Content uploaded by Flavio Farroni
Author content
All content in this area was uploaded by Flavio Farroni on Dec 08, 2015
Content may be subject to copyright.
Abstract— This paper presents some results on the development
and testing of new solutions in the field of driving automation. The
introduction of increasing levels of vehicle automation aimed at
enhancing road safety requires a renewed approach to the research
and development process and needs a multi-actor environment where
the innovation can be tested. Indeed, vehicle automation spans
several scientific disciplines and it is becoming exceedingly difficult
and too costly for a single research innovation team to go in depth
into all technologies and solutions. This is shifting the innovation
process toward a multidisciplinary approach in which the only way to
ensure an easy, rapid, efficient and scalable introduction of the
required innovation is to adopt integrated and complex testing
platforms for the simulation of automation solutions, based on a
modular architecture, where independent components can be
developed and then integrated and tested in a multi-actor
environment. A platform for virtual testing is presented herein and
employed to assess the performance of an integrated driving
assistance solution based on computing appropriate surrogate
measures of safety that allow for the transition between different
automation logics in free-flow, car-following and emergency braking
conditions.
Keywords— Intelligent Transportation Systems; Advanced
Driving Assistance Systems; Autonomous Emergency Braking;
Driving Automation; Road Safety; Driving Behaviour; Adaptive
Cruise Control, Intelligent Speed Adaptation.
I. INTRODUCTION
DVANCED Driving Assistance Systems (ADAS) have been
designed, developed and tested for several years, and the
development of further devices is under way. ADAS are
increasingly adopted under an active safety paradigm. Indeed,
although innovation in the field of passive safety is yet to be
implemented, its growth is likely to be outperformed by that of
active safety, aiming to prevent collisions rather than mitigate
This work was partially supported by the DRIVE IN2 project
(B61H110004000005/PON01_00744) and by the APPS4SAFETY project
(B68F12001150005/PON03PE_00159_3 (sponsor and financial support
acknowledgment goes here).
Maria Russo Spena is with the Dipartimento di Ingegneria Civile ed
Ambientale, University of Naples “Federico II”, 80125 Naples – ITALY
(corresponding author to provide phone: +39 081 76 85956; e-mail:
maria.russospena@unina.it).
Francecso Timpone is with the Dipartimento di Ingegneria Industriale,
University of Naples “Federico II”, 80125 Naples – ITALY (e-mail:
francesco.timpone@unina.it).
Flavio Farroni is with the Dipartimento di Ingegneria Industriale,
University of Naples “Federico II”, 80125 Naples – ITALY (e-mail:
flavio.farroni@unina.it).
effects. Every day on Europe’s roads 71 people continue to die
in traffic accidents, in which drivers are acknowledged to play
a crucial role. It is generally estimated that around 90% of
road accidents are correlated with human error.
Studies [1] carried out in the USA show that a large
proportion of traffic accidents can be caused both by the
driver’s distraction or inadequacy with respect to the traffic
conditions, and by the driver’s incorrect interaction with
primary and secondary driving commands or other on-board
and personal devices. The driver’s inadequacy is particularly
dangerous when the vehicle cruises in a dense traffic stream
and, as the traffic increases, can be exacerbated by increasing
levels of inter-vehicle interactions. Errors and accidents also
occur because of performance errors (like overcompensation,
inappropriate directional control, etc.). Automated driving at
different levels has the potential to reduce accidents by
reducing the impact of the human factor [2], thereby
contributing in the long term to the reduction in road fatalities
[3]. The problem is also serious at intersections, with
automation attempts being suggested for such cases ( [4], [5] ).
Indeed, in the near future rapid development is expected of
medium-to-high automated vehicles, a phenomenon which will
be boosted by the spread of car fleets owned by big market
players and used by drivers on the basis of a non-ownership
approach ( [6] ). However, the more advanced the system is,
the more complex the integration in the vehicle will be, and the
overload of information to the driver can sometimes be
unproductive. There are common concerns in the human factor
community that ADAS may fail to alleviate the workload and
can even introduce a new source of workload due to the need
to attend to new tasks. Thus, ADAS need to be carefully tested
before being implemented on vehicles ( [7] ). Moreover, they
should not be seen (at least, not yet) as a substitute for drivers,
but as a co-driver that does not exclude the driver from the
control loop even if his/her direct involvement in an increasing
number of driving tasks can be replaced by automation [8].
The driver should still be able to intuitively understand the
logics of automated driving assistance [9]. This is even more
crucial [10] when interaction between the driver and the
automation is required, for example when the system acts as a
pure warning facility and/or the driver has to regain control of
the vehicle in the transition from automated driving to manual.
In these events the well-known out-of-the-loop syndrome [11]
can occur in the human interaction with automation,
characterized by poor mental models, low reaction times, and
low accuracy in vehicle control reclaim.
Virtual Testing of Advanced Driving
Assistance Systems
Maria Russo Spena, Francesco Timpone, and Flavio Farroni
A
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
300
The above argument does not mean that the potential
positive impacts of automation should be left out of the
equation; rather, full automation is a long-term objective [12]
and the full benefits can only be gained in the long run.
Potential benefits can be achieved by gradually shifting the
main driving role from the driver to the vehicle. Moreover, for
acceptance of increasingly higher levels of interaction and
automation, we need to understand drivers’ needs as well as
possible reactions of the automated vehicle to both human and
automatic driving control logics. This issue arises at an early
stage in the development and deployment process, before the
pre-commercial testing phase at which ADAS are commonly
considered in the car-making process. Indeed, several
paradigms (and formal procedures too) have been established
for pre-commercial testing of automation/assistance solutions,
targeting the deployment of advanced components. These
procedures are employed by car-makers and tier-one suppliers
and range from application of international standards (e.g.
