On Social Interactions and the Emergence of Autonomous Vehicles

Carolina Centeio Jorge and Rosaldo J. F. Rossetti

Departamento de Engenharia Inform

atica, Faculdade de Engenharia da Universidade do Porto,

Rua Dr Roberto Frias, s/n – 4200-465, Porto, Portugal

Keywords:

Autonomous Vehicles, Social Interactions, Transport Systems, Human-Technology Interaction.

Abstract:

Nowadays and in the contemporary age, the reality of an all-autonomous trafﬁc seems closer and closer.

However, this transition period casts a lot of cards onto the table. Although technology can be replacing people

at the driver seat, it has not as yet gained our full trust in what concerns communication in real time and safety.

Humans interact on a daily basis in their various activities, and trafﬁc is no exception. Most actions performed

on the road rely on our perception of others’ awareness and potential reactions. For instance, pedestrians seek

for an eye contact before crossing the road, drivers seek for a gesture before starting a manoeuvre, and so forth.

Thus, the question remaining is what happens when someone is seeking such a communication interaction and

the car has no driver, nor has it someone who even knows what the car is doing. Moreover, people seating in

the car might be performing any other activities but driving. Other questions also arise such whether people

will accept the idea of trusting self-driving vehicles, or whether will they feel safe when walking amongst such

machines. In this paper we pursue a rather social perspective and will raise questions, covering the literature

so as to understand what practitioners, researchers and the industry have been doing to overcome the lack of

conﬁdence in self-driving cars and improve their trustworthiness towards more efﬁcient and smarter mobility,

as well as to identify trends and approaches to answer these emerging questions.

1 INTRODUCTION

As technology moves forward at an immensely accel-

erated pace, the old-fashioned utopia of self-driven

cars is an ever-closer reality. Bearing this in mind,

many questions arise, such as what impacts these au-

tonomous vehicles will have on pedestrians’ lives.

Undeniably, this subject has been the target of much

research, leading to many other still unanswered

questions.

Although technology can be replacing people at

the driver seat, trustworthiness of such new technolo-

gies is still to be proven. People are used to a cer-

tain kind of interactions, and this leads to constant and

patterned behaviours such as seeking for eye contact

before crossing the road. The same happens in cross-

roads when drivers (or a cyclist and a driver) look at

each other before yielding, to give the other assurance

that they are safe if they proceed. With a partially

autonomous trafﬁc, it is necessary to try and foresee

how people trust the car behaviour and what they do

when confronted with one. And later on, in an all-

autonomous trafﬁc, it is important to state whether

people should trust blindly the car sensory and de-

cision system and risk crossing the road. Recent re-

search shows that people feel reticent about crossing

the road in front of a car without a driver they can look

at so as to get any kind of signal.

In this paper, we aim to evaluate this kind of sce-

narios, summarizing the current state of the art and re-

ﬂecting upon what has been done, what will be done

and what needs to be done. This literature review

may lead to new questions and research ideas, since

the most recent research will be analysed and counter-

weighted with older studies, models and simulations,

as well.

The rest of this paper is structured as follows.

After giving an insight into the current scenarios in

road networks, Section 2 discusses on the pedes-

trian behaviour as well as the driver behaviour at

marked crossings. A brief explanation of the tradi-

tional models of trafﬁc dynamics from the perspective

of the driver’s decision-making processes are also pre-

sented. Section 3 focuses on recent studies tackling

the problems of partially autonomous trafﬁc interac-

tions and what has been done to address the many is-

sues arising in this context. Finally, we reﬂect on all

the issues stated above and suggest some ideas for fu-

ture work, as we reach the conclusion of this paper.

Centeio Jorge, C. and Rossetti, R.

On Social Interactions and the Emergence of Autonomous Vehicles.

DOI: 10.5220/0006763004230430

In Proceedings of the 4th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2018), pages 423-430

ISBN: 978-989-758-293-6

423

2 SOCIAL INTERACTIONS IN

TRAFFIC

Human beings interact by attempting to create a mu-

tual behaviour adaptation process with other individ-

uals through verbal and non-verbal communication.

Such non-verbal communication can be reﬂected in

posture, gaze and other kinds of body language (Lehs-

ing et al., 2016b). This section discusses on how so-

cial interactions in trafﬁc are an essential communi-

cation means that should also be taken into consider-

ation even when autonomous vehicles are to be de-

ployed.

