Semi-autonomous advanced parking assistants: Do they really have to be learned if steering is automated?

Abstract

Several studies have demonstrated positive effects of advanced parking assistants (APA) on driver comfort and parking performance. However, learning effects while handling the APA system and possible transfer effects on manual parking have not yet been discussed. In this study, N= 18 subjects parked parallel in a test area (26 manoeuvres) and in real traffic (nine manoeuvres). One half of the manoeuvres was done without a parking assistant, one half with a semi-autonomous APA system that utilises automatic steering. The APA system did not control speed by accelerating or braking. Parking performance and glance behaviour in selected manoeuvres were analysed as well as drivers' judgements and observations by an in-vehicle experimenter. Consistent with earlier studies, the APA system facilitates parking. Learning effects particularly appear in glance behaviour and maximum velocity during the first parking motion as also do the number of errors while handling the system. Using the APA system repeatedly might influence parking without an assistant as well: the more manoeuvres are carried out with the APA system, the more often the drivers look into the display during manual parking. The implications of this study are discussed.

Full text

Effects of ITS on drivers’ behaviour and interaction with the systems
123
SEMI-AUTONOMOUS ADVANCED PARKING
ASSISTS: DO THEY REALLY HAVE TO BE LEARNED
IF STEERING IS AUTOMATED?
Ingo Totzke, Susanne Jessberger, Dominik Mühlbacher, Hans-Peter Krüger
Center for Traffic Sciences (IZVW), University of Würzburg
Röntgenring 11, 97070 Würzburg, Germany
totzke@psychologie.uni-wuerzburg.de
ABSTRACT: A few studies have demonstrated positive effects of Advanced
Parking As-sists (APA) on driver comfort and parking performance. Learning
effects while handling the APA system and possible transfer effects on
manual parking have not been discussed yet. In this study, N = 18 subjects
parked parallel in a test area (26 manoeuvres) and in real traffic (9
manoeuvres). One half of the manoeuvres was done without the parking
assist, one half with a semi-autonomous APA system which utilized automatic
steering. The APA system did not control speed by accelerating or braking.
Consistent with earlier studies, the APA system facilitates parking. Learning
effects particularly ap-pear in glance behaviour and maximum velocity during
the first parking motion. Using the APA over a large number of manoeuvres
might influence parking without assistant: The more manoeuvres are done
with the APA, the more often the drivers look into the display during manual
parking.
1 INTRODUCTION
In recent years, an increased research and development activity in the field of
Advanced Parking Assists (APA) could be seen. The range of APA systems
varies from information systems (e.g. solely distance control or parking space
measuring) to full-autonomous parking assists (automated steering and speed
control; for an overview see [1]). Some of those sys-tems have already been
introduced to the market, especially in the area of distance control. Nowadays,
mainly ultrasonic systems inform the driver via acoustic sounds and/or visual
displays about the distance to objects or to cars which restrict the parking
space. At present, semi-autonomous APA systems for parallel parking
represent the most advanced type of parking assist systems on the market.
While steering is controlled by the system, the driver receives via displays a
step-by-step instruction on how to move the vehicle (accelerating and braking).
Up to now, efficiency of APA systems has been analyzed empirically, for
instance, respecting reduced operational demands for the driver during parallel
parking or regarding an improved parking performance (e.g. [2] – [4]). However,
learning effects while handling the APA sys-tem and possible transfer effects on
manual parking have not been discussed so far. Accord-ing to the so-called
“power law of practice” [5], it is expected that learning effects in handling the
APA system particularly take place during the first manoeuvres while parking:
The more often the driver parks with the APA system, the lower possible gains
in learning are. Whereas the “power law of practice” has been demonstrated for
learning processes concern-ing information systems in the vehicle (e.g. [6], [7]),
no comparable studies have been pub-lished for Advanced Driving Assistant
Systems (ADAS) yet. The present study’s aim was to examine these learning