ISO) to National Highway Traffic Safety Administration
(NHTSA) testing, and from OEM internal processes to
European New Car Assessment Programme (EuroNCAP)
tests. These involve car-makers at the deployment phase and
do not take into account at an earlier stage the possible
adoption of different solutions, technologies or different ways
to integrate the technologies with the vehicle. At the earlier
phases of development and adoption, car-makers and tier-one
suppliers rely on different tests, many of which are
implemented thanks to appropriate simulation platforms. In all
cases, the test scenarios aim to assess system performance
under different traffic conditions and/or different pre-defined
manoeuvres. In these contexts the vision of the driver’s role is
too often just to solicit the system, and the driver’s model is
intended as the mathematical or procedural representation of
such solicitations. The result is that testing scenarios lack
realism in terms of simulation of second-order effects, which
are those related to driver’s reactions to solicitations. Our
work aims to reverse the previous concept and to put the driver
side-by-side with the vehicle at the centre of the car-making
process. This enables ADAS testing in fully realistic scenarios,
starting from the earliest stages of the development and
adoption process, allowing the development of safer, more
robust, efficient and widely accepted/adopted ADAS solutions.
However, such a holistic and driver-centric approach
exacerbates the current trend that involves an increasing
number of technical disciplines in the automotive process,
making it exceedingly difficult and costly for a development
unit to treat all related technologies and study all the
consequences on the vehicle and on the driver but also on the
traffic and on traffic propagation on infrastructures and
networks. Indeed, different driving mechanisms interact in
terms of flow propagation, which could impact on how the
flow is propagated and on how the network has to be adapted (
[13], [14]) in order to take such effects into account. Indirect
impacts cam have effect at a transportation-planning level too
[15], as capacities and performances of infrastructures can be
drastically changed and the trade-off between technological
innovation and transportation supply management becomes a
real option.
In such a context, the only way to ensure an easy, fast,
efficient and scalable introduction into the market for the
required innovation is to adopt an appropriate development
and testing platform, based on a flexible architecture, where
the (reactive) driver’s model is central and independent
components can be developed and then exploited in a multi-
actor environment. A representation of this integrated platform
is shown in figure 1 below.
Fig. 1 – Integration platform
In this paper we present a first attempt to develop an integrated
platform based on a multidisciplinary and modular approach.
In section II the best-candidate platform is identified with
respect to the desired characteristics. The critical issues in
setting up the platform are verified and a second-best solution
identified for this early stage of development. Two of the main
modules of the platform, vehicle dynamics and the driver
model, are considered and developed for use in ADAS design.
In section III some general points are made on the
development of driving automation and three ADAS solutions
are identified, contextualized within the general automation
framework, and designed to be integrated synergistically. In
section IV the integrated ADAS solution is tested by means of
the developed platform and the results presented and
discussed.
II. SIMULATION PLATFORM
In order to allow for the necessary modular and
multidisciplinary approach, the platform requires some
particular characteristics. One of these is the integration layer;
it should have an appropriate level of flexibility (and/or
completeness) that allows the integration of different
multidisciplinarily developed modules. Analysis of the
available options steered our development towards adopting a
driving-simulation environment, namely Scaner Studio, by
Oktal [16], already installed on the car driving-simulation suite
(two twin compact driving simulators and a high-fidelity
dynamic one) at the University of Naples Federico II ( [17],
[18]). Indeed, it is peculiar to professional driving simulators
to possess a very sophisticated simulation environment for
road scenarios and traffic conditions, as well as to reproduce
driving feedback realistically (for instance with respect to
steering and braking, [19]). Obviously, for all previous aims, a
detailed representation of the vehicle dynamics is inherently
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
301
required in driving simulations ( [20] ). In our case, the Oktal
simulation environment is based on the Callas platform [21],
[22], which allows for a detailed characterization of the
vehicle dynamics. Callas software is a realistic simulator
validated by car-makers (including PSA), and research
institutions including IFSTTAR (formerly INRETS). The
Callas model also takes into account vertical dynamics
(suspension, tyres), kinematics, elasto-kinematics, tyre
adhesion and aerodynamics.
Finally, the driving-simulation environment allows co-
simulation, that is integration into the simulation environment
of interacting control logics thanks to APIs developed in
Matlab/Simulink.
The driving-simulation environment is our best-choice as an
integration platform. Obviously, it requires the development
and/or fine-tuning of a great variety of sub-models. Our choice
here is to focus first on two such models: the vehicle dynamics
and the driver.
Given the early stage of development of our platform and
the chosen modules to be developed, a simpler integration
platform can be employed. In particular, the Matlab/Simulink
environment (that is contained in the Oktal platform) was used
in this pilot work without loss of generality.
As a consequence, the development in the Matlab/Simulink
environment of the driver model and the vehicle dynamic
model is described below.
A. The driver model
By driver model we mean the mathematical representation
of the driver’s reaction to stimuli from both the traffic and the
driving assistance system. It is worth noting that at this early
stage a driver model is used, but our (fully) integrated platform
will allow interaction with the simulation of a real driver.
In this section we develop a driving model intended to
simulate the reaction to a collision warning system (CWS). In
our scheme, in car-following conditions the headway with
respect to the leading vehicle can reach a value that activates
the intervention of a CWS; the CWS warns the driver to focus
on a potentially unsafe time-headway. Once the warning signal
is raised, the driver’s reaction can vary according to the actual
headway with respect to the leading vehicle. In other words, if
the warning signal is raised later (for shorter headways), the
driver reacts more, while if the signal is raised very cautiously,
the driver tends to react less or more slowly.