2.1 Current Scenario

Needless to say that interactions between road users

(e.g. pedestrians, drivers and cyclists) is not only

based on a set of road rules but it is also dependent

on a large range of informal social protocols such as

eye contact, gestures and other social cues. There-

fore, the less the road is signalised with trafﬁc lights

and other signs, the more a safe navigation depends

on the ability of a person (either a driver or a pedes-

trian) to perceive and interpret these social cues (Earl

et al., 2016). The behaviour of a pedestrian on marked

crossings, a cyclist waving to signal his/her willing to

turn left and even a driver changing lane are situations

that require quick problem-solving thinking based on

visual perception. Research in the ﬁeld of transport

engineering and trafﬁc psychology has applied meth-

ods to better understand how such interactions occur

so that dangerous situations and fatalities can be ef-

fectively reduced (Lehsing et al., 2016b).

2.1.1 Pedestrian Behaviour at Marked Crossings

In its safety reminders for pedestrians the U.S. De-

partment of Transportation (US DOT) recommends

people to make eye contact with drivers as they ap-

proach you to make sure you are seen (NHTSA,

2012). With just a look, one can estimate the car

speed and the distance of the car from the cross-

ing, highlighting the importance of eye contact in

pedestrian-driver interactions. Bearing this in mind,

we calculate the time we have to cross without get-

ting hit by a car or another vehicle; ﬁnally, we decide

whether we can cross the road or need to wait. How-

ever, more often than expected, one factor that af-

fects our decision is the various signs given by drivers.

Waiting for these signs, which range from eye contact

to the waving of a hand or the ﬂashing of headlights, is

a consequence of the fact that drivers who give way to

pedestrians usually prefer to lower their speed rather

Figure 1: Pedestrian Intention (Sucha et al., 2017).

than to bring their car to halt. Pedestrians feel the

need to ensure their own safety. In fact, as we can see

in Fig. 1, 84% of pedestrians wait on the pavement

showing their intent to use the crossing sought to es-

tablish eye contact with the driver of an approaching

car (Sucha et al., 2017).

2.1.2 Driver Behaviour at Marked Crossings

Although the great majority of the pedestrians are

proactive in seeking eye contact or trying to estab-

lish other forms of explicit communication with the

drivers, only 39% of the drivers give them feedback,

as shown in Fig. 2. Thus, it makes it hard for a pedes-

trian to know whether the car will yield the expected

reaction. On the other hand, drivers are inﬂuenced

by the expression of a pedestrian. When pedestrians

are engaged in activities such as talking on the phone,

texting, reading (e.g. newspapers), talking to another

pedestrian or any other non-transport-related activity,

a driver might perceive this as a lack of willingness or

preparedness by the pedestrian to cross the road.

Figure 2: Driver Feedback (Sucha et al., 2017).

2.1.3 Uncontrolled Mid-block Crossings

The interaction during trafﬁc conﬂicts at mid-block

cross walks can be portrayed as a competition for

limited resources, mostly time and space. A study

shows that game theory is applicable to the analysis

of the interaction between vehicles and pedestrians as

such theory is based on the rational behaviour exhib-

ited in interpersonal conﬂicting situations. In order

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

424

to deal with the limited rationality in virtually most

decision-making processes, evolutionary game theory

is presented as an extension of the classical paradigm

towards bounded rationality in the study by (Chen

et al., 2016). In such cases, the developed model

is calibrated and validated using real data collected

at Jianshe First Road in Wuhan, China. The results

show that the proposed model is able to properly sim-

ulate the interaction between vehicles and pedestrians

(Chen et al., 2016).

2.1.4 Traditional Models in Trafﬁc

Other social interactions worth understanding are the

ones that happen in trafﬁc, amongst drivers. Mixed

trafﬁc conditions are commonly characterised by the

presence of different types of vehicle and behaviours.

Different manoeuvring capabilities of different vehi-

cle types lead to vehicle-type-dependent longitudinal

and lateral movement driving behaviours. Weak lane

discipline allows drivers to simultaneously look for

possible lateral movements while progressing longi-

tudinally. All of this gives rise to various driving phe-

nomena and microscopic models — an approach that

gives focus onto a ﬁner grained perspective of trafﬁc

ﬂow and the inner-workings of its individual particles

(Munigety and Mathew, 2016). That being said, it is

important to understand how drivers behave in these

situations as an attempt at predicting the problems in

the interactions between drivers and autonomous ve-

hicles.

Figure 3: Microscopic Models (Munigety and Mathew,

2016).

According to (Munigety and Mathew, 2016), a

tree classiﬁcation of different car-following and lane-

changing models are shown in Fig. 3. An exten-

sive review of these driver behavioural models is also

described by the authors. Lane changing models

are gaining prominence within the scientiﬁc research

community, since they are one of the most frequent

interactions implemented in trafﬁc simulations.

