centers, digital maps, onboard diagnostics, and logistics. Digital twins
foster model reuse and encourage the use of multiple models at different scales of resolution. The role of
digital twins w ill continue to grow as models become more tightly integrated to the physical systems they
represent. This will drive such models towards uniqueness (matching a par ticular vehicle or road), adapt-
ability (evolving with changing conditions and subject to wear and tear), and interpretability (conveying
useful information to a human user). A maturing connected infrastructure and the development of smart
cities will cause the number of new digital twin services to explode in a myriad of unforeseen ways.
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Several t rends are converging to make digital twins
viable and attractive in the transportation domain,
particularly in the fast-moving field of connected and
automated vehicles (CAVs). The Internet of Things
(IoT) movement seeks to connect large numbers of dev ices
with embedded sensors and sof tware [1], so large that In-
ternet addressing will need to shift from IPv4 to IPv6 pro-
viding an address space 1,028 times larger. Developments
on the IoT front promise to revolutionize manufacturing
and industry. This vision is captured under the heading
of Industry 4.0 [2]. The proliferation of this technology is
driving an evolution in the relationships between software
and hardware, between systems and models, and between
complex systems and humans who need to interact with
them [3]. A closely related concept is that of the cyber-phys-
ical system (CPS), in which tight integration is maintained
between digital and physical components. CPS and IoT are
key technologies in smart cities and intelligent transporta-
tion systems (ITS) of the future.
The idea of t winning goes back to NASA using dupli-
cates of their Apollo spacecraft on the ground while astro-
nauts were on mission in space. They later offered the first
formal definition of a digital twin that focused on high-
fidelity, lifecycle, and interdependent vehicle systems [4].
The twinning concept was specific enough to match a des-
ignated tail number. The automotive equivalent would be
to match the Vehicle Identification Number (VIN) to cap-
ture all relevant information about vehicle trim, opt ional
packages, and maintenance history.
Meanwhile, the concept of the digital twin was also de-
fined in the context of product lifecycle management at the
University of Michigan, even if the label was not yet applied
[5]. At the very hear t of the idea was that the digital twin
would be linked w ith its physical counterpart throughout
its entire lifecycle so that the two parts could evolve in tan-
dem. Grieves split the lifecycle of a product or system into
four phases: creation, production, operations, and disposal.
More recently, Barricell i et al. [1] defined the lifecycle
phases as designed, development, operational, and dismiss-
al [1], while Liu et al. used the ter minology design, manu-
facturing, service, and retire [6]: same meanings—different
labels. Grieves et al. furthermore asserted that traditional
system engineering models such as the waterfall model,
the spiral model, and the Vee model are idealistic and have
the potential to lead to fragile systems. The reality, on the
other hand, is that system design is highly iterative. In
other words, the loop among modeling, simulation, proto-
typing, and testing is much tighter in practice. The digital
twin concept embraces this reality and pushes it to its logi-
cal conclusion—they can in fact be simultaneous.
As computing power drove advances in simulation
capabilities, modeling and simulation found new appli-
cations in research, design, testing, and validation. Small-
scale applications in the 1960s gave way to wel l-developed
simulation tools in the 1980s. Model-based design began
to revolutionize manufacturing at the turn of the centur y,
and now digital twins hold out the promise of simulation
adding value throughout the entire lifecycle of a device [7].
Whereas historical modeling and simulation methodolo-
gies iterate back and forth between protot yping and build-
ing physical systems, a digital twin evolves over time with
its physical counterpart throughout its life, perhaps even
into its retirement. The distinctions among iteration (the
typical approach in creating virtual proving grou nds and
tuning models), model-based design, and digital twins is
diagrammed in Figure 1.
The virtual proving grou nd was popularized in the 1990s
[8] and is still in use today [9]. A virtual proving ground in
the automotive space consists of a high-fidelity model of
vehicle, or vehicle subsystem as well as the environment
with which it interacts [10]–[12]. Ideally, an environ ment
model would duplicate a real-world environment—a test
track or proving ground, rather than a nonexistent world.
Then test results obtained on the track can be used to
verify and validate the models and simulation results. The
third par t of a virtual proving ground is the driver, whose
role may be fulfilled by a model or by a human in a driv-
ing simulator [8], [13], [14]. The role of the driver has been
automated with the help of standardized vehicle maneu-
vers and robotic steering actuators are often installed in
vehicles to obtain repeatable results w ith J-turns, double
lane changes, understeering tests, and so on. This simpli-
fies the problem of modeling a human driver and closes the
experimental gap between virtual and real tests [15], [16].
The automotive industry was an early adopter of model-
based design, specifically for electronics and embedded
systems [17]. The adoption of model-based design has en-
couraged the use of modeling and simulation in the prod-
uct design cycle and is associated with the techniques of
model-in-the-loop, software-in-the-loop, and hardware-
in-the-loop testing. Model-based system engineering ex-
pands the concept to systems as well as systems of systems
[18], [19]. Virtual proving grounds, model-based design, and
model-based system engineering have traditionally been
concerned with the design, testing, verification, and vali-
dation of a product. To the extent that they result in more
reliable products, t hey also indirectly affect later phases of
the lifecycle by reducing the occurrence of manufacturing
delays and safety recalls.
Mass adoption of simulation-based methodologies re-
quires new tools and processes geared towards the vir tu-
alization of physical components, systems, and systems of
systems. Virtualization was discussed by Schoeggl et al. in
the context of advanced driver assistance systems (ADAS)
and automated driving systems (ADS) in 2018 [20], and
a year later by Howard for electronics development and
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
design [21]. Through the process of virtualizing and validat-
ing a vehicle, test environment, and scenarios, Shoeggl esti-
mated that testing time could be reduced by 80 to 90%. Rather
than replacing the physical with the virtual, the integration
of the two into evermore complex CPS has created enormous
opportunities and challenges [22]. The scope of CPS is very
broad and overlaps with the definitions of other terms in this
section. But in contrast to virtual proving ground, model-
based design, and virtualization, CPS typically refers to an
operational system rather than a design methodology. For
example, the problem of distributed control of a network of
CAVs, a complex CPS, was treated by Feng et al. [23].
The numbers of sensors, electronic control units, and
computers have exploded over the last couple of decades.
The introduction of ADAS and ADS required vehicles to
become aware of their surroundings, make decisions, and
act on them. This proliferation continues unabated with
increasing market penetration of infotainment and active
safety systems. Now the introduction of connected-vehicle
technology and applications has injected an IoT ethos into
the transpor tation system and motivated brand-new applica-
tions of modeling and simulation, including the digital twin.
What Are Digital Twins?
The presence of communication is a key differentiator for the
digital twin, so much so that additional definitions for digital
model and digital shadow have been introduced specifically
to avoid confusion. A digital model mimics a physical object
or system but there is no communication or data exchange
between the two. A digital shadow has a one-way flow of com-
munication from the physical model to the digital model. In
contrast, the digital twin exchanges data in both directions [2],
[24], [25]. Fuller et al. noted several instances where digital
models or shadows were misattributed as digital twins in the
Digital Tuning
Digital Replica
Physical Tuning
Physical Entity
Driver Vehicle Traffic Actuation
(a) (b)
Digital Twin
Model-Based Design
FIG 1 The simulation methodologies. (a) The iteration to convergence is commonly used to switch back and forth between the physical and digital
spaces. (b) The model-based design often starts digital and incrementally swaps in physical components. (c) The digital twins maintain synchronized
versions of a physical system and its digital counterpart.
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
literature. To add to the confusion, many other terms have
been used for similar applications in different fields including
computational mega model, device shadow, mirrored system,
product avatar, and synchronized virtual prototype, among
others [1], [26].
