ArticlePDF Available

Abstract

In its 2014 position paper Towards safer roads in Europe. Nine key challenges for road safety research for the next decade, FERSI stated that, despite considerable earlier im-provements in road safety across Europe, the number of road fatalities is no longer de-creasing to the extent necessary to reach national and European targets. In some coun-tries, the number of seriously injured is even increasing, in particular among vulnerable road users (VRUs) such as pedestrians and cyclists. Two rapidly developing areas which are likely to have a large impact on European road safety over the coming decades are the fields of automated driving and intelligent transport systems. The recent rate of technological developments in these areas, coupled with the large degree of uncertainty regarding several key aspects of this development and its effects on traffic safety, has therefore motivated FERSI to draft a new position paper, building on previous statements, on the key questions related to traffic safety, automated driving (AD), connectivity and intelligent transport systems (ITS). This position paper has been prepared during the fall of 2017 by a designated Working Group consisting of leading experts from the FERSI member organizations. The paper first gives a background to the problem and then lists a number of questions which are of primary concern to FERSI, and which have implications for traffic safety. The paper ends with a summary of our recommendations. For sake of simplicity, and to efficiently convey the position of FERSI on this matter in a condensed format, our main concerns are also summarized in bullet-form in the section FERSI Code of principles for ”Safety through automation” at the beginning (page 3) of this document.
Safety through automation?
Ensuring that automated and connected driving
contribute to a safer transportation system
FERSI Position Paper – January 19, 2018
Automated
and
connected
driving and
transport
What is needed
to improve road
safety?
Which road
safety issues
are not
addressed?
What new road
safety problems
may be caused?
How to change
testing,
certification
and validation?
2
Table of contents
Table of contents 2
FERSI Code of principles for ”Safety through automation” 3
The potential of automated and connected driving and intelligent transport systems 4
Purpose 4
Background 4
Potential development and unknown factors 5
Main concerns from the point of view of FERSI 7
How can AD and ITS improve road safety? 7
Which road safety issues will likely not be solved by AD and ITS? 9
What road safety issues may be caused by AD and ITS? 10
How should testing, certification and validation methods be adapted, best-performing AD systems
and ITS best practices be established? 13
Related and important questions 15
Summary of recommendations 16
References 17
About FERSI 19
FERSI mission and objectives 19
Main contributors to this position paper 20
Members of the FERSI Working Group on ITS and Automation 20
3
FERSI Code of principles for ”Safety through automation”
1. Human Factors at the core
“Human-Centred Design” shall be put the core of development to prevent new risks of new levels of auto-
mation and to render unbiased acceptance by users.
2. All potential user profiles
Drivers of all types, backgrounds and ages should be catered for by systems designed in such a way that
all drivers experience automation as safe and comfortable.
3. Safety in mixed traffic
Given the expected increases of traffic situations where users of non-automated as well as of automated
vehicles of various levels will share road space, it is necessary to adapt infrastructure, vehicles and driving
education, to reduce safety risks caused by mixed traffic.
4. Safe communication between automated vehicles and providers of services
Effective, secured, and error-free communication channels between automated vehicles and providers of
services (vehicle manufacturers, mobility services, fleet operators, road administrations, infrastructure
operators, etc.) must be established and maintained to ensure safe interaction between all types of road
users and the connected road environment.
5. Safe communication between all road users
Safe and secured communication among drivers of partially as well as fully automated vehicles, as well as
with other road users – especially vulnerable and unprotected groups – needs to be in place.
6. Safety and automation benefits for vulnerable road users
It must be ensured that vulnerable road users (VRUs) – pedestrians, cyclists, and riders of powered two-
wheelers (PTWs) – also benefit from automation and that, where feasible, specific connectivity- or detec-
tion-based solutions are developed to increase the safety of these groups.
7. New training & testing
All drivers of automated vehicles are to be well trained, tested, and licensed, in order to cope with driving
in modes with different levels of automation, and, when required, be capable of resuming manual control.
Automated vehicles should be able to establish whether the driver is fit to drive and to stay in control.
8. New tests & tools
Test procedures and tools (both virtual and physical) are to be set in place that cover the comprehensive
set of scenarios needed to evaluate, validate, and certify automated systems – and updates and technical
inspections are to be integral part of that process.
9. Policy mechanisms for incorporating safety considerations
Policy mechanisms are to be set in place to ensure that automated and connected driving development
takes safety and human factor considerations into account – and brings about substantial safety benefits
in a timely manner for all road users.
10. Impact assessment
Evaluation methods and models are to be established to measure the impact of automation – and infor-
mation from crash investigation as well as data from automated vehicles and their use both in tests and
operation, should be made openly available to impartial research for the sake of further improving safety.
N.B. that even if these principles may seem self-evident and straightforward, the underlying questions
that need to be solved are complex and hard, and will require considerable effort, and, even so, for
many of them there will still not be satisfactory solutions for many years.
4
The potential of automated and connected driving and intelligent
transport systems
Purpose
In its 2014 position paper Towards safer roads in Europe. Nine key challenges for road
safety research for the next decade, FERSI stated that, despite considerable earlier im-
provements in road safety across Europe, the number of road fatalities is no longer de-
creasing to the extent necessary to reach national and European targets. In some coun-
tries, the number of seriously injured is even increasing, in particular among vulnerable
road users (VRUs) such as pedestrians and cyclists.
Two rapidly developing areas which are likely to have a large impact on European road
safety over the coming decades are the fields of automated driving and intelligent
transport systems. The recent rate of technological developments in these areas, coupled
with the large degree of uncertainty regarding several key aspects of this development
and its effects on traffic safety, has therefore motivated FERSI to draft a new position pa-
per, building on previous statements, on the key questions related to traffic safety, auto-
mated driving (AD), connectivity and intelligent transport systems (ITS). This position
paper has been prepared during the fall of 2017 by a designated Working Group consist-
ing of leading experts from the FERSI member organizations.
The paper first gives a background to the problem and then lists a number of questions
which are of primary concern to FERSI, and which have implications for traffic safety.
The paper ends with a summary of our recommendations. For sake of simplicity, and to
efficiently convey the position of FERSI on this matter in a condensed format, our main
concerns are also summarized in bullet-form in the section FERSI Code of principles for
”Safety through automation” at the beginning (page 3) of this document.
Background
Vehicle manufacturers and suppliers – but also other actors from the information technol-
ogy (IT) industry and fleet operators – are currently racing to introduce and implement
advances in AD, connectivity-based solutions and ITS technologies. Industrial actors in-
tend to gain shares of what is expected to become a large and completely transformed
transport market dominated by partially or completely automated and connected vehi-
cles, covering the whole range from light person or goods transport, via automated
public transport “pods” or buses, to heavy goods vehicles. In this line of reasoning,
AD and ITS are expected to lead to optimization of road use, to improvements in traffic
flows, and to reduction of congestion due to shorter inter-vehicular spacing as well as re-
duced pollution and fuel consumption (Fagnant & Kockelman, 2015; Willemsen, Stuiver,
& Hogema, 2015). This should be the result of either centralized transport route optimi-
zation, or more de-centralized decision making in distributed platforms. In both scenarios
transport planning will operate by mechanisms that better match transport supply and
demand, and e.g. by supplying mobility as a service (MaaS) to complement the current
public transport offering (first and last miles, transport ser-vices in rural areas).
Some also envision that a transport system with fossil-free, self-driving small “pods” cov-
ering the “first” and “last” mile thus complementing other forms of more rapid pub-
lic transport – will make private cars for commuting far less attractive. On a similar note,
in increasingly congested cities, shifting store deliveries to automatized night-time op-
eration could potentially solve issues of both congestion and traffic safety during the day.
There has been remarkable technical progress over the past few years regarding the
sensors and artificial intelligence (AI)-related algorithms required to enable increasing
levels of automation, beyond today’s driver support systems. Mixed-traffic medium- or
large-scale trials with AD vehicles are being planned or already underway particularly in
Europe, Asia, and North America.