In Fig. 4, vehicle M is the vehicle that wants to

make a lane change, vehicle C and vehicle D are the

Figure 4: Lane Changing (Nagahama et al., 2017).

leading and the following vehicles in the same lane as

vehicle M, respectively; vehicle B and vehicle A are

the leading and following vehicles, respectively, in the

target lane. Car-following is, as the name says, the ac-

tion of a car following another. For example, M is fol-

lowing C (Fig. 4) (Nagahama et al., 2017). Most lane-

changing models are based on the hypothetical no-

tion that when the vehicle of an adjacent lane changes

lane, the following vehicle of the target lane keeps a

uniform motion. However, this hypothesis does not

reﬂect the real lane-changing scenario. A paper has

already put forward a lane-changing model with the

consideration of car following behaviours in mind,

focusing on the kinematic behaviour of the lane-

changing vehicle in the process of accelerated lane

change (Xiaorui and Hongxu, 2013). The vehicle in-

teraction for lane changing is more complex than that

of a car-following event because more vehicles are in-

volved in the lane-changing event, which requires a

greater workload of the subject vehicle driver. The

lane-changing event is associated with a higher crash

potential due to complex interactions with neighbour-

ing vehicles (Oh et al., 2017) . According to the usual

driver behaviour, when a vehicle wants to run in front

of his own vehicle and he accepts this behaviour, then

he will speed down to ensure safety for both. Consid-

ering the car-following behaviour, and accounting for

the real lane-changing scenario, and mainly focusing

on the kinematic behaviour of the lane-changing ve-

hicle in the process of accelerated lane change, the

study devises a lane-changing model with the con-

sideration of a car-following behaviour (Xiaorui and

Hongxu, 2013).

2.1.5 Communication Barrier

As stated above, eye contact plays a very important

role on pedestrians’ sense of safety. It also avoids

danger when the driver, due to distraction or any other

reason, is not willing to stop. This kind of communi-

cation can be signiﬁcantly reduced when the driver’s

car has deeply tinted windows. This situation can be

compared to a car with no driver, since both make it

very difﬁcult to understand cues. Deeply tinted win-

dow glass transmits less light than less deeply tinted

glass and therefore reduces driver visibility. The task

of looking through the rear window in dangerous situ-

ations was stimulated in a laboratory setting with road

On Social Interactions and the Emergence of Autonomous Vehicles

425

users such as pedestrians and cyclists. The car was

always detected, but detection probability decreased

with reduced luminous transmittance for the child and

roadway debris targets. The results suggest that the

safety of backing manoeuvres is compromised for all

drivers at the darkest tinting levels studied (Freedman

et al., 1993). Correspondingly, the UK government

demands that the front windscreen must let at least

75% of light through and the front side windows must

let at least 70% of light through for all vehicles ﬁrst

used on April the 1

, 1985 or later (United Kingdom

Government, ). The way this scenario affects the be-

haviour of other road users may lead us to a better

understanding of some of the problems on the inter-

action between humans and driverless cars.

2.2 Simulating Current Scenario

2.2.1 Driver-pedestrian Interactions

In order to analyse this behaviour, in addition to the

research in the ﬁeld of transport engineering and traf-

ﬁc psychology to reveal underlying processes and key

factors in trafﬁc that lead to dangerous situations or

fatalities, several driving simulators have been imple-

mented. Some use very complex apparatuses in full

scale so as to reach full immersion, whereas others are

considered low-cost environments resorting to sim-

pler computer environments based on video games

(Rossetti et al., 2013).

A pedestrian simulator, based on motion track-

ing technology, combined with a driving simulator al-

lowed the participants to communicate non-verbally.

This approach approximates the study to real-life sit-

uations since it creates a communication bridge be-

tween both the driver and the pedestrian in pedestrian-

crossing situations. The interaction was measured

using cross recurrence quantiﬁcation analysis as de-

scribed in (Lehsing et al., 2016b).

A simulation model for pedestrian-vehicle inter-

actions at unsignalised mid-block cross walks cap-

tures the behaviour of both vehicles and pedes-

trians when approaching unsignalised interactions.

The model is based on a cellular-automata ant

metaphor that is calibrated with detailed behavioural

data collected and extracted from observations of

two unsignalised intersections in Nanjing, China.

In particular, this simulation model can replicate

pedestrian-vehicle interaction and pedestrian delay

with high accuracy and reliability (Chen et al., 2016).