For the driver model, let:
H be the actual time headway with respect to the vehicle ahead,
which can be computed as H=sp/Vf = (Δx+L)/Vf
Vf the actual speed of the vehicle;
sp the gross-spacing with respect to the vehicle ahead;
Δx the net spacing (measured bumper to bumper) with respect to
the vehicle ahead;
L the length of the vehicle ahead, that can be approximated with
average vehicle length;
fr the net spacing corresponding to the minimum headway, which
can be fixed as an external parameter, for instance at the value
of 1 m;
Hm the minimum vehicle headway that can occur during car-
following trajectories, for very short transients and typically
before overtaking takes place; it can be computed as
H=(fr+L)/Vf
Hs the time-headway that is considered to be fully safe, that is the
value that does not result in any stimulus to the driver. The
stimulus is computed only for values of current headway that
are less than Hs. Otherwise the stimulus is null; this value has
to be computed by observation of driving styles and is
dispersed across the population of drivers; it can also be
considered to correspond to the threshold of perception of a
slower vehicle ahead, according to the action point theory (
[23], [24], [25]);
ST the stimulus received by the driver because of the leading
vehicle;
RP the probability the driver actually reacts to a warning raised by
the assistance system, depending on the stimulus (ST);
RI the intensity of the driver’s reaction, also depending on the
stimulus (ST);
a the deceleration reaction the driver applies once he/she reacts
to the stimulus;
amax the maximum deceleration reaction the driver applies (here we
fix a value of – 4 m/s2);
amin a minimal value for the deceleration reaction. Below this
threshold the driver does not actually apply any deceleration
(here we fix a value of – 0.5 m/s2).
It is worth noting that the spacing can be measured by a
forward radar mounted on the front of the vehicle, which is
becoming an increasingly popular (and relatively cheap)
device in automotive research. The other measurement
required in real time is the cruising speed, which is a trivial
measure supplied by the vehicle’s on-board system.
Our model proposes to compute the stimulus with a logistic-
type function as follows:
)( 1
11
)(
minmax
min
)2(
STST
ST
e
hST
h
−
⋅
−
+
=
−−
αα
(1)
where α is a modelling parameter to be estimated and ST is a
function of the actual headway H via the standard headway h,
defined as:
)( )(
min
HH HH
h
s
s
−
−
=
(2)
The equation for ST(h) gives standardized stimuli, in the
range from 0 and 1. To this aim the maximum and minimum
stimuli have to be computed, corresponding to the values of
the (non-standardized) stimuli at the minimum (0) and
maximum (1) standard headway (h):
α
e
ST +
=11
min
(3)
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
302
α
−
+
=e
ST 11
max
(4)
The probability of responding to a warning raised from the
driving assistance system is also computed by using a logistic-
type function:
)( 1
11
)(
minmax
min
)2(
RPRP
RP
e
STRP
ST
−
⋅
−
+
=
−⋅−
ββ
(5)
Once again, RP(ST) gives a standardized probability
response, in the range from 0 and 1, and the values of the non-
standardized response probability corresponding to the
minimum (0) and maximum (1) standard stimulus have to be
computed:
β
e
RP +
=11
min
(6)
β
−
+
=e
RP 11
max
(7)
The standard response intensity (RI) is computed as a
function of the stimulus by using a logistic-type function as
well. In this case not only are the minimum and maximum
(non-standard) response intensity computed to properly scale
the result, but the minimum standard intensity of the response
(γ) is also fixed (it is assumed that if the driver reacts a
minimum reaction intensity is attained). The equation for RI is:
1)1(
)II( 1
I
11
)(
minmax
min
)2(
+−⋅
−
⋅
−
+
=
−⋅−
γ
δδ
RR
R
e
STRI
ST
(8)
where:
δ
e
R+
=11
I
min
(9)
δ
−
+
=e
R11
Imax
(10)
Finally, the deceleration the driver applies in response to a
warning raised by the assistance system depends on the
probability of a response being applied times the intensity of
this reaction; the resulting value (in the range from γ to 1) is
multiplied by the driver’s maximum applicable deceleration
(amax here fixed at the value |-4| m/s2). It is also assumed that if
deceleration is applied it is unlikely to be less than a minimum
value (amin here fixed at |-0.5| m/s2). In analytical terms:
( )
maxmin
))(())((,max)( ahSTRIhSTRPaha ⋅⋅=
(11)
An example of the resulting function for the applied
deceleration is depicted in figure 2 below, where the modelling
parameters were fixed at α = 3, β = 2, δ = 6, γ = 0.6
Fig. 2 – Results of the driver model
B. The model for the vehicle dynamics
To carry out a more realistic simulation a vehicle dynamics
model was coupled with the ADAS model. The purpose of the
implemented vehicle dynamics model is to simulate, in real
time, the vehicle’s dynamic behaviour, thanks to continuous
integration of the balance equations regulating the longitudinal
and lateral vehicle motion.
In order to make the simulations as close to reality as
possible, the implemented mathematical model also includes
working dynamics for wheels, tyres, the engine and braking
system, as well as a control system for automatic gearbox
activation. Furthermore, a model able to provide an estimation
of real-time fuel consumption of the vehicle in question is
intended to be developed: such a model is embedded in the
vehicle model discussed in the following, a further research
purpose being the optimization of consumption performance
during urban driving.