Barricelli et al. [1] pointed out that there must be at least
three types of communication around digital twi ns. First,
the digital and physical twins must communicate; second,
multiple digital twins in a complex simulated environment
should be able to communicate among each other; third,
there should be communication between digital twins and
domain experts through user interfaces [1]. A more com-
prehensive breakdow n is offered by the five-dimensional
digital twin model, introduced by Tao [27], [28]. The five
dimensions can be expressed in a formula as
,,,,PE VM Ss DD CN
, (1)
where PE are physical entities, VM are virtual models, Ss
are services, DD stands for data, and CN denotes connec-
tions. Connections are established between each possible
pairing of the other four dimensions.
The five-dimensional model adds two new ideas. The use
of high-fidelity models with high-bandw idth connections to
the world, and collected over long spans of time, establishes
digital twins f irmly in the category of big data. Even if the
model is not high-fidelity, there may exist several models,
each informed by a set of sensors. The model of the whole
system might be constituted in the cloud and populated by
multiple streams of data from the physical tw in(s), generat-
ing enormous amounts of data over time. The value added
by digital twins takes the form of serv ices, the fifth dimen-
sion. Ser vices can provide a variety of functions, such as
analyzing the operation of t he physical system, adapting to
changes in the env ironment, changing its behavior, pred ict-
ing failures before they occur, and more [29]–[31].
Jones et al. carried out a systematic review of digital
twins and included an eight-dimensional model defined by
the International Academy for Production Engineering’s
(CIRP) Encyclopedia of Production Engineering [32]. The
CIRP definition mentions the specificity of a unique prod-
uct, single or multiple lifecycle phases, models, informa-
tion, and data. In contrast to other definitions, it does not
emphasize the digital-to-physical connection, thus high-
lighting some of t he dif ferences in usage across different
sectors and application domains.
Rasheed summarizes eight ways in which digital twins
add value [26], and several of them are exemplified in the
CAV digital twin applications described in this article.
They include the following benefits:
real-time remote monitoring and control
greater efficiency and safety
predictive maintenance and scheduling
scenario and risk assessment
better intra- and inter-team synergy and collaborations
more efficient and informed decision support system
personalization of products and ser vices
better documentation and communication.
Alam and Saddik proposed a digital twin architecture
reference model that was designed to accommodate cloud-
based communication [33]. The challenges of fusing local
sensor data with cloud-based data from disparate sources
were addressed using a Bayesian belief network. Further
flexibility was introduced with fuzzy rule bases and hierar-
chical systems. The framework addresses many of the con-
ceptual challenges of implementing a digital twin service in
a cloud-based CPS. Given enough computing resources, an
entire transportation system can be modeled and maintained
as a digital twin in the cloud. Multiple services can then be
provided to drivers, vehicles, and traffic managers based on
analyses and predictions made with the digital twin.
Before presenting digital twin examples, we contrast
them against the other methods described earlier. In the
digital twin literature, it is tempting to apply the label to
new methodologies, even if one of the more traditional
terms, such as virtual proving ground, would be more ac-
curate. While misattribution is frequent, we conjecture that
the cause is simple misunderstanding or failu re to agree on
standard terminology. In fact, the complexit y of modern de-
sign frameworks is already so great t hat it is difficult to tell
when one methodology ends and another begins.
Nevertheless, it is useful to define terms, categorize meth-
ods, and understand where the state of the art lies. There is a
tendency to equate a digital twin with a highly accurate model
of a physical system or environment. In other words, as a model
gets more accurate, it converges to a digital twin [34]–[36]. We
take the opportunity in this article to push back against that no-
tion and reinforce definitions of digital twins that have emerged
over the last decade and were described earlier. Hallerback
et al. [34] conclude with a prototype-in-the-loop method, which
would meet the strict five-dimensional definition.
We note three ways that digital twins are distinct
from ot her methodologies, which often create confusion
in the literature. Communication is a key aspect of the
three-dimensional and five-dimensional models of digital
twins. In contrast, virtual proving ground and model-
based design methodologies often utilize iteration between
physical and virtual tests to rapidly design, validate, and
test a system. While the overall time needed for such testing
can be reduced dramatically, the process is still iterative
and often requires manual tuning by engineers in between
physical and virtual tests [37], [38]. The communication
channels built into digital twins are understood to be more
tightly coupled, usually simultaneous and real-time.
A second observation seems obvious but is easy to miss
when reading complicated descriptions of test environ-
ments. A digital twin maintains a connection between the
physical object and the model that represents it. Model-based
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
design and model-based system engineering methodolo-
gies make use of models that work in conjunction with
bits of software and hardware but are not required to pair
a model with its physical counterpart. Rather, an electron-
ic control unit is paired with a vehicle dynamics model;
a real vehicle drives through a virtual environment with
simulated traffic; a deep learning model for an ADS is
trained using simulated scenarios [36].
A third factor t hat puts digital twins more in the camp of
CPS than wit h traditional methodologies like virtual prov-
ing ground and model-based design is that digital twins
are useful throughout the lifecycle of product. However,
we do not see this distinct ion as being critical to the clas-
sification of a digital twin. One can certainly create a tw in
whose purpose is l imited to the design or manufacturing
phase of a product lifecycle as long as it has the five di-
mensions. As IoT and cloud-based applications mature,
digital twins that had been limited in their application will
naturally find new life throughout the vehicle’s lifecycle.
In practice, the testing framework can be so complex that
different methodologies coexist or trade off responsibilities
at different times [39].
Examples of Digital Twins
There are several examples in the digital twin literature
geared towards the development of CAVs. It is not clear in
several of these cases whether t he digital tw in is the cor-
rect label; however, we reiterate the fact that testing frame-
works have become increasingly complex and difficult to
capture with any single label. The examples documented
here are also summarized in Table 1. The included refer-
ences were selected through a Google Scholar search using
the terms digital twins, connected and automated vehicles,
virtual proving grounds, model-based design, cyber-physi-
cal systems, and parallel driving. Search results were fil-
tered dow n to focus on CAVs and exclude other digital twi n
applications. The cited papers invoke digital twins, either
in spirit or with their language. The methodology classi-
fications offered in Table 1 were assigned af ter reviewing
the examples.
Transition Applications
An early application of digital twinning that sits bet ween
Industry 4.0 and ITS is instructive and has parallels to
more recent applications in the CAV field. It concerns
a factory floor w ith an unmanned autonomous vehi-
cle. The cyber-guided vehicle described in the article
is a materials-handling robot in an industrial setting
[31]. The entire system was miniaturized and realized
with small Arduino-powered robots. The vehicle moves
around a predefined track and is tasked with deliveri ng
material to one of three different handling stations. The
digital twin was a discrete-event simulation that man-
ages scheduling of operations in the facility. Since it was
not concer ned with lower-level operations such as the
dynamic dr iving task, a si mple digital representation of
the vehicle sufficed.
Perhaps the most direct translation from Industry 4.0 ap-
plications to an ITS occurs in tightly controlled vehicle en-
vironments such as docks, shipping yards, and distribution
centers. Barosan et al. developed a digital twin to support
truck docking at a distribution center [40]. The digital twin
was constructed using a model-based system engineer-
ing methodology in a modeling environment composed of
IBM Rhapsody, MATLAB Simulink, and Unity. There may be
several trucks driving around the dock and the simulations
helped solve complex path-planning problems. The authors
recognized their so-called virtual simulation environment to
be a valuable testing tool for engineers, but apart from a brief
mention of scaled-down trucks with Raspberry Pi control-
lers, little is said about the physical components. Neither did
they focus on its potential to improve scheduling efficiency
and driving safety during everyday operations at a function-
ing distribution center, a logical next step for the digital twin
in the lifecycle of the facility.