The classical and common approach of driving

simulation as a tool in trafﬁc research is usually

limited to one driving simulator. However, the be-

haviour of the drivers depends on the trafﬁc environ-

ment. Therefore, it is important for the program to

gather different scenarios with a higher number of

programmed road users. Unfortunately, this is a bot-

tleneck due to programming skills required and soft-

ware limitations. The proposed method for linking

simulators is said to have the potential to create a

more human-like behaviour by means of opening in-

teraction channels between road users in a driving

simulation (Lehsing et al., 2016a). Nonetheless, other

studies succeeded in combining different simulators

to represent more complex systems, specially when

different model resolutions were combined (Macedo

et al., 2013; Perrotta et al., 2014; Perrotta et al., 2012).

Many simulation models have been studied lately

to better instil safety to road users, especially pedes-

trians at cross walks. Nonetheless, these models do

not include the non-verbal communication aspects, in

which most road users rely on, as discussed in parts

2.1.1 and 2.1.2.

2.2.2 Driver-driver Interactions

Research has also been done in order to model and

simulate traditional models (3) in trafﬁc. As stated

above, these models focus on the interactions between

drivers, such as lane changing and car following. In

(Morton et al., 2017), authors also refer methods suit-

able to learn such driving models from real-world

data. This, however, is not within the scope of this

work.

3 PARTIALLY AUTONOMOUS

TRAFFIC INTERACTIONS

New communication requirements emerge when au-

tonomous vehicles are introduced into the urban con-

text. Certainly they are expected to overcome many

of today’s trafﬁc limitations; however, the transi-

tion between vehicles steered by humans to fully au-

tonomous trafﬁc poses enormous challenges calling

for appropriate attention.

3.1 Understanding the Problem

As already stated, people are often dependant on and

inﬂuenced by other people’s behaviour. In (Laznyi

and Marczi, 2017) authors state that the level of dis-

positional trust towards autonomous systems among

young adults is low. As when trusting other hu-

mans, trusting smart systems depends on those sys-

tems sharing the users goals. Moreover, to gain opti-

mal acceptability, goals of the user should be shared

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

426

by the smart systems, and smart systems should pro-

vide information to their user (Verberne et al., 2012).

Also, authors argue that experiencing highly auto-

mated driving within driving simulators can increase

self-reported trust in automation. However, although

elsewhere (Choi and Ji, 2015) authors state that per-

ceived usefulness and trust are major important deter-

minants of intention to use autonomous vehicles, and

that three constructs (system transparency, technical

competence, and situation management) have a posi-

tive effect on trust, this study identiﬁed that trust has

a negative effect on perceived risk.

In order to reach a full autonomous scenario,

which will be publicly available in the near future,

there will be a time between non-autonomous trafﬁc

and full-autonomous trafﬁc, in which we need to cope

with the challenges of partially autonomous trafﬁc. It

can be assumed that such transition will not be in-

stantaneous. This being said, it cannot be assumed

that humans will be completely out of the scene or

off the roads during this transitory period (Driggs-

Campbell et al., 2017). Instead, ﬁnding humans’

place in this new scenario and understand how we will

react to driverless cars is imperative, not only when

we are driving, but also when we are crossing a road

or even when travelling as passengers inside these

vehicles. In (Driggs-Campbell et al., 2017) authors

present a literature review of relevant driver mod-

elling frameworks for cooperative driving and present

a non-parametric driver model that can be adapted to

many different applications in the so-called human-

in-the-loop predictions.

Furthermore, connected cars, which are net-

worked to trafﬁc sensory and on-line information

about road conditions, will take the driver’s seat, act-

ing in a way that the driver will not always understand

or will ﬁnd counter intuitive. Elsewhere (Koo et al.,

2015), authors suggest that autonomous cars need ap-

propriate means to explain their actions to the drivers

so as to increase overall safety and become more reli-

able.

In particular, and we will give special attention

to this scenario, pedestrians usually seek interactions

such as eye contact, posture, and gestures. Simulta-

neously, as automation is increasing, people become

unable to rely on this crucial factor — i.e. human

feedback. Thus, new ways of communicating are

necessary in order to warrant safe interactions within

autonomous-vehicle settings. For this, pedestrians

willingness to cross the street and their emotional

state in encounters with a seemingly autonomous ve-

hicle need to be explored. Studies show that pedes-

trians willingness to cross the street decreases with

an inattentive driver. In contrast, eye contact with

the driver leads to calmer, more comprehensive and

workable interactions (Lundgren et al., 2017). On the

other hand, gestures are usually difﬁcult to interpret

due to cultural variations. For instance, hand gestures

can be used for counting or expressing other mes-

sages, varying from culture to culture. Since these

elementary gestures are not universal or unambigu-

ous, then a work into the universality of movement

gestures is also necessary (Gupta et al., 2016).