The model does not only involve vehicle dynamics, but it
represents a general vehicle-behaviour model: besides the
‘pure’ vehicle dynamics equations, also the modelling of some
other essential vehicle components (such as the engine,
gearbox, braking system and others) was performed. In order
to make the vehicle dynamics model clearer and more
readable, it was organized into different sub-systems.
The main subsystem contains all the equations describing
the vehicle dynamics: the Longitudinal Balance Equation,
Lateral and Yaw Vehicle Dynamics and Vertical Load
Determination, taking into account load transfers by means of
the handling diagram/phase plane approach described in [26]
and of the influence of roll stiffness [27]. In the input/output
subsystem all the model’s exchange parameters with the other
subsystems and in particular with the ADAS model are
suitably adjusted and displayed. In the I/O subsystem also the
errors affecting the signals referring to the control variables
are calculated, with reference to the target ones, so that the
instantaneous error value can be deduced at any given time.
Error determination is essential for the vehicle dynamics
model, as this variable stands for the input signal controlling
the processor units, designated to monitor the vehicle’s
behaviour and to make the decisions to adapt its working
conditions to the desired ones (e.g. action of throttle and
braking system). In the control unit subsystem the vehicle’s
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
303
real components and systems behaviour were introduced in
order to make the final model as close to reality as possible. In
particular, this subsystem allows introduction of the engine’s
dynamic behaviour, as well as the behaviour of the braking
system, complete with an Antilock Braking System (ABS)
unit. The ABS working principle is based on the wheels’ speed
detection (through dedicated sensors fitted on board), signal
processing and a set of actuators directly operating the
hydraulic braking system. The implemented ABS logic can be
easily de-activated by simply operating on dedicated manual
switches, specifically inserted in the model. Thanks to ABS
logic, it is even possible to take into account the possibility of
tyres locking, following the very intense application of braking
and/or poor traction driving conditions (e.g. on snow, ice, rain
etc.). As regards the variations in road frictional conditions,
fundamental in ABS actuation, a specific physical grip model
was employed [28]. In the ABS block two proportional-
integral controllers were used to estimate the magnitude of the
action to be performed on the vehicle’s systems to let it move
according to the ADAS control logics. The input signal for
these controllers is the error magnitude, while the output signal
is the action which has to be carried out in order to minimize
the error itself. In the tyre subsystem, an essential part of the
whole vehicle model, an evolved MF set of equations able to
simulate wheel behaviour was introduced.
In this subsystem it is possible to determine the tyres’ actual
grip at any one time and to calculate the longitudinal and
lateral interaction forces the tyre is able to perform. The
above-mentioned interaction forces are the friction actions the
tyres exert on the ground, depending both on the local friction
coefficient inside the contact patch, strongly connected to
fundamental thermodynamic phenomena observable at the
tyre/road interface [29], [30], both on the vertical force acting
on the tyre, hence even depending on the load transfer during
vehicle handling. Knowledge of these forces is essential, since
they deeply affect the vehicle’s handling and road-holding: this
means that these forces are intimately connected to vehicle
safety, and hints at the importance of this subsystem and of the
equations it aims to introduce.
III. DRIVING ASSISTANCE AUTOMATION
Reconciling mobility needs with efficient and more
sustainable transportation is a key objective that includes
increased levels of road safety. Accident analyses have shown
that human factors are responsible for up to 90% of road
accidents [2]. Therefore, automatic driving systems (ADS) are
considered one of the possible ways to reduce the road
fatalities. Automation levels have been classified; for instance,
as reported by ERTRAC [12], the Society of Automotive
Engineers (SAE) considers six automation levels, ranging from
0 to 5. These approximately correspond to the five levels
(from 0 to 4) established by the National Highway Traffic
Safety Administration (NHTSA) [31], with SAE levels 4 and 5
corresponding to NHTSA level 4. Level 0 corresponds to no
automation and level 5 to full automation. At automation level
1 (Driver Assistance) the vehicle is able to assist the human
driver in some driving tasks related to both
acceleration/deceleration and steering, but the main
responsibility for monitoring the driving context is up to the
driver, who is also in charge of fallback. At level 2 (partial
automation) the vehicle accomplishes some specific driving
tasks (both longitudinal and lateral driving), while monitoring
and fallback are in charge of the driver. At level 3 (conditional
automation) the vehicle executes driving tasks and monitors
the driving environment. This applies to specific driving
modes and the driver is required to resume the control in the
case of fallback. At level 4 (high automation) the vehicle is
responsible for all actions (including fallback) but this is
applied to specific driving modes. At level 5 (full automation)
the definitions of level 4 apply to all driving modes.
Various levels of driving automation have already started to
be progressively introduced in the automotive arena [32].
Level 1 of automation has been widely applied for several
years (e.g. adaptive cruise control, lane keeping assistance,
etc.), level 2 systems have emerged (e.g. automated parking,
adaptive cruise control with stop-and-go and/or truck
platooning) and introduction of level 3 is now discussed (e.g.
combination of adaptive cruise control and lane changing).
Here we present a combination of ADAS solutions that
work synergistically. This approach allows the resulting
integrated solution to be classified at automation level 3. The
main idea is to define the field of application of each of these
systems appropriately in order to ensure the most correct
application of any of them and the most suitable transition
from one to the other. Intelligent speed adaptation (ISA),
advanced cruise control (ACC); collision warning system
(CWS) and, finally, autonomous emergency braking (AEB)
are applied according to appropriate surrogate measures of
safety, evaluated at run-time. Moreover, the systems are
applied in such a way that the driver perceives for most of the
time (that is when the ACC runs) that the automation is
human-like and consistent with his/her own behaviour. This
ensures that the driver is always in-the-loop of the driving
control process. The driver interacts with the automation
system in the case of intervention of the CWS; in this event the
driver’s reaction is simulated by adopting the driver model
described in section II.A. All control algorithms result in the
need to apply accelerations and decelerations to the vehicle.