An example of a transition application between model-
based design and digital twin involved the design of a brake
system module [37]. The authors proposed an iterative
scheme in which the model was optimized to accurately
represent the physical system, a process they termed vir-
tual and physical simulation technology. A similar process
for testing the design of an automatic electric vehicle charg-
ing system was proposed by Shikata et al. [38]. Culley et al.
presented a digital twin for an autonomous racing vehicle
that could swap out the real vehicle for a simulation of the
vehicle, sensors, and track environment [41]. These applica-
tions share a common theme with many in the digital twin
literature. The digital and physical versions of the common
object do not appear to communicate in real time. Rather,
the user tunes the simulation and compares its output with
that of the physical system and then iterates. Model-based
design commonly connects a model of a controller with the
physical system, or vice versa. However, this approach and
the loosely coupled communication implied by iteration do
not strictly meet the definition of a digital twin. But the ac-
curate models developed in this process could be used dur-
ing operation to provide real-time diagnostics and optimize
performance, thus earning the name. Alternatively, a testing
paradigm like the ones described here could benefit from
establishing real-time communication between the device
under test and its simulation. For example, online model
identification techniques could be used to keep up with
changes in the hardware.
Automated Driving System Testing
The problem of comprehensively testing an ADS is so dif-
ficult and requ ires so much testing that it cannot be done
without the aid of simulation. Towards that end, much of
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
the recent literature is dedicated to the development of
simulation frameworks that can be applied to ADS testing.
A digital twin framework developed by Ge et al. proposed
three levels of testing: a pure virtual test, a real sensor
data-based test, and a vehicle-based test [42]. The frame-
work is exemplary of the combined met hodologies that
characterize modern complex systems. Models of the ve-
hicle, traffic, environment, sensors, and control algorithm
are developed in the first phase. The second phase demon-
strates a digital tw in application in which real sensor data
is fed into the simulation through a 4G/5G communication
channel and cont rol decisions could be fed back over the
network (at the time of publication, it did not seem that the
sensor data was collected in real time but fed from a data
center). In the thi rd phase, a real test vehicle is embedded
into a hybrid physical-digital environment that includes
simulated traffic and pedestrians. To the extent that such
testing also includes a simulation of the test vehicle, a digi-
tal t win is present. The mixture of a real vehicle with si m-
ulated traf fic is something else—an evolution of a virtual
proving ground into a full-f ledged CPS.
This type of CPS, one that represents a hybrid between
a digital twin and a virtual proving ground with a model-
based design flavor, has been used with great success at
Michigan’s Mcity test facility [43]. Another example was
developed with Unity, SUMO, and Python [44], since SUMO
contains a model of the test vehicle and communicates
with the physical vehicle, which meets the definition of a
digital twin. Rong et al. developed a framework in Unity
that integrates data from real sensors is the SVL simulator
by LG [45]. The environments are labeled as digital twins
because they model real-world locations, but there is no
requirement in the testing paradigm that a test vehicle
would need to drive on those same roads. A loose commu-
nication loop, established through iterative testing, was
introduced using the SVL simulator to develop a formal
Ref No. Lead Author Methodologies Component(s) Twinned Software/Frameworks Service
[31] E. Bottani CPS Environment, decision policy C++ Industry 4.0 process development
[37] V. Dygalo MBD, iteration Vehicle subsystem MATLAB, Amesim, Carsim ADAS design and test
[38] H. Shikata MBD, iteration Vehicle, Electronic Control Unit custom EV charging design and test
[39] V. Salehi MBSE None (all simulated) SysML, Vector DYNA4,
Cameo System Modeler, Unity,
Valet parking design and test
[47] I. Barosan VPG, MBSE Vehicle (tractor-trailer) SysML, MATLAB Simulink,
Autonomous truck control design
and test
[40] J. Culley VPG Vehicle, sensors, track ROS, Gazebo Autonomous racing vehicle design
and test
[41] Y. Ge MBD, CPS Vehicle not described Autonomous vehicle design and
[42] Y. Feng VPG, CPS Vehicle not described CAV test
[43] M. Szalai VPG Vehicle, pedestrian Unity, SUMO Autonomous vehicle test
[44] G. Rong MBD, VPG,
Environment (test track) SVL by LG, Unity, Apollo,
Autoware, ROS
SiL/HiL testing, V2X testing,
synthetic data generation
[46] D. J. Fremont VPG Environment (test track) SCENIC, VerifAI, SVL by LG Formal scenario-based testing
[47] Y. Laschinsky MBD Vehicle Vires VTD ADAS design and test
[49] X. Chen CPS Drivers MQTT, Unity, MATLAB Driver behavior risk analysis
[50] M. Buechel CPS Environment, traffic ROS, Apollo, NVIDIA DRIVE Vehicle behavioral planning
[51] Y. Liu CPS Traffic entities Unity ADAS
[52] Z. Wang CPS Vehicle, driver, environment,
not described ADAS, performance evaluation
[53] C. Schwarz CPS Vehicle subsystem, environment MATLAB ADAS enhanced ESC
[54] S. Liu CPS Vehicle, sensors, environment Unreal Engine Provide optimized routes and
trajectories for cooperation
[55] F.-Y. Wang CPS Environment, traffic iHorizon ADAS, energy management
MBD: model-based design, MBSE: model-based system engineering, VPG: virtual proving ground.
Table 1. A summary of digital twin examples related to CAVs.
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
testing method with the SCENIC scenario language and
the VERIFA I toolkit [46]. A testing environment that uses
Vires VTD sof tware with a vehicle-in-the-loop configura-
tion examined a n advanced headlamp application called
Active Safety Light [47].
Cyber-Physical Systems
One of the ironies of automation, highlighted by Bainbridge,
is that as automation increases, the role of the human oper-
ator may become more challenging and one for which they
are ill-equipped [48]. Another such irony is that as vehicles
become more connected and automated, they need to pay
more attention to the human operator, not less. A digital
twin application has been proposed for CAVs that models
the behavior of human drivers [49]. In fact, human behavior
models constitute the paper’s main digital t win, rather than
vehicle or environment models. The authors recognized
the social dimension to driving and the need to consider
human behavior when analyzing safety risks. The novelty
enabled by connected-vehicle safet y systems is the shar-
ing of driver behavioral models between vehicles. While
knowledge of vehicle state (position, velocity, orientation,
acceleration) can yield short-term predictions about future
state, knowledge about driver behavior adds a layer of rich-
ness to this prediction that considers human decision-mak-
ing and indiv idual differences. The env isioned application
creates a digital twin of a vehicle’s operator and passes that
model along to surrounding vehicles so that they can use it
to make better pred ictions. The authors develop this idea
into a risk analysis system that computes likelihood of colli-
sions bet ween vehicles. The idea could make automated ve-
hicles smarter in the context of mixed traffic. The proposed
digital twin is intended to function in the operational phase
of the lifecycle rather than during development.
The next CAV example uses sensors arou nd the infra-
structure to monitor vehicle traffic and create its digital
twin [50]. Information and recommendations derived from
the traffic digital tw in would then be fed to participating
vehicles to improve upon their safety, mobility, and ef-
ficiency (all services in the digital twin paradigm). The
authors describe the specifications and development of a
protot ype CAV called Fortuna. The communication chan-
nel between the vehicle and the digital twin of the environ-
ment would be accomplished through 5G cellular.
A common strategy that bridges developments in IoT,
CPS, and digital twins is the adoption of cloud computing.
The use of centrally managed simulations can bring to
bear more computing power to model a local ITS. A novel
approach for fusing data obtained from vehicle sensors to
data obtained from a cloud-based digital tw in was pro-
posed to improve visual guidance information for drivers
[51]. The authors lay out several challenges and algorithms
for dealing with the fusion of such disparate data sources.