In the public debate, it is often claimed that automated driving has the potential to im-
prove road safety. This reasoning is partly based on estimates indicating that,
5
presently, human errors cause or contribute to more than 90% of crashes (Singh, 2015;
Thorpe, Jochem, & Pomerleau, 1997). Consequently, if the driving task could be shifted
to optimal, error-free systems for automated driving, thus removing elements of human
error from the operation of vehicles, risks for collisions should be reduced, and, ulti-
mately, eliminated. Perhaps, however, a more realistic view is that we may be replacing
at least partially crashes associated with human error by crashes caused by imperfect au-
tomated systems in the intermediary stages, when “mixed traffic” likely will prevail.
Another potential contribution to road safety, enabled by large-scale cloud-connectivity
of vehicles, will be the possibility to instantaneously propagate more precise, detailed
individualized and thereby relevant information, which can be used by the driver or the
vehicle itself to take actions to improve traffic safety (e.g. by lowering speed, avoiding
obstacles or accidents, allowing free passage for emergency vehicles approaching, etc.
One example would be providing crowd-sourced and critical incident, hazard, or road
condition information to other road users (Eriksson, Lindström, Seward, Seward, &
Kircher, 2014) via low-latency ITS solutions (NordicWay consortium, 2017). While the
potential for positive traffic safety effects stemming from digitalization and vehicle con-
nectivity could be large, one cannot expect such services to emerge spontaneously, and
even if some do (e.g. driven by commercial interests), others which are perhaps more
traffic-safety or society-oriented will perhaps not be included. Work with both connectiv-
ity-based and infrastructure enablers and these types of services must therefore be ac-
tively pursued, while at the same time of course taking all actions necessary to eliminate
risks of misuse, fraud, or information/data quality problems.
If combined with fossil-free propulsion technology and the electrification of road infra-
structure, it is also envisaged in the public debate that a combined development of
shared mobility by virtue of self-driving and connected vehicles is a way of cutting the
proverbial Gordian Knot of future mobility solutions in a single blow.
Potential development and unknown factors
There are large uncertainties remaining regarding several of the fundamental factors
conditional for the development described above. An important factor that needs consid-
eration is if and how pre-existing vehicles and road infrastructure could (and should) mix
and co-exist with ITS solutions and vehicles capable of varying levels of automated driv-
ing. Even if technology development is rapid, infrastructure changes and vehicle fleet re-
newal will still take decades. Therefore, this type of mixed traffic will likely dominate
over the next several decades. Many questions related to traffic safety need to be ad-
dressed in practice during that time, for instance by technology development, by intro-
ducing of new types of rules and regulations, and through education. Whilst an immedi-
ate uptake, and substantial reduction in accident rates may not be feasible at low adop-
tion rates of automated vehicles, there could still be significant benefits to be reaped
from uptake rates as low as 2–5%, as well as from partial automation of particular driv-
ing tasks. Some claim that such benefits include improved traffic flow and a reduction in
phantom traffic jams (Fountoulakis, Bekiaris-Liberis, Roncoli, Papamichail, &
Papageorgiou, 2017; Ioannou, 1998; Roncoli, Papamichail, & Papageorgiou, 2015). Other
researchers, however, have shown that in certain scenarios, energy use and greenhouse
gas emissions will be nearly doubled with high levels of AD penetration in the vehicle
fleet (Wadud, MacKenzie, & Leiby, 2016).
Part of the reason for these apparently contradictory predictions, is that a key determin-
ing factor is the extent and individual willingness of sharing transport (Pernestål
Brenden, Kristoffersson, & Mattsson, 2017). If there is an advantage and a willingness to
share transport, automated driving will result in improved traffic efficiency, fewer vehi-
cles and less congestion – sometimes known as automation heaven, a scenario in which
most people transport becomes “public transport” and where “mobility as a service” pre-
vails and provides optimal use of transport resources. However, if shared mobility does
not increase, the technological development may lead to the exact opposite outcome,
known as automation hell: since vehicles in a future scenario drive themselves, possibly
at low running costs due to electrification, this will allow vehicle “occupants” (since they
6
are technically not drivers) to do other tasks while queueing in individual, self-driving
cars. Some of these cars could even be completely empty – causing more traffic – with
potentially more incidents, accidents and increased congestion.
Yet another fundamental issue to be solved concerns the interaction between automated
vehicles on the one side, and regular vehicles and vulnerable road users, on the
other. Fundamental questions are for instance how to indicate to other road users
whether a particular vehicle is self-driving or not, or how to communicate to a pedestrian
or cyclist that they have indeed been “seen” by the self-driving vehicle.
Other major questions in need of solving are related to physical and digital road infra-
structure, road stretch designation, legal and policy regulations, allocation of insurance
costs and liability, driving/riding licensing and training as well as traffic safety questions
regarding the human interfaces to AD and ITS systems.
And finally, the view that traffic safety could be improved by removing the human
driver from “the loop”, is overly simplified. As pointed out already by Bainbridge
(1983), automation of industrial processes may sometimes aggravate rather than reduce
problems with the human operator. This is particularly true if you resort to the “standard”
approach of leaving the operator with responsibility for abnormal conditions, which is
precisely what so-called level 3 systems, according to the SAE scale1 for AD, attempt to
do. Assistance systems at the lower end of the SAE scale (levels 0 – 3) can all be said to
require active supervision, since they rely on a human driver remaining ultimately re-
sponsible for the driving task at all times, whereas the highest levels (4 – 5) should not
require manual intervention, and can be termed un-supervised. However, due to the
way level 5 is specified, it will likely be unattainable in the next decades, and only of the-
oretical interest, since it requires dealing with any situation and environmental condition
manageable by man2. Consequently, the highest level of automation attainable – at least
from an immediate and practical standpoint – will be the SAE Level 4. It must also be
noted that a rapid transition from Level 2 to Level 4 automation for the entire traffic sys-
tem is impossible, as state-of-the-art Level 2 vehicles are to date typically limited to mo-
torways or other particular road environments where traffic is less complex. Indeed, a
similar approach of successive introduction will likely be necessary for Level 4 systems.
Initial systems for level 4 AD are likely to only function in specific environments, for in-
stance, on motorways only, whilst collecting data in areas outside of their operational de-
sign domain (ODD) until reliable behaviors can be ensured in an extension of the ODD.
If AD technology was already perfected, the potential for vehicles under the truly unsu-
pervised levels of automation, Level 4 with an extensive ODD or full (level 5) automation,
would theoretically be significant. The general argument is that, unlike humans, AD sys-
tems do not become tired or distracted, or lose attention, nor do they drive under the
influence of alcohol or drugs. They would be programmed to always keep within speed
limits, keep a safe distance to the vehicle in front and never overtake unless margins
are well on the safe side. Sensor placement can provide continuous data flow with 360-
degree views, and sensor technology furthermore allows AD systems to use frequency
ranges (and combinations thereof), well outside human perceptive abilities, such as RA-
DAR, LIDAR and IR.
In addition, recent advances in contemporary AI algorithms have already led to a situa-
tion where detection and control systems are trained on globally acquired visual and
other sensory training data which represents many orders of magnitude more data
than any individual human driver could ever encounter during a lifetime. On-line prolifer-
ation of updated decision schemes, based on newly encountered and resolved situations,
so-called fleet learning, is another way forward. Advanced algorithms are capable of way
shorter reaction times than human drivers and can easily base decisions on multiple data
sources (both on-line and historic, as well as correlations). Furthermore, communication
1 See https://www.sae.org/news/3544/
2 This does not mean that every human handles every situation well – only that some humans indeed cope where
others fail – nor that the AD system must always cope.
7
and connected ITS systems could enable several road safety-enhancing services, such
as sharing of road friction or road temperature information, coordinated emergency brak-
ing, early and directed warnings for queues, incidents and hazards up ahead which can
consequently be avoided (Eriksson, Lindström, Seward, Seward, & Kircher, 2014), etc.