These represent the stimuli viewed as from the vehicle (and
not the driver). Application of these stimuli and the resulting
effects on the vehicle are simulated with the support of the
vehicle dynamics model described in section II.B.
A. Integrated Logic Control
Integrated control logic is based on the run-time evaluation
of headway (H) and time-to-collision (TTC), which are based
on continuous measurement of the follower’s speed and
relative speed and spacing with respect to the vehicle ahead.
TTC is computed when the relative speed, measured as the
leader’s speed minus the follower’s one, is lower than zero,
that is where the follower approaches the leader. It is the time
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
304
after which a collision between the two vehicles will occur if
the collision course and speed difference are maintained [33].
H and TTC can be computed at each instant by:
H=(Δx+L)/Vf and TTC= Δx/ Δv
where, in addition to variables already introduced in section
II.A, Δv is the relative speed with respect to the vehicle ahead.
The relative speed can also be measured by a forward radar. It
is worth noting that H and TTC are also considered surrogate
measures of safety [34] [35]. Hence they are appropriate to
manage the transition between different solutions that are
conceived for different safety-related conditions and tasks. The
core of our integrated system is an ACC. Once activated, it
allows the driver to set a desired cruising speed. This speed is
maintained in free-flow traffic conditions until a slower vehicle
ahead influences the controlled vehicle. From this point on, the
vehicle runs in a car-following condition. If active, an ISA
overrides the driver’s desired speed and suggests to the ACC
the speed posted by a digital map or by an I2V communication
system. The need for the driver to regain control of the vehicle
can be requested by the on-board CWS. Indeed, if for any
reason, the headway (H) between the controlled vehicle and
the vehicle ahead is below a critical threshold, the CWS asks
the driver to re-establish conditions of greater safety. The
CWS acts as a real safety assistant, monitoring unsafe
headways and warning the driver to regain control of the
vehicle; the driver model simulates the response to such a
warning. Should not only the headway be below a given
threshold but also the TTC decrease below another threshold,
the danger is not only potential but also imminent. In this
event, the reaction has to be as prompt (and intense) as
possible, quicker than the perception and reaction time of the
driver to the CWS warning. This is the role played by the
AEB, which automates braking in order to avoid the incident
or (more likely) to reduce impact, damage and injuries. The
overall (integrated) control logic is depicted in figure 3 below.
NO
ISA
(posted-speed)
Driver’s
desired speed
Is ISA
activated?
Is the
integrated
system
activated?
Human
reaction
Free-flow
target speed
ACC
Is headway
critical?
CWS
NO
YES
YES
NO
YES
NO
YES
Is TTC
critical?
AEB
YES
Does the
driver react
to the
warning
NO
Fig. 3 - Integration logic
B. Behavioural models for ADAS
The transition and integration across the different ADAS
logics is ensured by the headway and TTC, as described in the
previous subsection. Moreover, each single logic (ISA, ACC,
CWS and AEB) potentially interacts with the driver. Logics
for ISA and AEB do not involve the driver in an active role
and interaction is very low or zero. Indeed, the posted speed of
the ISA solution is applied by the vehicle automatically, no
intervention is required by the driver and, assuming that the
posted speeds are smooth and consistent, there are no
particular issues to be addressed about the acceptability of
such automation. In the case of ISA the interaction of the
solution is with the vehicle (hence with the model developed in
section II.B) rather than with the driver. In the case of
intervention of the AEB the impact on the driver (and on the
vehicle) is extreme but the interaction with the driver is very
low; given the emergency conditions, the driver is completely
excluded from the control loop of the vehicle, and regaining
control over the vehicle (if at all, if the accident has been
completely avoided) occurs after the vehicle has come a
complete (or almost complete) halt. Interaction of the AEB
with the vehicle (hence with the model of the vehicle
dynamics) is intense and crucial. Another different aspect is
the effect of AEB accuracy on the propensity of drivers to use
the system without switching it off. It is evident that unreliable
AEBs (e.g. which have a high rate of false alarms) induce
drivers to deactivate the solution. However, such a long-term
interaction is not of a run-time type.
Events that strongly interact with the driver are the
intervention of the CWS and the running of the ACC. As
regards the CWS, the driver model described in section II.A
allows for simulating the driver’s interaction with the system,
thus enabling the logic to be tested in a virtual environment.
With regard to the ACC, the logic to be applied by this system
has to be specified. It is worth noting that an appropriate logic
is required, which avoids any undesired effect in terms of
safety. It has been shown elsewhere that in the case of highly
automated solutions the reduced ability of the driver to regain
control of the vehicle in the event of failure of the automatic
system entails a high risk [36]. Indeed, even with moderate
automation, as in the case of ACCs, drivers could experience
problems regaining control, perhaps because of overreliance
on vehicle systems and/or reduced situational awareness (SA)
[37]. In order to avoid reduced SA, the ACC logic here
employed was designed according to the human-like approach
described in [38]; other approaches to human-likeness for
driving skills can be found in [39]. Human-likeness is ensured
by an on-demand calibration process of the parameters of the
linear model, assumed to take place while the driver still has
full control of the vehicle. Indeed, during the control-learning
phase the driver actually drives the car and the ACC observes
his/her behaviour. A run-time calibration procedure is able,
during the learning phase, to translate the observed behaviour
in terms of parameters of a linear stimulus-response model.