The digital twin in this case consists of models of traf fic
agents including vehicles, drivers, a nd pedestrians, all
simulated in the cloud. The aut hors used the Unity game
engine as a simulation environment for human-in-the-
loop test, but it was not a key technology for the digital twin
The idea was further developed in a collaboration be-
tween Toyota InfoTech Labs and the University of Califor-
nia, Riverside [52]. As before, a digital twin in the cloud
simulated traffic participants and provided data to par-
ticipating vehicles over a 4G-LTE cellular communication
channel. The authors developed a cooperative ramp merg-
ing application as an example of what could be achieved.
They found that communication delays and packet losses
could be kept to within acceptable limits and that the sys-
tem could provide useful information to a driver in real
time. This is an example of a communication ser vice
between the digital twin and the domain expert (driver)
called out by Bar ricelli et al. [1].
We have described digital tw ins that model the driver
and traffic part icipants, but they must also model the driv-
ing env ironment. The most important piece of this environ-
ment is the road and digital maps of road networks continue
to grow in fidelity and resolution to keep up w ith demand-
ing applications of CAVs. Even lower-fidelity digital maps
can provide useful information as a service to a connected
vehicle. One way to think about digital maps is as a nontra-
ditional ty pe of sensor, one that is not limited by line of sight
[53]. A greater challenge is to keep digital maps up to date
as real-world conditions change. Recently a cloud-based
digital twin environment was proposed to merge traffic in-
formation with the high definition (HD) digital map [54]. To
the extent that the road digital maps can ref lect real-time
conditions and provide value-added ser vices to CAVs, they
represent a valuable digital twin application.
Parallel Driving
A concept very similar in scope and intent to the digital
twin is called parallel driving [55] and emerged from a
study of parallel control concepts in cyber-physical sys-
tems [56]. The parallel driving framework has three levels:
the physical world, t he mental world in which a road user
develops mental models and exhibits cognitive behaviors,
and the artificial world that contain computer-based mod-
els and simulations. Furthermore, parallel driving also
defines services that provide value by linking the three
layers in useful ways. Wang et al. provided two use cases
for parallel driving. The first describes parallel testing,
parallel learning and parallel reinforcement learning to
combine data from the physical and artificial worlds to
train a self-parking feature. The second was labeled as
intelligent horizon (iHorizon) and developed an extension
of eHorizon technology developed by Bosch, Continental,
and HERE. They made use of digital maps and driving style
recognition algorithms to make short and long-term speed
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
predictions. The services that could be prov ided by such
technology include traffic calming and intelligent energy
management technology.
Opportunities and Challenges
Ongoing IoT and CPS trends will continue to facilitate new
digital twin applications in the coming years. As CAVs gain
market penetration, the benefits of an increasingly connect-
ed traffic system will begin to multiply and realize the prom-
ise of connected safety warning systems. There are several
natural lines of development for digital twins that extend
already mature technologies. Other applications will be fun-
damentally new and shift the paradigm of what is possible.
The challenges that lie ahead for digital twins have
been well documented [1], [2], [24]. Data security and pri-
vacy are omnipresent watchwords for digital twin, CPS,
and IoT applications. Model interoperability and stan-
dardization enable the creation of models that can talk to
one another, and its absence continues to be a barrier to
large-scale systems. Issues around wealth and equit y must
be addressed before t his (and other) technological innova-
tions can be enjoyed worldwide. Data qualit y and validated
models are necessary to support critical services, but veri-
fying suf ficient accuracy is no easy matter. User interfaces
that allow humans to understand digital twin performance
and explain communicated results are diff icult to create
and user trust difficult to foster. Technical limitations re-
main that requ ire high-speed infrastructure and large-
scale computing platforms. On the other hand, this must
be balanced with the cost of developing and deploying a
new technology. We mention a few opportunities specific
to digital twins in the development and operation of CAVs.
Safety Critical Services
Digital tw in services for CAVs range from convenience to
crit ical. The sustainability and mobility of the transpor-
tation system will benef it from drivers receiving recom-
mendations on lane selection, speed, and route. However,
the promise of CAV technology includes safety applications
that w ill help to drive crashes to zero. In order to f ield an
effective safety serv ice, the digital twin needs to be accu-
rate, reliable, and have low latency.
Traffic Management Centers
Centrally managed locations for traffic observation and
analytics will gain powerful new capabilities to dynami-
cally advise and control traffic, from individual vehicle
alerts and recommendations to adaptive control over traf-
fic signals. This jump in capability w ill come from im-
proved cloud-based communication through 5G cellular
(and beyond), and from the addition of a simulation layer
that creates a true digital tw in of the traffic that can be
used for prediction and counterfactual scenarios. The co-
operative ramp merging application descr ibed earlier is
only one small example of the type of service that could be
supported by such a network.
Privacy and security are often presented as challenges
related to increasing connectivity, and they are indeed. It
will be possible to identify exact vehicles in traffic, down
to thei r license number and V IN, more than ever before.
There may need to be mult iple layers of data, highly se-
cure ones that contain personally identifiable information,
and less protected ones that are scrubbed of such data.
The potential fragmentation of connected transportation
networks is also of concern. It may be that existing traffic
management centers give up ground over time to networks
privately controlled by manufacturers or other institutions.
Then communication across networks to obtain complete
coverage w ill need to be negotiated and paid for.
Digital Maps
Digital maps are in the middle of a revolution, with simple
links and nodes being replaced by HD maps that contain
information about lane lines, road furniture, curbs and
elevation profiles, signals, and so on. It will take time for
such maps to proliferate, but they of fer tremendous oppor-
tunities for digital twin applications. The promise is not
simply adding higher definition to the map, but to main-
tai ning a live, dynamic digital t win of real roads. Weather,
time of day, work zones, accidents, rockslides, and other
events all contribute to temporary lane closures. Map sta-
tus can be crowd-sourced with d river reports and using
CAV sensors. Recommendations on route, lane choice,
speed, and caution zones are some of the services that
HD map digital twins can provide. The makers of such
maps, aware of the business opportunities being created,
are keeping tight control over access. While this will help
keep security concerns in check, it creates other concerns
around equity and accessibility to benefits provided by the
technology. Standardization of for mats is also required
to support nation-wide services such as quickly updating
work zone locations.
Onboard Diagnostics
Models created in the design phase as part of a virtual prov-
ing ground or model-based design methodology deserve a
new lease on life as onboard models that operate in paral lel
with the vehicle. The proliferation of sensors inside the ve-
hicle as well as outside will facilitate high-speed commu-
nication between the physical and digital twins. There are
many ser vices one could envision to make the vehicle more
reliable, including real-time diagnostics and failure pre-
diction, dynamic estimation of the coefficient of friction,
more accu rate service adv ice, and advanced controllers
that optimize safety, mobility, and efficiency parameters
during a drive. Additionally, the presence of onboard
models creates opportunities for edge computing applica-
tions t hat take some of the computing load out of the cloud.
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Logistics and scheduling are digital twin applications
more commonly associated with Industry 4.0 initiatives.
Such applications w ill continue to develop in factories,
shippi ng centers, hospitals, and airports, and t hey will find
ways to jump into transportation and integrate wit h CAVs.
The sma rt cities initiative w ill create channels of com-
munication, aggregate new types of big data, and gener-
ate opportunities for digital twin services that have not yet
been thought of. Road users may be more willing to plan
multimodal t rips that incorporate mass transit or biking as
the mobility gains become clear. The challenge with this,
as with all new technologies, is to ensure that the final ef-
fect makes our lives easier, better, and happier, rather than
busier, more fragmented, and less private.