In general, however, the actual performance levels of AD and ITS systems beg further
scientific scrutiny, and research efforts should be directed towards ensuring the lev-
els of safety of automated systems, as they develop in terms of their ODD. One pro-
posed method is to specify minimum performance levels in selected scenarios. This is by
no means easy: to improve on human drivers, an AD system must typically have a fatal
crash rate better than one fatality in billions of kilometers, making testing and verifica-
tion of performance a difficult or even impossible task (Kaira & Paddock, 2016). This may
indeed become a bottle-neck for the development of AD, since we currently lack the
methods to specify and assess safety levels of AD systems. Until such methods, for in-
stance based on formal mathematical proofs (Shalev-Shwartz, Shammah, & Shashua,
2017), are developed, we are unlikely to see the improvements expected by society. An-
other specific aspect related to the data-driven schemes currently dominating the algo-
rithmic side of AD systems is that of bias in the data, and the risk that excessive
amounts of training data on one type of road or situation, will still not be helpful or carry
over to other road types (or other situations).
What is stated above and elsewhere in this document may seem to stand in contrast to
more technology-positive messages in the media from large actors who claim they will
have completely automated vehicles in regular operation on a large scale in major cities
of the world shortly. Our fear is that this view is tainted by the economic driving forces
of these actors and that several of the concerns in this paper are too easily over-
looked by those actors as well as others who share and propagate similar views.
Main concerns from the point of view of FERSI
This FERSI position paper focuses on four key questions:
I. How can automated and connected driving and ITS improve road safety? What
conditions should be met, and which actions taken?
II. Which road safety issues will likely not be solved by AD, connectivity and ITS? In
particular, are there groups of road users which could benefit from AD and
ITS, but are unlikely to do so unless special action is taken?
III. What road safety issues may be caused by AD, connectivity and ITS? What ac-
tions can be taken to avoid this?
IV. How should testing, certification and validation methods be adapted and how
should “best performing” AD/connected systems and ITS best practices be
identified?
Each question is briefly discussed and then followed by a proposal for research topics
that need to be addressed over the coming years. For FERSI it is crucial that these
questions are tackled for AD, connectivity and ITS to successfully contribute to
a smart, green, and integrated transport system which is also safe.
How can AD and ITS improve road safety?
Considerable safety gains can already be achieved with contemporary technologies and
systems at the lower end of the SAEJ3016 scale1. This scale is used to define the level of
automation of vehicles. It ranges from 0 (“no automation”) through 5 (“full automation”).
For instance, systems for Automated Emergency Braking (AEB, a Level 1 “driver assis-
tance” system), could save lives and have been shown to reduce injuries, by intervening
and reducing the speed of or even stopping a vehicle completely when collision with a
solid object is imminent and/or unavoidable. For instance, at speeds below 50 km/h, AEB
was shown to halve the risk of being the striking party in injury-inflicting rear-end
crashes (Rizzi, Kullgren, & Tingvall, 2014).
8
Level 2 “partial automation” systems, for example adaptive cruise control (ACC), when
used in combination with lane keeping assistance (LKA) and optionally with lane depar-
ture alert (LDA) are already in operation in many vehicles. They are expected to help
avoiding rear-end collisions and lane-change crashes caused by human inattention
and/or human inability to visually observe and constantly monitor vehicle surroundings.
However, no clear such effect has been observed in the Field Operative Trials (FOTs) car-
ried out so far, sofurther studies are needed to verify that this is indeed the case. An-
other level 2 technology which is more promising is Intelligent Speed Adaptation (ISA),
which is known to reduce accidents and their effects, also on VRUs (Carsten & Tate,
2005).
Level 3 is a rather special case in the “supervised” category, called conditional automa-
tion, where the vehicle operates autonomously under many conditions, but where control
could be returned to a human driver within a moment’s notice, e.g. when automation
fails. Level 3 automation vehicles are highly automated and drive themselves under spe-
cific, but not all, conditions. They allow the driver to take both feet, hands, and eyes off
the driving task, but may still require manual intervention, at short notice, in situations
which the vehicle cannot handle. Traffic safety researchers have shown that the faith put
on human capabilities, specifically on attention and readiness are overestimated, thus in-
troducing novel and unacceptable accident risks in traffic. Prior knowledge regarding hu-
man behavior, and in particular the human inability to stay alert and respond adequately
at short notice (as would be the case) has given rise to several recent studies on simu-
lated, as well as on-road, interaction with AD systems, showing that reaction times are
comparatively long, typically in the order of 4-6 seconds (Eriksson & Stanton, 2017), and
the ability of drivers to react appropriately in the driving situation is severely diminished
during long stretches of time, and increase even further when the driver is engaged in a
secondary task. FERSI has therefore concluded, as have several major car manufacturers
(Level Three Automation Could Be Skipped, 2017; Davies, 2015; Why Every Car Maker
Should Skip Level 3, 2017; Volvo to skip Level 3 autonomous mode, 2017), that level 3
automation is unrealistic, and should for the practical and traffic safety purposes in this
position paper be disregarded.
Indeed, Level 2 systems are already presenting problems, since drivers overestimate
function, range and reliability. Some research indicates that most drivers do not really
understand how systems work and whether they are engaged or not (Banks, Eriksson,
O'Donoghue, & Stanton, 2018).
Another potential benefit of AD and ITS is that these technologies may provide safe driv-
ing solutions to groups which are either unauthorized (due to lack of a driver’s license),
unable (e.g. due to physical inability, old age or other types of impairment) or unfit to
drive (e.g. due to alcohol consumption, sleep deprivation, or under the influence of pre-
scription drugs). Higher levels of AD could potentially (albeit perhaps only theoretically)
provide mobility solutions for some (or all) of these groups.
However, the above reasoning effectively compares the present safety performance of
humans who are driving in an imperfect state against the possible safety level of a per-
fect Level 4 system. Unfortunately, one can safely assume that – at least for decades to
come – Level 4 systems will likely be restricted to very specific ODD’s where complexity
is low and there is a negligible risk of failures, whilst potentially continuing to operate at
a Level 3 stage, which is neither perfect, nor capable of ideally operating in all scenarios.
Consequentially, it is far from obvious that the promising AD-related safety prospects will
ever fully materialize, especially since comfort and mobility improvements seem to be the
main driver for the AD and ITS developments, along with the expected associated sheer
economic gains. The current version of the ERTRAC automation road map (ERTRAC
Working Group "Connectivity and Automated Driving", 2017) predicts imminent develop-
ment in parking and highway technologies, whereas road safety priorities across Europe
rather target VRUs in urban areas, and people in vehicles and on powered two-wheelers
(PTWs) in rural regions. It is therefore recommended for policy makers and other
stakeholders in road safety to specifically identify and promote certain areas in need of
9
development. Otherwise we run the risk of these areas gaining insufficient attention, and
of the resulting development affecting traffic safety adversely.
Questions to be solved for positive traffic safety effects to occur include the fol-
lowing:
1. How to ensure that AD and ITS development incorporates relevant and appropriate
safety and human factor considerations? It is necessary that societal actors,
legislators, and traffic authorities have a proper understanding and decision-
making power to steer the development of AD and ITS into the desired direction.
In scenarios where commercial forces in the vehicle and IT industry assume con-
trol over the development of AD, ITS and mobility services, some of the benefits
for society, e.g. improved safety for other road users, risks not being addressed
adequately. We therefore need a mechanism for policy-makers to direct advanced
driver assistance systems (ADAS) and AD system development so that they bring
substantial safety benefits in a timely manner. Operational definitions will need to
be defined with regards to safety levels – should it be as safe as the average
driver, as the top 10th percentile of drivers or defined according to the ALARA (as
low as reasonably achievable) principle?