To further explore the consequences of these en-

counters, a proof-of-concept study was implemented

at a cross walk and a trafﬁc circle. In the study, par-

ticipants encountered a vehicle that appeared to have

no driver — the Ghost Driver experience. This vehi-

cle was driven by a human confederate hidden inside

a car seat costume. Pedestrians who encountered the

car reported that they saw no driver, yet they man-

aged interactions smoothly, except when the car mis-

behaved by moving into the cross walk just as they

were about to cross. In light of the aforementioned,

practices such as making eye contact or making hand

gestures will no longer be a reliable means of commu-

nication. Road users cannot observe any head move-

ments indicating that they have been noticed to ad-

vance. This is not only connected to the pedestri-

ans’ safety but also to their comfort when walking

on the road. Not only can pedestrians increase their

own safety by interacting with car drivers through

signals and gaze, but also with cyclists, since drivers

tend to make decisions about their intentions by look-

ing at the cyclist’s face. A driver’s gaze goes ﬁrst

onto the face of a cyclist and remains there for longer

periods than any gaze directed at the cyclist’s hand

signs (Rothenbucher et al., 2016). A Wizard of Oz

(WOZ) methodology for evaluating these interactions

with autonomous vehicles has also been implemented

more than once (Lundgren et al., 2017; Rothenbucher

et al., 2016). The WOZ technique enables unimple-

mented technology to be evaluated by using a human

to simulate the responses of a system. The wizard —

a person who pretends to act as an autonomous sys-

tem — is hidden, observing the user’s actions. The

wizard then simulates the system’s responses in real

time. Since participants do not suspect that a wiz-

ard, i.e. a human entity, was behind the feedback

they were receiving, the method can be said to be

successful. Also, this is an easily implemented but

nonetheless powerful tool for gathering information

at this early stage of the shift to an autonomous traf-

ﬁc, as it requires little software resources to yield a

well-presented deception. The WOZ method sim-

ulated interaction in realistic trafﬁc situations, al-

though achieving repeatability in such dynamic set-

tings can be challenging sometimes (Rothenbucher

On Social Interactions and the Emergence of Autonomous Vehicles

427

et al., 2016). In another experience (Lundgren et al.,

2017), all pedestrians (N = 13 of 13) that met the

standstill vehicle stated that they would only cross the

street once they had established eye contact with the

driver. The willingness to cross was reduced when

the driver was talking on the phone, (N = 10 of 13).

But it was further reduced when the driver was read-

ing a newspaper (N = 5 of 13), or even when there

was no driver present in the vehicle (N = 5 of 13). All

pedestrians (N = 13 of 13) stated that eye contact with

the driver, and the driver behaviour and interaction in

general, changed the experience completely. When

asked about the safest encounter, all pedestrians (N =

13 of 13) afﬁrmed they felt most safe when they got

eye contact with the driver (Lundgren et al., 2017).

3.2 Possible Solutions

To sustain perceived safety when eye contact is dis-

carded due to vehicle automation some solutions have

lately been proposed. For instance, a project treated

autonomous vehicles as social robots; instead of es-

tablishing eye contact, humans could acknowledge

vehicles’ intentions through audio and visual cues.

This could assure other road users of the vehicle’s in-

tentions. Without this assurance, road users can eas-

ily get scared while walking in shared zones. The

pedestrian has no clear way of ensuring that the driver

sees him/her moving around the vehicle as there is

no driver to interact with. And people inside an au-

tonomous vehicle have no sure way of knowing that

the car is noticing the pedestrian and whether has it

the intention to move on or slow down and let the

pedestrian cross. As a result, the pedestrian may not

be the only one not feeling safe with the idea of au-

tonomous vehicles; passengers can feel scared too!

So far a number of researchers have studied ways

of turning the vehicle more sociable. This is il-

lustrated by showing motion intentions of the au-

tonomous vehicles, such as the route destination and

mission state, through a LED message board. An

audio cue, as ﬁrstly referred, in the form of music

has also been broadcast through the speaker while the

vehicle was driving autonomously to capture the at-

tention of the surrounding pedestrians and other road

users, who otherwise might not notice that the vehi-

cle did not have a human driver. This LED strip to

broadcast obstacle detections from a LiDAR (remote

sensing technology that measures distance by illumi-

nating a target with a laser and analysing the reﬂected

light) was found to be an effective method to acknowl-

edge pedestrian presence (Florentine et al., 2016). A

research has been carried out in order to catalogue

the gestures of transport authorities of different states.