This article began as an effort to understand digital twins
and the role they play i n the evermore complex landscape
of CAV modeling and simulation. Digital t wins have en-
joyed a decade of fast grow th and w ide adoption in vari-
ous sectors including manufacturing, health care, and
transportation. They can be characterized using five key
dimensions: the digital model, the physical system, the
communication between them, the big data that is created
from this process, and the serv ices that can be deployed to
add value. Digital twins enjoy the benefits of reduced cost,
reduced power consumption, and miniaturization of sen-
sors brought about by the more fundamental trends of IoT
and cloud computing.
We presented several published applications that used
digital twins in the context of CAVs and ITS. It is true that
some uses of the term do not fully comply with the five di-
mensions, and their status as authentic digital twins can be
called into doubt. These cases usually corresponded to old-
guard methodologies of virtual proving ground and model-
based design rather than new-approaches of IoT and CPS.
However, we find these distinctions to be paper-thin. There
is nothing to prevent the use of five-dimensional digital twin
earlier in the lifecycle of a CAV. Likewise there is no reason
that models and simulation tools created during a vehicle’s
design cannot be deployed as a digital twin during the op-
erational phase of its life. In some cases, it merely requires a
mindset shift to reimagine a tool for development and testing
into a digital twin that continues to add value in new ways.
Simulation-based testing of CAVs will continue to evolve
to encompass methodologies that combine virtual prov-
ing ground, model-based design, CPS, and digital twins.
As such, they defy simple classification, and the commu-
nity should take care not to oversimplify them with an all-
encompassing label. Digital twins foster model reuse and
encourage the use of multiple models at different scales of
resolution. The role of digital twins will continue to grow
as models become more tightly integrated to the physi-
cal systems they represent. This w ill drive such models
towards uniqueness (matching a part icular vehicle or
road), adaptability (evolving with changing conditions and
subject to wear and tear), and interpretability (conveying
useful infor mation to a human user). A maturing IoT in-
frastructure applied to ITS and smart cities will cause the
number of new digital twin services to explode in a num-
ber of unforeseen ways.
About the Authors
Chris Schwarz (chris-schwarz@uiowa
.edu) earned his Ph.D. degree in elec-
trical and computer engineering from
the University of Iowa. He is a research
engineer and Director of Engineering
and Model ing Research at the National
Advanced Driv ing Simulator, Universi-
ty of Iowa, Iowa City, Iowa, 52242, USA. His research inter-
ests include all types of advanced driver assistance
systems, connected vehicles, warning systems, automated
vehicles, and driver impairment modeling. He is a member
of SAE and a Senior Member of IEEE.
Ziran Wang (
earned his Ph.D. degree from the Uni-
versity of California, Riverside. He is a
principal researcher at Toyota Motor
North America R&D, InfoTech Labs,
Mountain View, California, 94043,
USA. His research interests include in-
telligent vehicle technology, cooperative automated driv-
ing, d river beh avior model ing, and veh icula r cyber- physical
systems. He is a Member of IEEE.
[1] B. R. Barricelli, E. Casiraghi, and D. Fogli, “A survey on d igita l twi n:
Definitions, characteristics, applications, and design implications,”
IEEE Access, vol. 7, pp. 167,653167,671, Nov. 2 019, doi: 10.1109/AC-
[2] A. Fuller, Z. Fan, C. Day, and C. Barlow, “ Digit al twin: Enabling
tech nologies, challenges and open resear ch,” IEEE Acc ess, vol. 8, pp.
108, 952108,9 71, May 2020. doi : 10.1109/ACCESS.2020.2998358.
[3] E. Negri, L. Fumagalli, and M . Macchi, “A rev iew of the roles of digital
tw in in CPS-based product ion system s,” Procedi a Manuf., vol. 11, pp.
939948, Jan. 2017, doi : 10.1016/j.promfg.2017.07.198.
[4] M. Shafto et al., “ Modeling, simulation, i nformation technology & pro-
cessing roadmap,” National Aeronautics and Space Administration,
Washi ngton, DC, USA, 2012. [Onli ne]. Available: https://www.nasa
[5] M. Grieves and J. Vickers, “Digit al twin: Mit igating unpred ictable ,
undesirable emergent behavior in complex systems,” in Transdisci-
plin ary Per spect ives on Com plex Sy stems, F. J. Kahlen, S. Flu merfelt,
and A. Alves, Eds. Cham: Springer-Verlag, 20 17, pp. 85113.
[6] M. Liu, S. Fang, H. Dong, and C. Xu, “Review of digital twin about con-
cepts, technologies, a nd industrial applicat ions, J. Manuf. Syst., vol.
58, no. par t B, pp. 3463 61, Jan. 20 21, doi: 10.1016/j.jmsy.2020.06.017.
[7] S. Boschert and R. Rosen, “Digital twin—The simulation aspect,” in
Mechatronic Futures, P. Hehenberger and D. Bradley, Ed s. Cha m,
Switzerland: Springer-Verlag, 2016, pp. 5974.
[8] E. J. Haug, K. K. Choi, J. G. Kuhl, and J. D. War go, “Virtual protot yping
simu lation for design of mec hanical systems,” J. Mech. De s., vol. 117,
no. B, pp. 6370, 1995, doi: 10.1115/1. 2836472.
[9] D. Negrut, R. Serban, A. Elmquist, D. Hatch, E. Nutt, and P. Sheets,
Autonomou s vehicles i n the cyb erspace: Accelerat ing testing via
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
computer simulation,” SAE Internat ional, Warrenda le, PA, USA, SA E
Tech. Pap er 2018-01-1078, 2018.
[10] Y. Zhang, A. Ta ng, T. Palmer, and C . Hazard, “Vi rtua l provi ng
grou nd—An integrated technology for full vehicle a nalysis and simu-
lation,Int. J. Veh. Des ., vol. 21, no. 4/5, pp. 450470, 1999, doi: 10.150 4/
[11] K.-H. Mo, K.-W. Suh, and S.-G. Hong, “New approach in veh icle dura-
bility evaluation, v irtual proving grou nd,” SAE International, Wa rren-
dale, PA, USA , Tech. Paper 2 000 -05- 0030, 2000.
[12] G. G. Wa ng, “Defi nition a nd review of vir tual pr ototy ping,” J. Com put.
Infor m. Sci. E ng., vol. 2, no. 3, pp. 232 236, 2003, doi: 10.1115/1.1526 508.
[13] E. J. Haug, L. Chen, Y. Papelis, and D. Solis, “Vir tual pr oving g round
simu lation for h ighway safety resea rch and vehicle design,” in Virtual
Nonlinear Multibody Systems, W. Schiehlen and M. Valá sˇek, Eds. Dor-
drecht, The Netherlands: Springer-Verlag, 2003, pp. 213232.
[14] C. Schwarz et al., “ Development of an off-road agr icultu ral virtual
proving ground,” in Proc. Dr iving Si mul. Conf., No rth Ame rica ( DSC-
NA 2003), 2003.
[15] H. Albrecht, F. S . Barickman, and S. C . Schnel le, “Advanced t est tools
for ADA S and ADS,” NHTSA, Washington, DC , USA, DOT H S 813 083,
[16] K. Salaani, S . J. Rao, J. G. Howe, D. Elsasser, and S. C. Sch nelle, “An ap-
plied review of simulat ion validation approaches on a veh icle dynam-
ics model,” NHTSA, Washington, DC , USA, Tech. Rep. DOT HS 813 177,
[17 ] P. Struss and C. Price, “Model-based s ystems i n the automotive indus-
try,AI Mag., vol. 24, no. 4, p. 17, 2003, doi: 10.16 09/ai mag.v24i4.172 8.
[18] J. A. Estefan, “Su rvey of model-based systems engineer ing (M BSE)
methodologies,” Incose MBSE Focu s Group, vol . 25, no. 8, pp. 1–12,
[19] A. L. Ramos, J. V. Ferreira, and J. Barceló, “ Model-based systems en-
gineering: An emerging approach for modern systems,IEEE Trans.