2. How to adapt and develop in-vehicle technology as well as urban and non-urban in-
frastructure, so that AD and ITS enhancements lead to substantially increased
safety levels? There is no point in promoting AD for the happy, select few, limited
to a number of stretches such as straight motorways, where road safety is already
high and not much is to be gained in this respect. There is need for more research
into the necessary adaptations of the road infrastructure for optimal use of AD
and ITS, e.g. by connected traffic lights, proliferation of warnings, coordinated
emergency braking and many other enhancements made possible by recent tech-
nology developments.
3. How can information from other services be integrated and harmonized with AD
functions? What requirements should be put on data and information sharing
between operators of mobility services, between vehicle manufacturers, fleet own-
ers and operators, owners of digital and physical infrastructure? What could be
the role of IoT (Internet of Things)? Research is needed on how to combine AD
functions with the booming area of IoT, towards enhancing the level of automa-
tion and facilitating the interaction with road users and the road environment, al-
ways in the benefit of road safety.
Which road safety issues will likely not be solved by AD and ITS?
Most AD and ITS development focuses on car and truck drivers and passengers. Quite
some effort is also put on automation of relatively slow-moving public transport “pods”.
Several groups of road users will likely not benefit at all, or only in a limited way,
from the introduction of such vehicle automation and connectivity-based “intelligent” so-
lutions in the transport system. These include large groups at risk among vulnerable road
users, which constitute the bulk of the remaining challenge to reach the European road
safety goals. Already today, VRUs are more numerous than car passengers amongst the
traffic victims that are hospitalized. Unless specific focus is given, cyclists, pedestrians,
motorcycle, and moped riders (i.e. VRUs) are not likely to gain immediate benefits from
AD and ITS technology, or at least not to the same extent as car and truck occupants.
The same holds, for instance, for people driving on rural roads – where it might be too
expensive to adapt the road infrastructure, for elderly – who can no longer adapt to new
“driving modes”, and for young people – who will have to learn to drive in different
modes (which might be more challenging than is currently the case). Due to the larger
cost of AD vehicles (and vehicles equipped with advanced ADAS systems), they will ini-
tially be bought by older rather than younger people, resulting in an age bias in the user
10
base for quite some time before these vehicles eventually become common on the sec-
ond-hand market.
A conservative estimate of the share of accidents which cannot be avoided by virtue of
AD, ITS or similar technology can be made based on in-depth research and crash statis-
tics. Based on such data (ACEM, 2009; ERSO, 2008; Schepers, Stipdonk, Methorst, &
Olivier, 2017) we estimate that in 10% of road fatalities in Europe no motor vehicles that
can be automated are involved as victim or crash opponent. Of these, around 4% are cy-
clists and 6% PTWs. For severe injuries this share is hard to quantify but is likely sub-
stantially higher.
Also, looking to traffic situations, there are many conditions and types of events that
might not be addressed by the industrial vehicle automation actors, at least not initially,
since they are costly and difficult to solve. Among these are several dangerous traffic sit-
uations such as snow, heavy rain, roads with poor-quality or no road markings, difficult
intersections and crossings, narrow and curved roads etc.
Questions to be solved therefore include
4. How to make sure that different types of VRUs will also reap the benefits of AD and
ITS developments geared towards cars and trucks? There is a strong need to in-
tensify traffic safety research regarding VRUs, both in general but also specif-
ically related to the interaction between representatives from these groups with
vehicles featuring differing levels of AD.
5. What specific AD and ITS developments are needed for cyclists, moped drivers, mo-
torcyclists, and pedestrians? How can VRUs benefit from ITS or connected vehi-
cle solutions, or how could the latter be enhanced or modified to yield customer
utility also for specific VRUs?
6. How to ensure that both older and younger drivers will fully benefit from con-
nected and automated driving and ITS, and not experience it as an obstacle for
safe and comfortable driving?
What road safety issues may be caused by AD and ITS?
Under the supervised levels (up to SAE level 3) of AD, there are many issues related to
the interface (human-machine interface or HMI) between the human rider/driver and the
vehicle.
These systems are intended to aid the driver and ease the process of safely driving his or
her vehicle, but could at the same time place additional demand on human perfor-
mance, specifically on attention and readiness, and may therefore introduce new risks
in traffic (Banks, Eriksson, O'Donoghue, & Stanton, 2018). These need to be further
studied, and it may well be the case that specific and additional driver training is needed
for supervised levels (0 – 3) of automation to be acceptable. Even for level 4 systems, if
there are limitations in operation defined by restrictions in the vehicle’s ODD, users need
to be aware of these and understand their consequences.
An AD vehicle will also need to interface with other (physical) road users, both other
vehicles and vulnerable road users such as pedestrians and cyclists. Both users of AD ve-
hicles and other road users may need to acquire a deeper comprehension of how differ-
ent levels of automation work and behave, in order to understand the automation fea-
tures of the vehicles they encounter and to properly interact with such vehicles. Indeed,
even discerning between AD and regular vehicles may be difficult enough. There is a defi-
nite need to investigate the advantages (and disadvantages) of differentiation between
different types of vehicles, either by clear visual features (adapted for human perception)
and/or electronically (for AD vehicles, and the connected road infrastructure). It is obvi-
ous that this will be a complex and long process that will not follow smoothly and hence
will lead to dangerous situations that may, in turn, lead to crashes.
11
In addition, any AD vehicle is likely to interact not only with a single connected service,
but with several which may reside in different vehicle “clouds” related to the specific
vehicle, brand, or fleet, or to independent information suppliers. From all these sources,
information needs to be extracted and possibly harmonized or at least allowed to co-exist
with information from other channels. While this is a prerequisite for developing efficient
mobility solutions, it adds a yet unseen level of complexity to the transport system which
may of course cause conflict and possibly errors. Such harmonization, quality assurance
and standardization of information flows may be essential to achieve many of the ad-
vantages of automated and connected vehicles and ITS.
Drivers must also be aware of the limits of the ADAS and/or automated features of
their cars such that they do not rely too much on these features – and hence risk to lose
control. A concrete example is that current Lane Keeping Assist (LKA) systems will not
work properly when there are no lane markings; when the speed of the vehicle is too
low; when the curve is too narrow; when something is blocking the camera that detects
the lane markings; when snow or leaves cover the lane markings; when there is too
much rain, or fog, etc. for the camera-based computer vision system to detect the mark-
ings. More specifically, drivers must know that systems have both technical limitations
limited range or domain of competence (ODD), which may result in sudden loss of AD ca-
pability.
As mentioned previously, human limits in staying alert and acting as fallback when sys-
tem automation fails may in fact preclude the introduction level 3 AD systems com-
pletely, which would coincide with FERSI’s general view. While several of the traditional
vehicle manufacturers advocate skipping level 3 altogether (Davies, 2015; Level Three
Automation Could Be Skipped, 2017; Volvo to skip Level 3 autonomous mode, 2017;
Why Every Car Maker Should Skip Level 3, 2017), other actors promote level 3 strongly,
and make claims regarding their almost imminent introduction. This could lead to public
and/or political demands requiring drivers to practice “handover skills”, although
these skills are likely not possible to train. Instead the systems need to adapt to human
thresholds for resumption of control, (Eriksson & Stanton, 2017). All this could make
driver training more complicated and expensive. For established and particularly
older drivers, this can be quite some challenge since they have become accustomed to a
certain type of driving, using human-vehicle interfaces predominantly designed for direct
and immediate manipulation in traditional manual driving. Even if contemporary user in-
terface design in AD vehicles should be self-explaining and intuitive to use, chances are
that the mode of operation is still qualitatively simply too different for large groups.
As previously mentioned, there will be a lengthy transition phase while the number of
automated systems on the road is increasing and at the same time the capabilities of
those systems will be rapidly expanding. Separation of AD vehicles from non-AD vehicles
by use of designated road networks (or separation in time) is a luxury few regions in the
world can afford, so during a foreseeable time, we will have a mixed fleet.
While there has been some simulation work to examine the network level impacts of a
mixed fleet, the capabilities of existing microsimulation models to accurately represent
both the functionality of automation systems and the detailed kinematics of road user
manoeuvres are very limited. This is an area needing further attention.