This classiﬁcation of hand and body movements has

not only contributed to understanding the psychology

of such gestures, but also to the categorisation of hand

rules. This research can provide revelations about the

various elements involved in such gestures, which in-

clude arm movement, eye behaviour, body posture

and head movement (Lundgren et al., 2017).

3.3 Future Approaches

Researchers say there is a need to draw conclusions

about other trafﬁc situations (e.g. crowded crossings,

turning manoeuvres) and other cultural settings, since

one of the WOZ tests were carried out in a small

Swedish population (Lundgren et al., 2017). Sim-

ilar methods to the Ghost Driver should be imple-

mented to investigate other types of encounters with

an autonomous vehicle. In particular, pedestrians’ be-

haviour when coming across an autonomous car in

which the only visible person in the car displays atyp-

ical behaviour inside the car. From eating, to reading,

applying make-up or any other kind of leisure activ-

ity in the autonomous vehicle instead of driving, it

should be noted how pedestrians respond to the per-

son in the car and whether they would attempt to com-

municate back with the pedestrian. Since these people

do not necessarily know it is an autonomous vehicle,

it might be a cause for concern when the person in-

side the car is doing an unrelated task. Finally, the

research about the gestures could extend the study to

peoples interpretation of these gestures. Future work

will also address the question whether any of these

agreed gestures are also used by pedestrians and hu-

man drivers. Therefore, autonomous vehicles need to

understand the intentions beyond gesture signs.

4 CONCLUSIONS

Taking all this into account, interaction in trafﬁc will

change as the number of autonomous vehicles circu-

lating on the roads increases. Nowadays, pedestrians’

feeling of being safe is much dependent on eye con-

tact and other non-verbal forms of communication. In

a near future, road users will have to communicate

with autonomous vehicles. Not only pedestrians but

also cyclists and other drivers will have to interpret

signals from a machine to know how to proceed in

trafﬁc.

This literature review aims to show how road users

interact with each other, approaching vehicles’ inter-

actions’ traditional models such as car-following and

lane-changing, pedestrian-driver behaviour at both

marked and unsignalised cross walks.

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

428

Current concerns are discussed as well as are

the latest research done on social interactions with

autonomous vehicles. In our opinion, the future

can bring new approaches to address these concerns,

which may vary from audio and visual cues on the au-

tonomous vehicle to real-time body language recogni-

tion. Finally, some ideas of future work that we con-

sidered worth investigating are presented at the end

of the recent studies section. These ideas range from

new data collections about other trafﬁc situations, dif-

ferent cultural settings and its analysis to new fake en-

counters that test people’s reactions.

We believe this paper helps understanding the

biggest concerns that are triggered when we talk

about inserting autonomous vehicles in a mostly

human-based trafﬁc. In addition, it has highlighted

which questions need answers and further investi-

gation. Indeed, as future work, we would like to

explore how tinted windows affect the communica-

tion between the driver and the other road users (i.e.

other drivers, cyclists and pedestrians). Due to such

glass characteristics, the light reﬂected prevents peo-

ple who are outside the car to easily see the driver’s

gestures and expressions. We believe that testing the

interaction between a driver in a tinted-windowed car

and a pedestrian at a cross walk, as well as another

driver, can be an easy way of testing people’s be-

haviour in a situation in which it is hard for them to

establish eye contact or to get another kind of feed-

back from other gestures, in order to better manage

their own safety.

Also, this can be useful to test the results of us-

ing the headlights of an autonomous vehicle as a

means to signal pedestrians or any other road user

about the vehicle’s intentions. Different communica-

tion metaphors can be easily tested even with low-

cost driving simulators (Gonc¸alves et al., 2012; Alves

et al., 2013; Gonc¸alves et al., 2014), in which sub-

jects play drivers and pedestrians in virtual environ-

ments may give important insights into how such in-

teractions emerge when visual clues are not present.

Once we get the data about people’s behaviour in a

situation in which it is hard for them to establish eye

contact or to get another kind of feedback, we aim to

build a dataset that we intend to use for a social sim-

ulation approach to reﬁne the existing models or even

ﬁnd a more appropriate behavioural mining approach.