Syst ., Man, Cybe rn. C, Ap pl. Rev., vol. 42, no. 1, pp. 101111, 2011, doi:
10.1109/ TSMCC.2011. 2106495.
[20] P. Schoeggl, M. Oswald, R. Vo egl , P. Clement, M. Stol z, and E.
Ramschak, “Driver assist ance and automated d riving: Opportun ities,
challenges, solutions,” in P roc. 30th Int. AV L Conf., 2018 , pp. 7585.
[21] D. Howard, “The dig ital t win: V irtu al validation i n electronics devel-
opment a nd design ,” in Proc. 2019 Pan Pacific Mic roele ctron . Symp.
(Pan Pacif ic), pp. 1–9, doi: 10.23919/Pan Pacific.2019.8696712.
[22] D. Serpanos, “The cyber-physical systems revolution,” Comput er, vol.
51, no. 3, pp. 7073, 2018, doi: 10.1109/ MC.2018.173105 8.
[23] Y. Feng, B. Hu, H. Hao, Y. Gao, Z. Li, and J. Tan, “Design of dist ributed
cyber–physical systems for con nected a nd automated vehicles w ith
implementing methodologies,” I EEE Tra ns. Ind. Informat., vol. 14, no.
9, pp. 4200 4211, 2 018, doi: 10.110 9/TI I.2018.28 05910.
[24] P. Balakrishnan, K . Ramesh Babu, C. D. Naiju, and M. Madiajagan,
Design and implementation of digit al tw in for pred icting failu res in
automobiles usi ng machi ne learning algorith ms,” SAE I nternational ,
Warrendale, PA, USA, Tech. Pap er 2019-28-0159, 2 019.
[25] W. K ritzinger, M. Karner, G. Traar, J. Henjes, and W. Sihn, “Digita l
tw in in ma nufacturing: A categor ical literatu re review and classifi-
cation,IFAC-PapersOnLine, vol. 51, no. 11, pp. 10161022, 2018, doi:
[26] A. Rasheed, O. San, and T. Kvam sda l, “Digital t win: Values, challenges
and enablers from a modeling perspec tive,” IEEE Access, vol. 8, pp.
21,98022,012, Jan. 2020, doi: 10.1109/ACCESS.2020.2970143.
[27] F. Tao et a l., “Five-di mension d igita l twin model and it s ten applica-
tions,Comput. In tegr. Manuf. Syst., vol. 25, no. 1, pp. 1–18, 2019.
[28] Q. Qi et a l., “Enab ling tec hnologies a nd tools for d igital t win,” J. Manuf.
Syst., vol. 58, no. Part B, pp. 3–21, Oct . 29, 2019.
[29] S . Boschert, C. Heinrich, and R. Rosen, “Next generation dig ital twin,”
in Proc. TMCE, Las Pal mas de Gra n Canar ia, Spain, 2018, pp. 209217.
[30] T. H. J. Uhlemann, C. Lehmann, and R. Steinhilper, “The dig ital
tw in: Rea lizi ng the cyber-physical production system for industr y
4.0,” Procedia CIRP, vol. 61, pp. 335340, De c. 2 017, doi: 10.1016/j.
procir.2 016.11.152.
[31] E. Bottani, A. Cammardella, T. Muri no, and S. Vespoli, “From t he cy-
ber-physical system to t he digital tw in: The process development for
behaviour model ling of a c yber guided vehicle in M2M log ic,” in Pr oc.
22nd Su mmer Sc hool Francesco Turc oInd. Sy st. Eng., 2 017, pp. 1–7.
[32] D. Jones, C. Snider, A. Nassehi, J. Yon , and B. Hicks, “Characterising
the digital twin: A systematic literatur e review,” CIRP J. Manuf. Sci.
Tech nol., vol. 29, pp. 3652, May 2020, doi: 10.1016/j.cirpj.2020.02.002.
[33] K. M. Alam and A. E. Saddik, “C2P S: A digital tw in arc hitect ure refer-
ence model for the cloud-based cyber-phy sical sy stems,” IEEE Access,
vol. 5, pp. 20502062, Ja n. 2 017, doi: 10.1109/ACCESS.2017.2657006.
[34] S. Hallerbach, Y. Xia, U. Eberle, and F. Koester, “Simulation-based
ident ificat ion of critical scenarios for cooperat ive and autom ated vehi-
cles,SAE Int. J. C AV, vol. 1, no. 2, pp. 93106, 2018, doi: 10.4271/2018-
01-10 66.
[35] A. A. Korostelkin, O. I. Klyavin, M. V. Aleshin, W. Guodong, W. Suifeng,
and L. Jingyi, “O ptim ization of frame mass in crash testing of off-road
vehicles,” Russian Eng. Res., vol. 39, no. 12, pp. 10 211028, 2019, doi:
10.3103 /S1068798X1912 0116.
[36] Z. Saigol and A. Peters, “Verifyi ng automated d riving systems i n simu-
lation: Framework and challenges,” in Proc . 25th ITS World Co ngr.,
2018 .
[37] V. Dygalo, A. Keller, and A . Shcherbin, “P rinciples of application of vir-
tua l and physical simu lation technolog y in production of d igital twin
of acti ve vehicle safety systems,” Tran sp. Res. Procedia, vol. 50, pp.
121129, Nov. 2020, doi: 10.1016/j.trpro.2020.10.015.
[38] H. Shikata, T. Yam ashita, K . Arai, T. Naka no, K. Hatanaka, and H . Fu-
jikawa, “ Digital twin envi ronment to integrate vehicle simulation and
physical verification,” SEI Tec h. Rev., no. 88, p. 19, 2019.
[39] V. Salehi, J. Taha, and S. Wang, “Application of Mu nich ag ile concept
for MBSE by means of automated valet parking fu nction s and the 3D
environ ment-data,” in P roc. ASME 2020 Int. Des. Eng. Tec h. Conf. Com-
put. Inf. Eng. Conf., vol. 9, p. V009T09A058, doi: 10.1115/DETC2 020 -
[40] I. Barosan, A. A. Basmenj, S. G. Chouhan, and D. Manrique, “Devel-
opment of a v irt ual simulation environment and a digital twi n of an
autonomous driv ing tr uck for a dist ribution c enter,” in Pro c. Eur. Conf.
Softw. Arch it., 2020, pp. 5425 57, doi: 10.1007/978-3-030-59155-7_39.
[41] J. Culley et al., “System design for a dr iverless autonomous rac -
ing vehicle,” i n Proc. 2 020 IEEE 12th In t. Sym p. Commun. Syst.,
Netw. D igit . Signa l Proce ss. (CSNDSP), pp. 1– 6, doi: 10.1109/CSN D-
[42] Y. Ge, Y. Wang , R. Yu, Q. Han, and Y. Chen, “D emo: Resea rch on
test method of autonomous driving based on dig ital t win,” in
Proc. 2019 IEEE Veh. Netw. Conf. (V NC), pp. 1–2, doi: 10.1109/
VNC48 660.2019.9062813.
[43] Y. Feng and H. X. Liu, “Augmented reality for robocars,” IEEE Spectr.,
vol. 56, no. 12, pp. 2227, 2019, doi: 10.1109/MSPEC.2 019.8913 829.
[44] M. Szalai, B. Va rga , T. Tettamanti, and V. Ti hanyi, “M ixed rea lity test
envi ronment for aut onomous ca rs using unity 3D and SUMO,” in Proc .
2020 IE EE 18th World Sy mp. Appl. M ach. Int ell. Informat . (SAMI ), pp.
7378, doi : 10.1109/SA MI48414.20 20.910 8745.