The extensive transition phase will provide opportunities to improve safety but there will
also be new risks introduced. As yet, we have little idea about how we should manage
the process to get the greatest road safety gains as quickly as possible. There is an
urgent need to start developing suitable strategies for that, rather than allowing develop-
ment to be driven solely by other forces in operation today. The latter obviously include
technological advances, but also consumer demand, e.g. for the comfort of being chauf-
feured. On the business side the vehicle industry has a strong urge to increase sales by
introducing new models with new features, while actors originating in the IT and commu-
nication technology sphere aim at developing new business by providing new types of
mobility services which are profitable, perhaps taking market shares from the traditional
vehicle industry.
12
It should also be added that alcohol and drugs consumption may increase if (and
when) unsupervised levels of AD are introduced, if intoxication under automated
transport is not regulated. Aside from obvious negative health effects of such a (tenta-
tive) consumption increase, there are also general road safety risks with an increased
share of individuals under influence of alcohol or drugs in the transport system. In addi-
tion to this, there are also specific risks that a “driver” or passenger under impairment
can misuse AD and ITS systems, or, for example resume vehicle control when intoxicated
(even if it may be unlawful). In critical situations where for instance an automated vehi-
cle ends up outside its ODD, an intoxicated person will be even slower to react, or even
fail to do so, and is of course more prone to respond inappropriately or perform errone-
ous and dangerous or even fatal actions.
For the above reasons, a set of questions which would need special attention
are:
7. How to manage the transition to automated driving while guaranteeing road
safety, and ensuring that mixed traffic conditions do not lead to increased
safety risks? One radical, but likely expensive, measure would be to separate au-
tomated vehicles from other road users - either permanently by designated and
barred-off lanes, blocks or regions - or temporally, by only allowing their use on
designated road stretches during specific times of the day. At the other extreme,
where completely mixed traffic would be allowed, research must instead focus on
somehow verifying the road safety aspects of interaction between AD vehicles and
all other types of road users in a multitude of locations, situations, road, light and
weather conditions.3
8. How to support the driver’s automation mode awareness in view of decreasing
accident risks? Research is needed regarding how difficulties in understanding the
level or state of automation (both for drivers and other road users) relates to acci-
dent risks, and what remedies are at hand, e.g. by virtue of HMI solutions or
training.
9. What changes are needed for driver training, both initial driving training courses
and upgrade courses later in life? Drivers of vehicles featuring supervised AD will
need to learn to drive in “non-automated mode”, and in different “assisted
modes”. What is the role and responsibility of car dealerships in this respect?
10. How to ensure that the resumption of control by drivers of supervised AD vehi-
cles happens smoothly and does not create additional risks? Research is needed
on how to support resumption of control, in critical situations as well as under
more normal conditions (Eriksson & Stanton, 2017; Eriksson & Stanton, 2017). A
specific issue lies in how the system should establish that the driver is actually in
a state where he or she is fit to drive (and not, for instance, drowsy, sleepy, fa-
tigued, agitated, distracted or even intoxicated).
11. How to adapt the driver licensing system and driver license categories to take
into account the variety of situations (with consequences for liability) resulting
from development in ITS, automated and connected driving? More research is
needed in this area and policy-makers need to discuss this at European or prefer-
ably even at global level. How will AD affect professional drivers, and the legis-
lation and other regulations regarding driving and resting times? What new
types of professional driving license categories may be needed? What will be
the license and training requirements on the driver of a rented car or a car-pool
vehicle?
3 This is indeed a challenging task, since recent estimates state that automated vehicles
need to be driven between 1.6 million and 11 billion miles (depending on the safety
thresholds), taking up to 500 years for a fleet of 100 vehicles, to statistically assess the
safety of automated driving technology (Kaira & Paddock, 2016)
13
12. How to ensure that drivers will still be able to handle extremely critical or unfore-
seen situations while driving under supervised conditions? How to avoid that ex-
tensive support systems and partial automation use will lead to less concern for
accidents? Automation-induced inattention, unawareness, or unfitness to act
are areas in need of investigation. So-called selective de-skilling may also be a
potential hazard: relying heavily on ADAS or AD functionality may over time en-
courage drivers to forget key skills, especially in critical or stressful situations,
when they may lose situational awareness or vehicle control.
13. How to make sure that road users with different levels of automation communi-
cate appropriately with each other in view of avoiding traffic conflicts? Research is
needed regarding how to prevent accidents due to inappropriate actions by others
(particularly vulnerable road users) and, in particular, how to handle their com-
munication and interaction with AD vehicles.
14. How to ensure that unmanned vehicles can communicate appropriately with
vulnerable road users? Which specific requirements (in terms of speed, road
paths, safe stopping distances, communication/alert systems) are to be defined to
make sure that they do not endanger the safety of road users? Research is
needed on such requirements and also on how these could (and should) be en-
forced.
How should testing, certification and validation methods be adapted,
best-performing AD systems and ITS best practices be established?
AD and ITS systems will form an integral part of the traffic system in the future and sev-
eral such systems (both old and more recent) are already in fact doing that, e.g. satellite
navigation systems and radio data system (RDS) traffic announcements, with the im-
portant distinction that the latter are channeled through a physical person driving.
AD systems will consist of a physical vehicle, its sensors, the connected information
sources, and the algorithms designed to detect, understand, and drive in relation to all
the information at hand. Therefore, several issues regarding testing, certification
and validation become considerably more complicated and need to be examined anew.
In addition, technical validation and system certification is only one side of the coin. The
other concerns human users of AD systems (Kyriakidis, o.a., 2017). As mentioned ear-
lier, with different levels of automation, the role of the human in the frame of AD vehi-
cles changes from “a driver driving”, a person in charge of “driver monitoring”, to ulti-
mately being only one of the “passengers/driver not driving” in a vehicle of tomorrow
(Banks & Stanton, In press). As has been discussed above, specific (and differing) risks
are related to these three different options, and research is required regarding how hu-
mans will take on, potentially accept, and perform their tasks and liabilities in these dif-
ferent roles.
Since the highest levels of AD do not really exist on the road today, one way to anticipate
their future use by humans is by performing virtual simulation based on human mod-
els, as well as driving simulations with the “human-in-the-loop”. Indeed, even if an AD
system is technically validated, a remaining challenge is to understand how this system
will be used (or misused) by human beings, and what types of interaction will take place
with other vehicles as well as with humans inside or outside the vehicle in question. Con-
sequently, virtual testing methods and/or new types of integrative simulation tools – in-
cluding both the simulated AD systems and human model(s) of the future users – need
to be developed and used before introducing real AD systems on the public road. They
are essential in predicting potential difficulties or risks for end users, and to prevent new
types of accident likely to occur, much as was the case when automated systems were
introduced in aviation. One additional application of such simulation tools is to help in-
crease future acceptability of AD at the individual and societal levels. Virtual testing
should be understood in a wide meaning, incorporating vehicles, systems, humans,
14
physical as well as digital infrastructure etc. During a forseeable future (until virtual test-
ing and purely mathematical proofing takes over), it must also be accompanied by field
testing and longish periods of naturalistic driving.
The above also means that driving licenses may take on a new meaning, with new lev-
els or pre-conditions attached to different types of licenses.
Furthermore, from a transport system point of view, the boundaries between AD testing
and traffic safety development are by no means closed, but in fact interdependent. When
running, AD systems and connected vehicles generate huge amounts of data on traffic
situations, interactions, and incidents, both within and beyond the ODD which the sys-
tems are designed to cope with. Machine learning tools may be used to improve exist-
ing algorithms and the performance of “virtual drivers” in preventing and coping with se-
rious incidents. The same tools also provide the obvious way of successively extending
the operating domains of AD systems. Therefore, the potential is high to make large
traffic safety improvements at early stages of the learning curve, if the proper re-
quirements for data gathering and sharing between different stakeholders is accom-
plished.