Ideally, mining large volumes of recorded data to

extract useful information about behaviour patterns

of individuals and groups of individuals in the area

under study, and monitoring the scene in real-time

in order to provide an immediate response would be

a desirable outcome. Learning spatio-temporal be-

haviour patterns from a public space is frequently of

intrinsic commercial or security interest for users to

gain more knowledge about activity patterns in pub-

lic spaces they are responsible for (Hospedales et al.,

2012). Having said that, we could use behavioural

mining to identify typical behaviours in pedestrian-

driver social interactions. The nature of behavioural

patterns in a given scenario could be deﬁned by a wide

variety of factors, affecting the activity of a single ob-

ject both over space and over time. Building models

general and ﬂexible enough to represent all these as-

pects of behaviour is an open research question need-

ing further investigation (Hospedales et al., 2012). In

the next steps of our research, we intend to study these

topics and also integrate social simulation tools to ve-

hicle simulators (Pereira and Rossetti, 2012) so as to

investigate how autonomous cars can be prepared to

communicate with pedestrians and other road users.

Furthermore, outcomes of research in this emerging

ﬁeld may yield enormous contributions to improving

legislation and trafﬁc regulatory directives in the new

age of autonomous vehicles.

REFERENCES

Alves, P. R. J. A., Gonc¸alves, J., Rossetti, R. J. F., Oliveira,

E. C., and Olaverri-Monreal, C. (2013). Forward col-

lision warning systems using heads-up displays: Test-

ing usability of two new metaphors. In 2013 IEEE

Intelligent Vehicles Symposium (IV), pages 1–6.

Chen, P., Wu, C., and Zhu, S. (2016). Interaction be-

tween vehicles and pedestrians at uncontrolled mid-

block crosswalks. Safety Science, 82:68–76.

Choi, J. K. and Ji, Y. G. (2015). Investigating the impor-

tance of trust on adopting an autonomous vehicle. In-

ternational Journal of HumanComputer Interaction,

31(10):692–702.

Driggs-Campbell, K., Govindarajan, V., and Bajcsy, R.

(2017). Integrating intuitive driver models in au-

tonomous planning for interactive maneuvers. IEEE

Transactions on Intelligent Transportation Systems,

18(12):3461–3472.

Earl, R., Falkmer, T., Girdler, S., Dahlman, J., Rehnberg,

A., and Falkmer, M. (2016). Visual search strategies

of pedestrians with and without visual and cognitive

impairments in a shared zone: A proof of concept

study. Land Use Policy, 57:327–334.

Florentine, E., Ang, M. A., Pendleton, S. D., Andersen,

H., and Ang, Jr., M. H. (2016). Pedestrian notiﬁca-

tion methods in autonomous vehicles for multi-class

mobility-on-demand service. In Proceedings of the

Fourth International Conference on Human Agent In-

teraction, pages 387–392, New York, NY. ACM.

Freedman, M., Zador, P., and Staplin, L. (1993). Effects

of reduced transmittance ﬁlm on automobile rear win-

dow visibility. Human Factors, 35(3):535–550.

Gonc¸alves, J., Rossetti, R. J. F., and Olaverri-Monreal, C.

(2012). Ic-deep: A serious games based application to

On Social Interactions and the Emergence of Autonomous Vehicles

429

assess the ergonomics of in-vehicle information sys-

tems. In 2012 15th International IEEE Conference on

Intelligent Transportation Systems, pages 1809–1814.

Gonc¸alves, J. S. V., Rossetti, R. J. F., Jacob, J., Gonc¸alves,

J., Olaverri-Monreal, C., Coelho, A., and Rodrigues,

R. (2014). Testing advanced driver assistance sys-

tems with a serious-game-based human factors anal-

ysis suite. In 2014 IEEE Intelligent Vehicles Sympo-

sium Proceedings, pages 13–18.

Gupta, S., Vasardani, M., and Winter, S. (2016). Con-

ventionalized gestures for the interaction of people in

trafﬁc with autonomous vehicles. In Proceedings of

the 9th ACM SIGSPATIAL International Workshop on

Computational Transportation Science, IWCTS ’16,

pages 55–60, New York, NY. ACM.

Hospedales, T., Gong, S., and Xiang, T. (2012). Video be-

haviour mining using a dynamic topic model. Inter-

national Journal of Computer Vision, 98(3):303–323.

Koo, J., Kwac, J., Ju, W., Steinert, M., Leifer, L., and

Nass, C. (2015). Why did my car just do that? ex-

plaining semi-autonomous driving actions to improve

driver understanding, trust, and performance. Interna-

tional Journal on Interactive Design and Manufactur-

ing (IJIDeM), 9(4):269–275.

Laznyi, K. and Marczi, G. (2017). Dispositional trust

x2014; do we trust autonomous cars? In 2017 IEEE

15th International Symposium on Intelligent Systems

and Informatics (SISY), pages 000135–000140.

Lehsing, C., Benz, T., and Bengler, K. (2016a). Insights

into interaction - effects of human-human interaction

in pedestrian crossing situations using a linked simu-

lator environment. IFAC-PapersOnLine, 49(19):138–

143.