Once the model is calibrated, the sampler ends the learning
phase and switches to the running phase; it takes control of the
vehicle and activates the identified behaviour. The learning-
running approach has been shown to be feasible in [38] and
[40], where the linear stimulus-response model has been
validated with respect to both synthetic (laboratory-generated)
and real-world observed car-following trajectories.
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
305
IV. USING THE PLATFORM
The simulation platform was validated by:
- comparing driving trajectories simulated by the
platform with real-world observed trajectories; this
validates the virtual testing framework with respect
to its ability to reproduce real data;
- analysing the driving trajectories in the case of
adoption of the integrated ADAS logic and
assessing consistency with expectations of the
simulated results.
As the simulation platform integrates the driver model
(section II.A), the model of vehicle dynamics (section II.B),
and the behavioural ACC model (section III.B), validation was
carried out with reference to the integrated performance of
these models, without trying to separate the impacts or validate
the modules separately.
Figure 4 below shows the accordance of a simulated vehicle
trajectory with an observed one; as the vehicle is mainly in
car-following conditions, the ACC logic and the model of the
vehicle dynamics are most important in terms of modelling
accuracy. Accuracy in terms of spacing is much more critical
than that in terms of speed, as an effect of integral errors.
Fig. 4 – Modelling accuracy
Figure 5 below shows the result in terms of activation of the
CWS and AEB logics. Some interventions of the CWS can be
highlighted; these occur when the spacing with respect to the
vehicle ahead is excessively reduced. Analysis of the
decelerations suggests that the second of the first two (very
close) warnings results in a reaction on the part of the drivers.
Differently, after more than 300 seconds of simulation a
warning raised by the CWS remains unheard, and as a
consequence the AEB is invoked, with a sudden intense
deceleration. Data replicates the initial condition and the
boundary conditions of an observed car-following trajectory.
In the observed data the ADAS are not in place. However, the
ADAS interventions cause only local divergence of the
simulated trajectory with respect to the observed one. Once
again, the fitting is very satisfactory. Importantly, the
trajectory of vehicle speed (as well as that of acceleration) is
smoother (apart from AEB intervention) in the simulation than
for the observed data; this is an encouraging property for
integrated control logic.
Figure 5 – Activation of CWS and AEB
Figure 6 below shows another case in which a very
aggressive driver is observed. For this driver (as for all others
participating in our experiments) the parameters of the ACC
logic are calibrated thanks to a short learning phase. The
aggressiveness of the driver is well captured by the model and
the simulation replicates it, with the need of frequent
(sometimes neglected) interventions by the CWS. Analysis of
the acceleration plot shows several small decelerations (CWS
activation and – in many cases – driver’s intervention) and one
intense deceleration. In this case the smooth speed and
acceleration profiles activated by the ISA and the ACC are
biased by the frequent decelerations imposed by the safety
logics (CWS and AEB).
Figure 6 – Activation of the ADAS logics for an aggressive driver
V. CONCLUSION
A platform was described and tested, allowing integration of
different simulation models developed by specialists in a
cooperative multi-actor environment. Two modules of the
platform received particular attention in terms of development:
the driver’s model and the model of vehicle dynamics.
Simulations were carried out, corresponding to real-world
observations mainly collected during the DRIVE IN2 project (
[41] [42]). The system works as expected, conditions of
potential and imminent danger are eliminated, but the
implemented driving behaviour is very similar to that which
real drivers would have applied. An increase can be observed
in road safety with an (integrated) control strategy that is likely
to maintain the driver within the control loop of the vehicle. It
is confirmed that the ACC behaves human-likely. If used
alone, it induces few situations of potential danger, but a
significant number of imminent dangerous situations (in a few
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
306
cases up to the collision, with almost null TTC values). This is
consistent with the proposed ACC, which does not embed a
safety logic that is left to the CWS and the AEB. The
behaviour learned by the ACC is always applied. In some
conditions this means that (transient) drivers’ aggressive
behaviours are moderated by the ACC (few potential dangers),
while in other conditions (typically, rough braking of the
vehicle ahead) the ACC underperforms with respect to the
human driver who would have promptly modified his/her
behaviour in order to avoid potential or imminent danger
conditions. Of course, these potential and imminent danger
conditions, not dealt with by the ACC, are recovered once the
CWS and AEB are considered.
In the future, further analyses should be devoted to
acceptability among drivers of the integrated system. This can
be properly done by supporting the integration of testing
procedures in a driver-in-the-loop (DIL) approach, based on
the driving simulator. This could also allow the system to be
tested with the real perception and reaction times of the drivers
with respect to CWS warnings, as well as with observed
imposed decelerations.
REFERENCES
[1] NHTSA - National Technical Information Service, «National Motor
Vehicle Crash Causation Survey, DOT HS 811 059,» Springfield,
Virginia 22161, 2008.
[2] iMobility Forum, “Automation in Road Transport: roadmap,” May
2013. [Online]. Available:
http://www.imobilitysupport.eu/library/imobility-forum/working-
groups/active/automation/reports-3/2185-auto-wg-automation-roadmap-
final-report-june-2013/file.
[3] EPoSS, «European Roadmap: Smart systems for automated driving,»
2015. [Online]. Available: http://www.smart-systems-
integration.org/public/documents/publications/EPoSS%20Roadmap_S
mart%20Systems%20for%20Automated%20Driving_V2_April%20201
5.pdf.