However, this also raises issues related to the timely validation and certification of
these improvements (into new vehicles) as well as the methods to their integration in ex-
isting vehicles – in order to limit vehicle diversity that may be detrimental to traffic
safety.
Such development also has strong implications for related questions regarding safety and
security, cf. questions i-iv on page 15.
Primary questions in need of investigation are the following:
15. What are the new testing methods and/or tools to be specifically designed, de-
veloped and used to support vehicle automation systems evaluation, validation
and certification? How can the wide set of scenarios experienced in normal driving
(the high frequency events) be represented? How can the scenarios that result in
crashes with human drivers (the high-risk scenarios) be represented? How can the
scenarios which become more challenging for an automated system be repre-
sented?
16. What human user models are required in order to evaluate future use (or mis-
use) of AD systems?
17. How can a “Human-Centered Design” of vehicle automation be used to prevent
new risks when introducing AD? How can the same type of approach be used to
increase acceptance and facilitate deployment?
18. What will be the role of virtual testing and what will be the role of physical test-
ing? What are the consequences if, as predicted, the former will become more
prevalent than the latter?
19. Once a vehicle is validated, how will vehicle updates, especially of software, be
handled in terms of validation and certification?
20. What will be the role of the periodical technical inspection, and who will assume
responsibility in an increasingly shared mobility eco-system? If vehicle fleets are
mostly leased to operators or individuals, this question may potentially be handled
among the market actors themselves, as is already the case in other fields of
product safety.
21. Should the driving license be linked to the level of automation, and if so – in
what way? (This is paralleled today by licenses being conditioned in certain mem-
ber states e.g. on using an automatic rather than stick-shift gear-box, or requiring
the driver to wear corrective glasses or contact lenses to compensate for poor
eye-sight.
15
22. How can we measure the real-world impact of connected and automated vehi-
cles (CAVs)? Current accident investigation structures tell us very little about the
effectiveness of crash avoidance technologies since a good system will avoid
crashes. Currently vehicles with state of the art ADAS are relatively rare in the ve-
hicle fleet and even when they do crash they are not likely to fall into representa-
tive accident sampling schemes. This means that the need for the analysis of
stored data describing the operation of the system is even more essential, but
such data is likely to be encrypted and proprietary, making investigation and evi-
dence-based system improvement virtually impossible. Both direct and indirect
impact needs to be considered (Milakis, Van Arem, & Van Wee, 2017). One solu-
tion would be to fund independent evaluation by the research community of
industry’s innovations, starting immediately as they reach the market. Such an
evaluation program should have the purpose of classifying which AD, ADAS and
ITS solutions yield the best effects, and which AD, ADAS and ITS practices should
be promoted.
23. How can we ensure that AD testing and validation data as well as the huge
amounts of streamed live data from AD and connected vehicles come to optimal
use, from a traffic safety perspective? Large volumes of AD sensor data from dif-
ferent traffic situations, interactions and incidents are continuously generated, and
machine learning tools may profit on these to improve existing AD algorithms to
prevent and cope with serious incidents. What are the proper requirements for
data gathering and sharing between different stakeholders to ensure the maxi-
mum traffic safety benefits?
Related and important questions
The 23 questions brought up above are those which are of primary concern to FERSI,
given our organizations focus on traffic safety. It should however also be mentioned that
several other necessary questions need to be answered and prerequisites met for public
AD and ITS acceptance to reach sufficient levels, and for public safety an integrity to be
guaranteed Among these auxiliary but important questions FERSI finds the fol-
lowing of special interest:
i. How to address the privacy issues caused by data sharing and proliferation, both
for recorded and real-time data-streams? What requirements can and should Eu-
ropean and national traffic administrations put on vehicle manufacturers, fleet op-
erators and individual vehicle owners? Obtaining higher safety levels may come at
the expense of privacy. However, perceived problems with privacy may reduce
public trust in AD vehicles to the point where users refrain from adopting automa-
tion altogether. This in turn would mean missing out on the traffic safety benefits
AD promises to bring. Such data may typically also be (mis-)used for marketing
purposes, leading to further distraction and negative effects on traffic safety. It is
necessary to find a new equilibrium regarding availability and access of data re-
garding individual movement and behaviour on the road.
ii. How to protect road users from fraud and hacker attacks? Of primary concern is
that connected vehicle systems are not compromised, and vehicle and traffic
safety can be guaranteed. In addition, users must have confidence in automated
safety systems, otherwise these will not be adopted. Further investigation and
consequence analysis is needed to ensure effective positive benefits on traffic
safety.
iii. How can society protect its citizens from AD and ITS being used as tools for acts of
terrorism? In recent years, terrorists have resorted to using vehicles in several
attacks across the world, and un-manned vehicles would make for ideal weapons.
Similarly, the more centralized traffic control becomes (which may be necessary
for efficient and effective traffic control and steering), the more vulnerable it
16
becomes to any form of hostile attack, e.g. with the purpose of causing crashes or
blocking traffic.
iv. What is the cost-efficiency of automated driving systems? Is there a point of in-
creasing automation where the marginal costs of further improvements exceed
the benefits of improved casualty reduction, reduction of congestion, increased
mobility or reduced environmental impact? How does user acceptance influence
take-up of AD systems and affect cost-efficiency?
FERSI assumes that questions i-iv will primarily be driven by other organisations, but
since these questions border on traffic safety, their solution will likely also need involve-
ment from FERSIs member organizations.
Summary of recommendations
Automated Driving (AD), along with connected driving and Intelligent Transport Systems
(ITS) are likely to transform the European scene as regards transportation of both people
and goods. Many strong industrial and political driving forces exist, but improving road
safety gets insufficient priority in relation to current development.
In this paper, FERSI points out 23 separate questions, for convenience grouped into four
areas (I – IV as described on page 7 ), towards which pre-normative research and policy
actions should be directed already now, given the rapid rate of technical development in
the area. FERSI concludes that for each of these questions, more research (and possibly
subsequent regulation) is needed for AD and ITS to lead to significant steps forward in
road safety on our European roads. If left unattended, development with regard to these
questions may in fact lead to set-backs in road safety. FERSI therefore recommends that,
at the European level, action is taken to ensure
the prioritization of research funding addressing the 23 questions listed here,
the development of policies that take research findings into account,
the establishment of a scheme for evaluating industry’s innovations by the research
community to establish and proliferate best practices/best system knowledge
among member states, and
the promotion of involvement of other stakeholders, e.g. insurance companies, road
administrations/owners, municipalities/cities, representatives for law enforcement,
the research community as well as organisations representing different groups of
(for instance) vulnerable road users. This is a much-needed complement in addi-
tion to the vehicle and supplier industries which are currently the main actors,
pushing development and policy making.
It is the firm belief of FERSI that this is necessary for AD, connectivity and ITS
to successfully contribute to a smart, green, and integrated transport system
which is also safe.
17
References
ACEM. (2009). MAIDS, In-depth investigations of accidents involving powered two wheelers. Brussels:
Association of European Motorcycle Manufacturers.
Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775-779.
Banks, V., & Stanton, N. A. (In press). Analysis of driver roles: Modelling the changing role of the
driver in automated driving systems using EAST. Theoretical Issues in Ergonomics Science.
Banks, V., Eriksson, A., O'Donoghue, J., & Stanton, N. A. (2018). Is partially automated driving a bad
idea? Observations from an on-road study. Applied Ergonomics , pp. 68, 138–145.
Carsten, O., & Tate, F. (2005). Intelligent speed adaptation: accident savings and cost-benefit
analysis. Accident Analysis & Prevention, 37(3), 407-416.
Davies, A. (2015, November 11). Ford's skipping the trickiest thing about self-driving cars. Retrieved
from Wired: http://www.roboticslawjournal.com/news/level-three-automation-could-be-
skipped-23345642
Eriksson, A., & Stanton, N. A. (2017). Driving performance after self-regulated control transitions in
highly automated vehicles. Human Factors, pp. 59(8), 1233-1248.