[45] G. Rong et al., “ LGSV L simulator: A hig h fidelity simu lator for autono-
mous driving,” in P roc. IEEE 23rd Int. Conf. I ntell. Trans p., 2020, pp.
1–6, doi: 10.1109/ ITSC 45102. 2020.9294422.
[46] D. J. Fremont et al., “ Formal scenario-based testi ng of autonomou s ve-
hicle s: From si mulat ion to the real world,2020, arXiv:2003.07739.
[47] Y. Laschinsky, K. von Neumann-Cosel, M. Gonter, C. Wegwerth, R. Du-
bitzky, and A . Knoll, “Evaluat ion of an active safety light using v irtu al
test d rive within vehicle in t he loop,” in
Proc. 2010 IEEE Int . Conf. Ind.
Tech nol., pp. 11191112 , doi: 10.1109/ICIT.2010.5472583.
[48] L. Bainbridge, “I ronies of automation,” Automatica, vol. 19, no. 6, pp.
775779, Nov. 1983, doi: 10 .1016/00 05-109 8(83)90046-8 .
[49] X. Chen, E. Kang, S. Shiraishi, V. M . Preciado, and Z. Jiang, “ Digital
behav ioral t wins for safe connected ca rs,” presented at the
Proc. 21th
ACM/ IEEE I nt. Conf. Model Driven Eng. La ng. Syst., Copenhagen,
Denmark, 2018, doi: 10.1145/3239372.3239401.
[50] M. Buechel, M. Schellmann, H. Rosier, T. Kessler, and A. Knoll, “For-
tun a: Presenting t he 5G-con nected aut omated veh icle protot ype of the
project PROVIDENTIA,” 2019.
[51] Y. Liu, Z. Wa ng, K. Han, Z. Shou, P. Tiwari, and J. H. Ha nsen, “Sen-
sor fusion of camera and cloud digital twi n infor mation for intell i-
gent vehicles,” in Proc. I EEE Intel l. Veh. Sy mp., 2020, pp. 182187, doi:
10.1109/ IV47402.2020.93 0464 3.
[52] Z . Wang , X. Liao, X. Zhao, K . Han, P. Tiwari, M.J. Barth, and G. Wu, “A
digit al twin paradigm: Vehicle-to-cloud based advanc ed driver assis-
tance systems,” in P roc. 2020 IEEE 91st Veh. Tech nol. Conf. ( VTC2020 -
Spring), pp. 1–6, doi: 10.1109/ VTC2020-Spring48590.2020.9128938.
[53] C. Schwarz and K. Moran, “Digit al map en hancements of elect ronic
stabilit y control,” SAE I nternational, Warrendale, PA, USA, SAE Tech.
Paper 0148 -7191, 2010.
[54] S. Liu, B . Yu, J. Tang, and Q. Zhu, “Towards fully intelligent transpor-
tation through inf rastr ucture-vehicle cooperative autonomous driv-
ing: C hallenges and opportun ities,” 2 021, arXiv :2103.02176 .
[55] F.-Y. Wan g, N.-N. Zheng, D. Cao, C. M. Martinez, L . Li, and T. L iu, “Par-
allel driv ing in CPS S: A unif ied approach for transport autom ation and
vehicle intelligence,” IEEE/CAA J. Automatica Sinica, vol. 4, no. 4, pp.
577587, 20 17, doi: 10.1109/ JAS.2017.7510598.
[56] F.-Y. Wang, “Control 5.0: From Newton to Merton in popper’s cyber-
social-physical spaces,” IEEE/CAA J. Automatica Sinica, vol. 3, no. 3,
pp. 233234, 2 016, doi: 10 .1109/JAS.2016.750879 6.
Authorized licensed use limited to: Univ of Calif Riverside. Downloaded on January 13,2022 at 21:29:34 UTC from IEEE Xplore. Restrictions apply.
... The Digital Twin concept has been loosely defined and adopted in the transportation domain since its emergence, partly due to its similarity and connection with other technologies. In particular, the confusion among the roles of iteration (i.e., switches back and forth between the physical and digital spaces), model-based design (i.e., starts with digital components and incrementally swaps in physical components), and Digital Twins (i.e., maintain synchronized versions of a physical system and its digital counterpart) in connected vehicles are well discussed in the survey paper by Schwarz and Wang [32]. Nonetheless, many previous efforts related to the IoT and CPS in the automotive industry envision the development of the Digital Twin, since the majority of those proposed methodologies and/or algorithms were developed on multilayer system frameworks with physical entities (i.e., vehicles) and their digital replicas (simulation models/environments). ...
... The major difference between a Digital Twin framework with an iteration framework, or a model-based design framework is that, a Digital Twin framework always maintains synchronized versions of the physical system and its digital counterpart [32]. In the proposed MDT framework of our study, this is guaranteed by the communication plane between the physical and digital spaces. ...
Full-text available
A Digital Twin is a digital replica of a living or nonliving physical entity, and this emerging technology attracted extensive attention from different industries during the past decade. Although a few Digital Twin studies have been conducted in the transportation domain very recently, there is no systematic research with a holistic framework connecting various mobility entities together. In this study, a Mobility Digital Twin (MDT) framework is developed, which is defined as an Artificial Intelligence (AI)-based data-driven cloud-edge-device framework for mobility services. This MDT consists of three building blocks in the physical space (namely Human, Vehicle, and Traffic), and their associated Digital Twins in the digital space. An example cloud-edge architecture is built with Amazon Web Services (AWS) to accommodate the proposed MDT framework and to fulfill its digital functionalities of storage, modeling, learning, simulation, and prediction. The effectiveness of the MDT framework is shown through the case study of three digital building blocks with their key micro-services: the Human Digital Twin with user management and driver type classification, the Vehicle Digital Twin with cloud-based Advanced Driver-Assistance Systems (ADAS), and the Traffic Digital Twin with traffic flow monitoring and variable speed limit. Future challenges of the proposed MDT framework are discussed towards the end of the paper, including standardization, AI for computing, public or private cloud service, and network heterogeneity.
... Recently, there have been some studies on DT and autonomous vehicles. For example, the role of DTs in connected and automated vehicles is discussed in [44]. Further, Reference [45] discusses a framework for vehicular digital twins, which includes data collection, data processing, and analytics phases. ...
Full-text available
The Internet of Vehicles (IoV), where people, fleets of electric vehicles (EVs), utility, power grids, distributed renewable energy, and communications and computing infrastructures are connected, has emerged as the next big leap in smart grids and city sectors for a sustainable society. Meanwhile, decentralized and complex grid edge faces many challenges for planning, operation, and management of power systems. Therefore, providing a reliable communications infrastructure is vital. The fourth industrial revolution, that is, a cyber-physical system in conjunction with the Internet of Things (IoT) and coexistence of edge (fog) and cloud computing brings new ways of dealing with such challenges and helps maximize the benefits of power grids. From this perspective, as a use case of IoV, we present a cloud-based EV charging framework to tackle issues of high demand in charging stations during peak hours. A price incentive scheme and another scheme, electricity supply expansion, are presented and compared with the baseline. The results demonstrate that the proposed hierarchical models improve the system performance and the quality of service (QoS) for EV customers. The proposed methods can efficiently assist system operators in managing the system design and grid stability. Further, to shed light on emerging technologies for smart and connected EVs, we elaborate on seven major trends: decentralized energy trading based on blockchain and distributed ledger technology, behavioral science and behavioral economics, artificial and computational intelligence and its applications, digital twins of IoV, software-defined IoVs, and intelligent EV charging with information-centric networking, and parking lot microgrids and EV-based virtual storage. We have also discussed some of the potential research issues in IoV to further study IoV. The integration of communications, modern power system management, EV control management, and computing technologies for IoV are crucial for grid stability and large-scale EV charging networks.