[4] F. Yan, M. Dridi and A. El Moudni, «An autonomous vehicle
sequencing problem at intersections: A genetic algorithm approach,»
International Journal of Applied Mathematics and Computer Science,
vol. 23, no. 1, pp. 183-200, 2013.
[5] J. Wu, A. Abbas-Turki and F. Perronnet, «Cooperative driving at
isolated intersections based on the optimal minimization of the
maximum exit time,» International Journal of Applied Mathematics and
Computer Science, vol. 23, no. 4, pp. 773-785, 2013.
[6] S. de Luca and R. Di Pace, “Modelling users' behaviour in inter-urban
carsharing program: A stated preference approach,” Transportation
Researrch Part A: Policy and Practice, vol. 71, pp. 59-76, 2015.
[7] D. Talaba and H. Erdelyi, “Virtual prototyping of mechanisma using
customized haptic feedback,” in 8th International Symposium on Tools
and Methods of Competitive Engineering, 2010.
[8] N. A. Stanton and M. S. Young, «Driver behaviour with adaptive cruise
control,» Ergonomics , vol. 48, no. 10, pp. 1294-1313, 2005.
[9] B. D. Seppelt and J. D. Lee, «Making adaptive cruise control ACC
limits visible,» International Journal of Man-Machine Studies, vol. 65,
no. 3, pp. 192-205, 2007.
[10] R. Bishop, Intelligent Vehicle Technology and Trends, Boston: Artech
House, 2005.
[11] J. G. Hollands, Engineering Psychology and Human Performance (3rd
ed.), Atlantic City, NJ.: Prentice Hall, 2000.
[12] «Automated Driving Roadmap, version 3.0,» 16 February 2015.
[Online]. Available:
http://www.ertrac.org/uploads/documentsearch/id35/ERTRAC_Automat
ed-Driving_draft3-web.pdf.
[13] M. Gallo, L. D'Acierno and B. Montella, «Global Optimisation of
Signal Settings: Meta-heuristic algorithms for solving real-scale
problems,» Advances in Intelligent Systems and Computing, vol. 262,
pp. 177-193, 2014.
[14] G. E. Cantarella, S. de Luca and A. Cartenì, «Stochastic equilibrium
assignment with variable demand: theoretical and implementation
issues,» European Journal of Operational Research, vol. 241, no. 2, pp.
330-347, 2015.
[15] E. Cascetta, A. Cartenì, F. Pagliara and M. Montanino, «A new look at
planning and designing transportation systems as decision-making
processes,» Transport Policy, vol. 38, pp. 27-39, 2015.
[16] Oktal, «Scaner Driving Simulation Engine,» Oktal, [Online]. Available:
http://www.scanersimulation.com/. [Consulted on 11 05 2015].
[17] F. Galante, F. Mauriello, A. Montella, M. Pernetti, M. Aria and A.
D'Ambrosio, «Traffic Calming along Rural Highways Crossing Small
Urban Communities: Driving Simulator Experiment,» Accident
Analysis & Prevention, vol. 42, no. 6, pp. 1585-1594, 2010.
[18] A. Montella, M. Aria, A. D'Ambrosio, F. Galante, F. Mauriello and M.
Pernetti, «Simulator Evaluation of Drivers’ Speed, Deceleration and
Lateral Position at Rural Intersections in Relation to Different
Perceptual Cues,» Accident Analysis & Prevention, vol. 43, pp. 2072-
2084, 2011.
[19] P. Bouchner and S. Novotny, “Development of advanced driving
simulator: Steering wheel and brake pedal feedback,” in 2nd
International Conference on Circuits, Systems, Control, Signals, 2011.
[20] P. Bouchner and S. Novotny, «Car dynamics model - Design for
interactive driving simulation use,» in 2nd International Conference on
Applied Informatics and Computing Theory, 2011.
[21] G. Schaefer, D. Lechner, Y. Delanne and V. Schmitt, «CALLAS: A
decisive step toward validity for 3D vehicle dynamics,» in Proceedings
of the 30th International Symposium on Automotive Technology and
Automation, Florence (IT), 1997.
[22] J. Stéphant, A. Charara and D. Meizel, «Vehicle sideslip angle
observers,» Control Engineering Practice, vol. 15, no. 7, p. 803–812,
2007.
[23] E. P. Todosoiev, «The Action Point Model of the Driver Vehicle
System, Report No. 202A-3,» The Ohio State University, Engineering
Experiment Station, Columbus, Ohio, 1963.
[24] R. Wiedemann, «Simulation des Strassenverkehrsflusses,» Tech. Rep.
Institut für Verkehrswesen. Universität Karlsruhe Heft 8 der
Schriftenreihe des IfV, Karlsruhe, 1974.
[25] G. N. Bifulco, L. Pariota, M. Brackstone and M. Mcdonald, «Driving
behaviour models enabling the simulation of Advanced Driving
Assistance Systems: Revisiting the Action Point paradigm,»
Transportation Research Part C: Emerging Technologies, vol. 36, pp.
352-366, 2013.
[26] F. Farroni, M. Russo, R. Russo, M. Terzo and F. Timpone, «A
combined use of phase plane and handling diagram method to study the
influence of tyre and vehicle characteristics on stability,» Vehicle
System Dynamics, vol. 51, no. 8, p. 1265–1285, 2013.
[27] F. Farroni, R. Russo, M. Russo, M. Terzo and F. Timpone, «On the
influence of anti-roll stiffness on vehicle stability and proposal of an
innovative semi-active magnetorheological fluid anti-roll bar,» in Raad
2012 Proceedings. 21th International Workshop on Robotics in Alpe-
Adria-Danube Region, Naples, 2012.