Eriksson, A., & Stanton, N. A. (2017). Take-over time in highly automated vehicles: non-critical
transitions to and from manual control. Human Factors, 59(4), 689-705.
doi:10.1177/0018720816685832
Eriksson, A., Lindström, A., Seward, A., Seward, A., & Kircher, K. (2014). Can user-paced, menu-free
spoken language interfaces improve dual task handling while driving? Human-Computer
Interaction. Advanced Interaction Modalities and Techniques, 394-405.
ERSO. (2008). Basic Fact Sheet Motorcycles and Mopeds . The European Road Safety Observatory.
Retrieved from http://erso.swov.nl/data/content/motorcycles_and_mopeds.htm
ERTRAC Working Group "Connectivity and Automated Driving". (2017). ERTRAC Automated Driving
Road Map version 7.0. ERTRAC.
Fagnant, D. J., & Kockelman, K. (2015). Preparing a nation for autonomous vehicles: opportunities,
barriers and policy recommendations. Research Part A: Policy and Practice, 77, 167-181.
Fountoulakis, M., Bekiaris-Liberis, N., Roncoli, C., Papamichail, I., & Papageorgiou, M. (2017).
Highway traffic state estimation with mixed connected and conventional vehicles:
Microscopic simulation-based testing. 17, 13-33.
doi:http://dx.doi.org/10.1016/j.trc.2017.02.015
Ioannou, P. (1998). Evaluation of mixed automated/manual traffic.
Kaira, N., & Paddock, S. M. (2016). Driving to safety: How many miles of driving would it take to
demonstrate autonomous vehicle reliability? Transportation Research Part A: Policy and
Practice, 94, 182-193.
Kyriakidis, M., de Winter, J. C., Stanton, N., Bellet, T., van Arem, B., Brookhuis, K., . . . Merat, N.
(2017). A human factors perspective on automated driving. Theoretical Issues in Ergonomics
Science, 1-27.
Level Three Automation Could Be Skipped. (2017, february 21). Retrieved from The Robotics Law
Journal: http://www.roboticslawjournal.com/news/level-three-automation-could-be-
skipped-23345642
Milakis, D., Van Arem, B., & Van Wee, B. (2017). Policy and society related implications of automated
driving: A review of literature and directions for future research. Journal of Intelligent
Transportation Systems, 1-25.
18
NordicWay consortium. (2017, 11 22). NordicWay. Retrieved from EU project NordicWay:
www.nordicway.net
Pernestål Brenden, A., Kristoffersson, I., & Mattsson, L.-G. (2017). Framtidsscenarier för självkörande
fordon på väg. Samhällseffekter 2030 med utblick mot 2050. Linköping: VTI.
Rizzi, M., Kullgren, A., & Tingvall, C. (2014). Injury crash reduction of low speed Autonomous
Emergency Braking (AEB) on passenger cars. IRCOBI Conference IRC-14-73.
Roncoli, C., Papamichail, I., & Papageorgiou, M. (2015). Model predictive control for motorway traffic
with mixed manual and VACS-equipped vehicles. Transportation Research Procedia, 10, 452-
461.
Schepers, P., Stipdonk, H., Methorst, R., & Olivier, J. (2017). Bicycle fatalities: Trends in crashes with
and without motor vehicles in The Netherlands. Transportation Research, Part F, 46B, 491-
499.
Shalev-Shwartz, S., Shammah, S., & Shashua, A. (2017). On a Formal Model of Safe and Scalable Self-
driving Cars. arXiv:1708.06374v4 [cs.RO] . Retrieved from https://arxiv.org/abs/1708.06374
Singh, S. (2015). Critical reasons for crashes investigated in the national motor vehicle crash causation
survey. NHTSA.
Thorpe, C., Jochem, T., & Pomerleau, D. (1997). The 1997 automated highway free agent
demonstration. IEEE Conf. on Intelligent Transportation System, ITSC '97. IEEE.
Wadud, Z., MacKenzie, D., & Leiby, P. (2016). Help or hindrance? The travel, energy and carbon
impacts of highly automated vehicles. Transportation Research Part A: Policy and Practice,
86(Supplement).
Why Every Car Maker Should Skip Level 3. (2017, June 30). Retrieved from Wonderhowto: Driverless:
https://driverless.wonderhowto.com/news/waymo-was-right-why-every-car-maker-should-
skip-level-3-0178497/
Willemsen, D., Stuiver, A., & Hogema, J. (2015). Automated Driving Functions Giving Control Back to
the Driver: A Simulator Study on Driver State Dependent Strategies. Automated Driving
Functions Giving Control Back to the Driver: A Simulator Study on Driver State Dependent
Strategies.
Volvo to skip Level 3 autonomous mode. (2017). Retrieved from Auto2xtech.com:
http://auto2xtech.com/volvo-to-skip-level-3-autonomous-mode/
19
About FERSI
FERSI, the Forum of European Road Safety Research Institutes (FERSI) is a not-for-profit
association in the area of road safety research.
The Association acts as a flexible network of European road safety research institutions.
It unites partners from 21 European countries who have a mandate of their governments
to implement a pre-normative road safety research. That means that FERSI participants
develop road safety solutions to be translated in legislation and norms, offer consulting to
national and European road safety authorities while implementing these solutions and
evaluate effects of implementation.
FERSI is open for cooperation with any road safety research institute or organisation,
both in exchanging research knowledge and in seeking opportunities for collaborative re-
search. Road safety research entities form outside Europe could join FERSI as associated
partners.
FERSI acts on behalf of its partners as their representative vis-à-vis the European Com-
mission, European Parliament, European Council while promoting road safety research
agenda and research results and acquiring funding for research projects.
FERSI represents its partners and their reconciled opinion in the dialogue with other in-
ternational organizations contributing to the road safety improvement, e.g. OECD, UN,
professional unions and branch associations.
The association supports its partners in disseminating the knowledge generated by them
and is actively involved in knowledge and information exchange.
FERSI mission and objectives
The primary mission of the Association is to promote or coordinate high quality research
on road safety issues, consult on implementation of research results and scientifically
evaluate implementation outcomes.
The objective of the Association is to contribute to road safety research and enhance the
road safety by ensuring that the relevant problems are addressed by best available ex-
perts and that solutions recommended by researchers are implemented in the most ef-
fective way.
In particular, FERSI may engage in the following (not exclusively) activities:
enhancing the scientific quality of road safety research;
proposing road safety topics for EU and national Research Agendas;
developing road safety research recommendations and proposing them to compe-
tent national and European authorities;
organizing seminars and meetings on road safety issues;
participating in road safety research work conducted by EU or other international
organizations;
encouraging the international exchange of researchers;
awarding prizes or scholarships financed by its resources to road safety research-
ers.
20
Main contributors to this position paper
Anders Lindström, VTI (editor and WG leader)
Wouter Van den Berghe, VIAS
Julie Maes, VIAS
Klaus Machata, KfV
Rob Eenink, SWOV
Astrid Linder, VTI
Alexander Eriksson, VTI
Pete Thomas, Loughborough University
Thierry Bellet, IFSTTAR
Stefan Siegrist, BFU
Lila Gaitanidou, CERTH
João Cardoso, LNEC
Merja Penttinen, VTT
Jindřich Frič, CDV
Members of the FERSI Working Group on ITS and Automation
Thierry Bellet, IFSTTAR
Rob Eenink, SWOV
Virginie Etienne, IFSTTAR
Jindřich Frič, CDV
Klaus Machata, KfV
Dominique Mignot, IFSTTAR
Lila Gaitanidou, CERTH
Astrid Linder, VTI
Anders Lindström, VTI
Juha Luoma, VTT
Julie Maes, VIAS
Merja Penttinen, VTT
Stefan Siegrist, BFU
Horst Schulze, BASt
Raschid Urmeew, BASt
Wouter Van den Berghe, VIAS
Nicole van Nes, SWOV
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
The automation of longitudinal and lateral control has enabled drivers to become “hands and feet free” but they are required to remain in an active monitoring state with a requirement to resume manual control if required. This represents the single largest allocation of system function problem with vehicle automation as the literature suggests that humans are notoriously inefficient at completing prolonged monitoring tasks. To further explore whether partially automated driving solutions can appropriately support the driver in completing their new monitoring role, video observations were collected as part of an on-road study using a Tesla Model S being operated in Autopilot mode. A thematic analysis of video data suggests that drivers are not being properly supported in adhering to their new monitoring responsibilities and instead demonstrate behaviour indicative of complacency and over- trust. These attributes may encourage drivers to take more risks whilst out on the road.