Conference Paper
Full-text available
Testing autonomous driving algorithms on real autonomous vehicles is extremely costly and many researchers and developers in the field cannot afford a real car and the corresponding sensors. Although several free and open-source autonomous driving stacks, such as Autoware and Apollo are available, choices of open-source simulators to use with them are limited. In this paper, we introduce the LGSVL Simulator which is a high fidelity simulator for autonomous driving. The simulator engine provides end-to-end, full-stack simulation which is ready to be hooked up to Autoware and Apollo. In addition, simulator tools are provided with the core simulation engine which allow users to easily customize sensors, create new types of controllable objects, replace some modules in the core simulator, and create digital twins of particular environments.
Conference Paper
Full-text available
In today’s complex world of development of functions for automating Driving Systems (ADS), methods, tools, systems and new approaches are necessary for a seamless application. Furthermore, it is important to apply new technics of simulation and visualization (Digital Twin) for the new ADS functions. To prototype and to test these functions in a physical manner is not only a costly and complex effort but also encounters legal and bureaucratic obstacles. The importance of simulation is very high. For that reason, this paper and corresponding research project will develop a consistent traceable System Engineering approach of autonomous driving functions and its environment based on Munich Agile Concept for Model-Based-Systems-Engineering (MBSE). MBSE is based on three important core pillar which is 1) Methods/Processes, 2) Language and 3) Systems. The purpose of the new developed Munich Agile Concept Approach is to handle the complexity over the entire ADS feature development from the system requirement definition process up to the test and validation of the system. The Munich Agile Concept contains six different levels which are System Requirement-, System Functional-, System Architecture-, System Validation-, System Test and the System-Usage-Level. For defining the first three-level, a graphical language called System Modelling Language (SysML) has been applied.
Full-text available
The creation of a digital twin requires appropriate content for a mathematical model of all the relevant processes and phenomena, correlating with the processes that occur within the real physical entities. A digital twin of a vehicle includes a model of the braking system, which is essential for ensuring safety. Modern automated braking systems are difficult to reproduce in a simulation, because of the sheer number of factors involved. Furthermore, there are additional processes that also have a substantial impact on both the system’s performance and the vehicle position in physical space. This paper covers the principles of applying a virtual and physical simulation technology to the production of a digital twin of active vehicle safety systems. We begin our study by analyzing the Vehicle–Driver–Road system and ranking its elements. Within the Vehicle system, in turn, we also build a hierarchy of the subsystems relevant to the simulation, ranking them by priority level. In the course of system analysis, we arrange all modules by priority, considering the traffic conditions. In the case of this particular study, we are dealing with the braking mode, and therefore give priority to the braking system module over the other subsystems and modules. We also suggest various ways of structuring the model of the braking system itself, depending on the task. The tasks are grouped by difficulty, in ascending order, from the task of designing an algorithm for controlling a single wheel to the task of controlling an entire automated braking system of a multi-wheel vehicle, across all possible automation levels (i.e. both the self-driving mode and the operation mode involving a real driver with unique physical and psychological characteristics). The virtual and physical simulation technology, as exemplified by the braking system, enables the improvement and adjustment of digital twins of both current and future vehicles.
Conference Paper
Full-text available
The rising popularity of autonomous vehicles has led to the development of driverless racing cars, where the competitive nature of motorsport has the potential to drive innovations in autonomous vehicle technology. The challenge of racing requires the sensors, object detection and vehicle control systems to work together at the highest possible speed and computational efficiency. This paper describes an autonomous driving system for a self-driving racing vehicle application using a modest sensor suite coupled with accessible processing hardware, with an object detection system capable of a frame rate of 25fps, and a mean average precision of 92%. A modelling tool is developed in open-source software for real-time dynamic simulation of the autonomous vehicle and associated sensors, which is fully interchangeable with the real vehicle. The simulator provides performance metrics, which enables accelerated and enhanced quantitative analysis, tuning and optimisation of the autonomous control system algorithms. A design study demonstrates the ability of the simulation to assist in control system parameter tuning-resulting in a 12% reduction in lap time, and an average velocity of 25 km/h-indicating the value of using simulation for the optimisation of multiple parameters in the autonomous control system.
Full-text available
Various kinds of engineering software and digitalized equipment are widely applied through the lifecycle of industrial products. As a result, massive data of different types are being produced. However, these data are hysteretic and isolated from each other, leading to low efficiency and low utilization of these valuable data. Simulation based on theoretical and static model has been a conventional and powerful tool for the verification, validation, and optimization of a system in its early planning stage, but no attention is paid to the simulation application during system run-time. With the development of new-generation information and digitalization technologies, more data can be collected, and it is time to find a way for the deep application of all these data. As a result, the concept of digital twin has aroused much concern and is developing rapidly. Dispute and discussions around concepts, paradigms, frameworks, applications, and technologies of digital twin are on the rise both in academic and industrial communities. After a complete search of several databases and careful selection according to the proposed criteria, 240 academic publications about digital twin are identified and classified. This paper conducts a comprehensive and in-depth review of these literatures to analyze digital twin from the perspective of concepts, technologies, and industrial applications. Research status, evolution of the concept, key enabling technologies of three aspects, and fifteen kinds of industrial applications in respective lifecycle phase are demonstrated in detail. Based on this, observations and future work recommendations for digital twin research are presented in the form of different lifecycle phases.
Conference Paper
Full-text available
With the rapid development of intelligent vehicles and Advanced Driving Assistance Systems (ADAS), a mixed level of human driver engagements is involved in the transportation system. Visual guidance for drivers is essential under this situation to prevent potential risks. To advance the development of visual guidance systems, we introduce a novel sensor fusion methodology, integrating camera image and Digital Twin knowledge from the cloud. Target vehicle bounding box is drawn and matched by combining results of object detector running on ego vehicle and position information from the cloud. The best matching result, with a 79.2% accuracy under 0.7 Intersection over Union (IoU) threshold, is obtained with depth image served as an additional feature source. Game engine-based simulation results also reveal that the visual guidance system could improve driving safety significantly cooperate with the cloud Digital Twin system.
Conference Paper
Full-text available
Developing a vehicle is always a long and complex task. This is especially true for autonomous cars. Tasks performed by the driver are taken over by the vehicle and must be performed with maximum reliability. Developing these systems is a difficult task, especially due to limited testing capabilities. Testing a vehicle in a closed environment is safe and controlled, but the variety of test scenarios is limited. By using mixed reality environments, one can create a diverse environment around a real test vehicle, with traffic, obstacles, and unexpected situations. The real movement of the test vehicle allows testing decision and motion planning level vehicular functions. Information from the virtual world can be considered as input to the vehicle sensor. Mixed reality or digital twin simulation environments can greatly assist the autonomous vehicle development process and also serve as a basis for validation procedures for such systems. This article introduces a mixed reality simulation environment that integrates a real test vehicle into a virtual environment, can handle other real and virtual obstacles, contains traffic simulations, and visualizes all of these.
Full-text available
Digital Twin technology is an emerging concept that has become the centre of attention for industry and, in more recent years, academia. The advancements in industry 4.0 concepts have facilitated its growth, particularly in the manufacturing industry. The Digital Twin is defined extensively but is best described as the effortless integration of data between a physical and virtual machine in either direction. The challenges, applications, and enabling technologies for Artificial Intelligence, Internet of Things (IoT) and Digital Twins are presented. A review of publications relating to Digital Twins is performed, producing a categorical review of recent papers. The review has categorised them by research areas: manufacturing, healthcare and smart cities, discussing a range of papers that reflect these areas and the current state of research. The paper provides an assessment of the enabling technologies, challenges and open research for Digital Twins.