[28] F. Farroni, M. Russo, R. Russo and F. Timpone, «A physical–analytical
model for a real-time local grip estimation of tyre rubber in sliding
contact with road asperities,» Proceedings of the Institution of
Mechanical Engineers, Part D: Journal of Automobile Engineering, vol.
228, no. 8, pp. 955-969, 2014.
[29] F. Farroni, D. Giordano, M. Russo and F. Timpone, «TRT: Thermo
racing tyre a physical model to predict the tyre temperature
distribution,» Meccanica, vol. 49, no. 3, pp. 707-723, 2014.
[30] F. Farroni, A. Sakhnevych and F. Timpone, «An Evolved Version of
Thermo Racing Tyre for Real Time Applications,» in World Congress
on Engineering, London, 2015.
[31] NHTSA, «Preliminary Statement of Policy Concerning Automated
Vehicles,» 30 May 2013. [Online]. Available:
http://www.nhtsa.gov/About+NHTSA/Press+Releases/U.S.+Department
+of+Transportation+Releases+Policy+on+Automated+Vehicle+Develop
ment.
[32] M. Van Ratingen, A. Williams, P. Castaing, A. Lie, B. Frost, V. Sander,
R. Sferco, E. Seger and C. Weimer, «Beyond NCAP: Promoting New
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
307
Advancements in Safety,» in Proceedings of the International Technical
Conference on the Enhanced Safety of Vehicles, 2011.
[33] C. Hyden, Traffic conflicts technique: state of the art, Green Series:
Traffic safety work with video processing, H. H. Topp (Ed.), University
Kaiserslautern, Transportation Department, 1996, vol. 37, pp. 3-14.
[34] A. Tarko, G. Davis, N. Saunier, T. Sayed and S. Washington,
«Surrogate measures of safety,» Subcommitee on surrogate measures of
safety, 2009.
[35] K. Vogel, «A comparative analysis of hotspot identification methods,»
Accident Analysis and Prevention, vol. 35, pp. 427-433, 2003.
[36] N. Strand, J. Nilsson, M. I. Karlsson and L. Nilsson, «Semi-automated
versus highly automated driving in critical situations caused by
automation failures.,» Transportation Research Part F: Traffic
Psychology and Behaviour, vol. 27, no. B, pp. 218-228, 2014.
[37] M. Vollrath, S. Schleicher and C. Gelau, «The influence of cruise
control and adaptive cruise control on driving behaviour – A driving
simulator study,» Accident Analysis and Prevention, vol. 43, no. 3, pp.
1134–1139, 2011.
[38] G. Bifulco, L. Pariota, F. Simonelli and R. Di Pace, «Development and
testing of a fully Adaptive Cruise Control system,» Transportation
Research Part C: Emerging Technologies, vol. 29, pp. 156-170, 2013.
[39] P. Cichosz and L. Pawelczal, “Imitation learning of car driving skills
with decision trees and random forests,” International Journal of
Applied Mathematics and Computer Science, vol. 24, no. 3, pp. 579-
597, 2014.
[40] G. N. Bifulco, F. Galante, L. Pariota and M. Russo Spena,
«Identification of driving behaviors with computer-aided tools,» in
UKSim-AMSS 6th European Modelling Symposium, Malta, 2012.
[41] G. N. Bifulco, L. Pariota, F. Simonelli and R. Di Pace, «Real time
smoothing of car-following data through sensor fusion techniques,»
Procedia - Social and Behavioral Sciences, vol. 20, pp. 524-535, 2011.
[42] G. N. Bifulco, L. Pariota, F. Galante and A. Fiorentino, «Coupling
Instrumented Vehicles and Driving Simulators: opportunities from the
DRIVE IN2 Project,» in 15th IEEE International Conference on
Intelligent Transportation Systems, Anchorage, Alaska (USA), 2012.
Maria Russo Spena (Naples, 11th of February 1988).
She received her M.Sc. degree in Transportation
Engineering in 2012 at University of Naples “Federico
II”. She is at her last year of Ph.D. at same University.
The focus of her research activity is the study of
driving behavior correlated at fuel consumption in
order to develop an eco-drive system. Dr Russo Spena
was also involved, as contractor, in the National
Italian project DRIVEIN2.
Francecso Timpone (Naples, 5th of September 1974)
received the M.Sc. degree in Mechanical Engineering
in 1999 and the Ph.D. degree in Thermomechanical
System Engineering in 2004 both from the University
of Naples “Federico II”. He is Assistant Professor at
the University of Naples “Federico II” and his research
interests include the dynamics and the control of
mechanical systems, involving motorsport and
automotive companies in cooperation activities.
Flavio Farroni (Naples, 7th of March 1985) received
his M.Sc. degree in Mechanical Engineering in 2010
and Ph.D. in Mechanical System Engineering in 2014
at University of Naples “Federico II”. He is technical
consultant for vehicle dynamics at Ferrari S.p.A. and
tyre modeler at GES racing department. Dr. Farroni's
recent work has focused on the development of
interaction models accounting friction and
thermodynamics phenomena and on experimental
activities in the field of contact mechanics for the
optimization of dry and wet grip performances. In February 2015 he has been
awarded as Young Scientist of the Year at Tire Technology International
Conference 2015 in Cologne.
INTERNATIONAL JOURNAL OF MECHANICS
Volume 9, 2015
ISSN: 1998-4448
308