Lehsing, C., Fleischer, M., and Bengler, K. (2016b). On the

track of social interaction - a non-linear quantiﬁcation

approach in trafﬁc conﬂict research. In 2016 IEEE

19th International Conference on Intelligent Trans-

portation Systems (ITSC), pages 2046–2051.

Lundgren, V., Habibovic, A., Andersson, J., Lagstrm, T.,

Nilsson, M., Sirkka, A., Fagerlnn, J., Fredriksson, R.,

Edgren, C., Krupenia, S., and Salur, D. (2017). Will

there be new communication needs when introducing

automated vehicles to the urban context? Advances in

Intelligent Systems and Computing, 484:485–497.

Macedo, J., Kokkinogenis, Z., Soares, G., Perrotta, D., and

Rossetti, R. J. F. (2013). A hla-based multi-resolution

approach to simulating electric vehicles in simulink

and sumo. In 16th Int. IEEE Conf. on Intelligent

Transportation Systems (ITSC), pages 2367–2372.

Morton, J., Wheeler, T. A., and Kochenderfer, M. J. (2017).

Analysis of recurrent neural networks for probabilistic

modeling of driver behavior. IEEE Transactions on

Intelligent Transportation Systems, 18(5):1289–1298.

Munigety, C. R. and Mathew, T. V. (2016). Towards behav-

ioral modeling of drivers in mixed trafﬁc conditions.

Transportation in Developing Economies, 2(1):6.

Nagahama, A., Yanagisawa, D., and Nishinari, K. (2017).

Dependence of driving characteristics upon follower-

leader combination. Physica A: Statistical Mechanics

and its Applications, 483:503 – 516.

NHTSA (2012). Trafﬁc safety facts 2010: A compila-

tion of motor vehicle crash data from the fatality

analysis reporting system and the general estimates

system. Technical report, National Highway Trafﬁc

Safety Administration, Washington, DC.

Oh, C., Choi, J., and Park, S. (2017). In-depth understand-

ing of lane changing interactions for in-vehicle driving

assistance systems. International Journal of Automo-

tive Technology, 18(2):357–363.

Pereira, J. L. F. and Rossetti, R. J. F. (2012). An integrated

architecture for autonomous vehicles simulation. In

Proceedings of the 27th Annual ACM Symposium on

Applied Computing, SAC ’12, pages 286–292, New

York, NY, USA. ACM.

Perrotta, D., Macedo, J. L., Rossetti, R. J., de Sousa,

J. F., Kokkinogenis, Z., Ribeiro, B., and Afonso, J. L.

(2014). Route planning for electric buses: A case

study in oporto. Procedia - Social and Behavioral Sci-

ences, 111:1004 – 1014. Transportation: Can we do

more with less resources? 16th Meeting of the Euro

Working Group on Transportation Porto 2013.

Perrotta, D., Ribeiro, B., Rossetti, R. J., and Afonso, J. L.

(2012). On the potential of regenerative braking of

electric buses as a function of their itinerary. Procedia

- Social and Behavioral Sciences, 54:1156 – 1167.

Rossetti, R. J. F., Almeida, J. E., Kokkinogenis, Z., and

Gonc¸alves, J. (2013). Playing transportation seri-

ously: Applications of serious games to artiﬁcial

transportation systems. IEEE Intelligent Systems,

28(4):107–112.

Rothenbucher, D., Li, J., Sirkin, D., Mok, B., and Ju,

W. (2016). Ghost driver: A ﬁeld study investigat-

ing the interaction between pedestrians and driver-

less vehicles. 25th IEEE International Symposium on

Robot and Human Interactive Communication, RO-

MAN 2016, pages 795–802.

Sucha, M., Dostal, D., and Risser, R. b. (2017). Pedestrian-

driver communication and decision strategies at

marked crossings. Accident Analysis and Prevention,

102:41–50.

United Kingdom Government. Tinted vehicle windows: the

law. https://www.gov.uk/tinted-vehicle-window-rules.

Last accessed: 2018-02-06.

Verberne, F. M. F., Ham, J., and Midden, C. J. H. (2012).

Trust in smart systems: Sharing driving goals and

giving information to increase trustworthiness and ac-

ceptability of smart systems in cars. Human Factors,

54(5):799–810. PMID: 23156624.

Xiaorui, W. and Hongxu, Y. (2013). A lane change model

with the consideration of car following behavior. Pro-

cedia Soc Behav Sci, 96:2354–2361.

VEHITS 2018 - 4th International Conference on Vehicle Technology and Intelligent Transport Systems

430