Article
Full-text available
Objective: This study aims to explore whether driver-paced, noncritical transitions of control may counteract some of the aftereffects observed in the contemporary literature, resulting in higher levels of vehicle control. Background: Research into control transitions in highly automated driving has focused on urgent scenarios where drivers are given a relatively short time span to respond to a request to resume manual control, resulting in seemingly scrambled control when manual control is resumed. Method: Twenty-six drivers drove two scenarios with an automated driving feature activated. Drivers were asked to read a newspaper or monitor the system and relinquish or resume control from the automation when prompted by vehicle systems. Driving performance in terms of lane positioning and steering behavior was assessed for 20 seconds post resuming control to capture the resulting level of control. Results: It was found that lane positioning was virtually unaffected for the duration of the 20-second time span in both automated conditions compared to the manual baseline when drivers resumed manual control; however, significant increases in the standard deviation of steering input were found for both automated conditions compared to baseline. No significant differences were found between the two automated conditions. Conclusion: The results indicate that when drivers self-paced the transfer back to manual control they exhibit less of the detrimental effects observed in system-paced conditions. Application: It was shown that self-paced transitions could reduce the risk of accidents near the edge of the operational design domain. Vehicle manufacturers must consider these benefits when designing contemporary systems.
Article
Full-text available
In this paper, the potential effects of automated driving that are relevant to policy and society are explored, findings discussed in literature about those effects are reviewed and areas for future research are identified. The structure of our review is based on the ripple effect concept, which represents the implications of automated vehicles at three different stages: first-order (traffic, travel cost, and travel choices), second-order (vehicle ownership and sharing, location choices and land use and transport infrastructure) and third-order (energy consumption, air pollution, safety, social equity, economy and public health). Our review shows that first-order impacts on road capacity, fuel efficiency, emissions and accidents risk are expected to be beneficial. The magnitude of these benefits will likely increase with the level of automation and cooperation and with the penetration rate of these systems. The synergistic effects between vehicle automation, electrification and sharification can multiply these benefits. However, studies confirm that automated vehicles can induce additional travel demand because of more and longer vehicle trips. Potential land use changes have not been included in these estimations about excessive travel demand. Other third-order benefits on safety, economy, public health and social equity still remain unclear. Therefore, the balance between the short-term benefits and long-term impacts of vehicle automation remains an open question
Article
Full-text available
Automated driving can fundamentally change road transportation and improve quality of life. However, at present, the role of humans in automated vehicles (AVs) is not clearly established. Interviews were conducted in April and May 2015 with twelve expert researchers in the field of Human Factors (HF) of automated driving to identify commonalities and distinctive perspectives regarding HF challenges in the development of AVs. The experts indicated that an AV up to SAE Level 4 should inform its driver about the AV’s capabilities and operational status, and ensure safety while changing between automated and manual modes. HF research should particularly address interactions between AVs, human drivers, and vulnerable road users. Additionally, driver training programs may have to be modified to ensure that humans are capable of using AVs. Finally, a reflection on the interviews is provided, showing discordance between the interviewees’ statements—which appear to be in line with a long history of human factors research—and the rapid development of automation technology. We expect our perspective to be instrumental for stakeholders involved in AV development and instructive to other parties.
Article
Full-text available
Objective: The aim of this study was to review existing research into driver control transitions and to determine the time it takes drivers to resume control from a highly automated vehicle in noncritical scenarios. Background: Contemporary research has moved from an inclusive design approach to adhering only to mean/median values when designing control transitions in automated driving. Research into control transitions in highly automated driving has focused on urgent scenarios where drivers are given a relatively short time span to respond to a request to resume manual control. We found a paucity in research into more frequent scenarios for control transitions, such as planned exits from highway systems. Method: Twenty-six drivers drove two scenarios with an automated driving feature activated. Drivers were asked to read a newspaper, or to monitor the system, and to relinquish, or resume, control from the automation when prompted by vehicle systems. Results: Significantly longer control transition times were found between driving with and without secondary tasks. Control transition times were substantially longer than those reported in the peer-reviewed literature. Conclusion: We found that drivers take longer to resume control when under no time pressure compared with that reported in the literature. Moreover, we found that drivers occupied by a secondary task exhibit larger variance and slower responses to requests to resume control. Workload scores implied optimal workload. Application: Intra- and interindividual differences need to be accommodated by vehicle manufacturers and policy makers alike to ensure inclusive design of contemporary systems and safety during control transitions.
Article
In recent years, car makers and tech companies are racing toward self driving cars. It seems that the main parameter in this race is who will have the first car on the road. The goal of this paper is to add to the equation two additional crucial parameters. The first is standardization of safety assurance --- what are the minimal requirements that every self-driving car must satisfy, and how can we verify these requirements. The second parameter is scalability --- engineering solutions that lead to unleashed costs will not scale to millions of cars, which will push interest in this field into a niche academic corner, which might drive the entire field into a "winter of autonomous driving". In the first part of the paper we propose a white-box, interpretable, mathematical model for safety assurance. In the second part we describe a design of a system that adheres to our safety assurance requirements and is scalable to millions of cars.
Article
This paper presents a thorough microscopic simulation investigation of a recently proposed methodology for highway traffic estimation with mixed traffic, i.e., traffic comprising both connected and conventional vehicles, which employs only speed measurements stemming from connected vehicles and a limited number (sufficient to guarantee observability) of flow measurements from spot sensors. The estimation scheme is tested using the commercial traffic simulator Aimsun under various penetration rates of connected vehicles, employing a traffic scenario that features congested as well as free-flow conditions. The case of mixed traffic comprising conventional and connected vehicles equipped with adaptive cruise control, which feature a systematically different car-following behavior than regular vehicles, is also considered. In both cases, it is demonstrated that the estimation results are satisfactory, even for low penetration rates.
Article
This paper discusses the ways in which automation of industrial processes may expand rather than eliminate problems with the human operator. Some comments will be made on methods of alleviating these problems within the 'classic' approach of leaving the operator with responsibility for abnormal conditions, and on the potential for continued use of the human operator for on-line decision-making within human-computer collaboration.
Article
How safe are autonomous vehicles? The answer is critical for determining how autonomous vehicles may shape motor vehicle safety and public health, and for developing sound policies to govern their deployment. One proposed way to assess safety is to test drive autonomous vehicles in real traffic, observe their performance, and make statistical comparisons to human driver performance. This approach is logical, but it is practical? In this paper, we calculate the number of miles of driving that would be needed to provide clear statistical evidence of autonomous vehicle safety. Given that current traffic fatalities and injuries are rare events compared to vehicle miles traveled, we show that fully autonomous vehicles would have to be driven hundreds of millions of miles and sometimes hundreds of billions of miles to demonstrate their reliability in terms of fatalities and injuries. Under even aggressive testing assumptions, existing fleets would take tens and sometimes hundreds of years to drive these miles?an impossible proposition if the aim is to demonstrate their performance prior to releasing them on the roads for consumer use. These findings demonstrate that developers of this technology and third-party testers cannot simply drive their way to safety. Instead, they will need to develop innovative methods of demonstrating safety and reliability. And yet, the possibility remains that it will not be possible to establish with certainty the safety of autonomous vehicles. Uncertainty will remain. Therefore, it is imperative that autonomous vehicle regulations are adaptive?designed from the outset to evolve with the technology so that society can better harness the benefits and manage the risks of these rapidly evolving and potentially transformative technologies.