Human Centred Design for Intelligent Transport Systems
124
effects while using a semi-autonomous APA system. It was assumed that
possible learning effects particularly prevail during the first manoeuvres with the
APA system.
2 METHOD
2.1 Semi-autonomous APA system
The APA system used in this study first supported the driver in finding a parking
space by showing free spaces in a display placed on the upper central console.
After stop-ping the car and changing into reverse, the APA system controlled
the steering. The driver manoeuvred the car by using the systems instructions
on the display regarding accelerating and breaking, while the system controlled
the steering while parallel parking autonomously. The manoeuvre ended as
soon as the pre-calculated parking position was reached. The systems’ support
continued when more forward and backward motions were necessary to reach
the pre-calculated parking position. The driver could monitor the whole
procedure on the display in the upper central console.
In addition, the car was fitted with a distance control system to check the
distance to any neighbouring objects using ultrasonic sound (UPA, Ultrasonic
Park Assist). Approximation to any object was indicated by an increasing signal
until a constant alert indicated the minor distance of less than 30 cm to the other
object.
2.2 Parking manoeuvres
The subjects had the task to park the car in parking spaces on a test area with
and without the described semi-autonomous APA system, respectively. The
same parking manoeuvres had to be done in real traffic. In the following, the
results for the test area will only be consid-ered.
The first part of the study was done on a test area in an industrial area of
Würzburg. A con-natural traffic situation was created (see Fig.1). Some cars
formed six parking spaces on both sides of a lane, 5.5 m wide, approachable
from both sides. The spaces were 2 m wide. One of the spaces was limited by a
displaced car (see Fig.1 down right). In addition, road marking was added for a
better identification of the lane and wooden beams were used as curb stones to
mark the parking spaces.

Effects of ITS on drivers’ behaviour and interaction with the systems
125
Fig.1. Schematic illustration of the study’s design in the test area.
In allusion to Lee (2006), two different space sizes had been created:
• 140 % of the car’s length (as the minimal space size, in which the APA
system relia-bly identified a suiting parking space), or
• 160% of the car’s length (as the minimal space size, in which the APA
system could realize a one backward-motion parking position), respectively.
The length of the 140%-space was approx. 6.78 m, the one of the 160%-space
approx. 7.74 m (the length of the test vehicle was 4.84 m). Parking into the
160%-space was judged as easier by the drivers compared to the 140%-space
(“How strenuous was parking”, m = 7.86, sd = 1.55 for 140%-space; m = 6.74,
sd = 1.79 for 160%-space on a 16-point scale ranging from 0 = “not at all” to 15
= “very much”, t-test for dependant samples: t(53) = -7.724, p = .000). In the
following, results for selected manoeuvres in the 140%-spaces will be
discussed only: four manoeuvres without APA (two manoeuvres of the training
stage and one manoeu-vre of block 1 and block 2 each; see chapter
“procedure”) and three manoeuvres with APA (allocated to the whole part in the
test area).
In some of the manoeuvres in the test area, staged situations were realized in
which obsta-cles were positioned in the vehicle’s pathway (e.g. a tethered toy
coupe was pulled into way of the car while backing up; a post was placed within
the parking space by an on-road ex-perimenter). These obstacles should stand
for passers-by or any not detected objects (for further information and results
see [4]).
2.3 Glance behaviour analysis
The glance behaviour of subjects was measured by means of the head-
mounted measure-ment device “Dikablis”. In the head unit of that device two
cameras have been installed, the one directed on the environment, the other on
the left eye of the subject. Through superim-position of both camera pictures the
glance behaviour of the subjects could be specified.
The superimposed videos have been coded manually. For this purpose five
areas of interest (AOI) were defined:

Human Centred Design for Intelligent Transport Systems
126
1. front windshield
2. mirrors
3. front side windows
4. display
5. rear windows
For each AOI the following parameters were calculated:
 duration of glances (mean, median, maximum)
 glance direction („no glance into AOI“ vs. „minimum one glance into
AOI“)
 relative frequency of AOI during the whole parking manoeuvre
 In the following, the glance behaviour of the driver during the first
backward motion will be discussed exclusively. This is because the
driver leaves the flowing traffic and might get in contact with obstacles
in the parking space. Therefore, particularly the first backward motion
addresses safety-relevant aspects while parking. The first backward
motion starts with a stop after passing a free parking space and ends
with a halt at the end of the first backward mo-tion.
2.4 Procedure
The experimental session included the following stages:
(1) training stage in a test area (8 manoeuvres)
(2) block 1 and block 2 in a test area (9 manoeuvres each)
(3) public traffic (9 manoeuvres)
The session started in the test area with the installation of the glance behaviour
measuring device. The subject was not given a detailed explanation of the
functionality of the APA sys-tem. Instead, there was only a short instruction
designed to be similar to a “car rental situa-tion”. Thereafter, the subject gained
experience in handling the car and the APA system by performing 4 parking
manoeuvres without as well as with the APA system (“training stage”). During
these manoeuvres the subject had to acquaint her-/himself with realizing the
system’s functionality. After the training stage, detailed system instructions were
given to the driver.
In the main part of the session (“block 1” and “block 2”), the driver had to
perform 18 ma-noeuvres (9 manoeuvres with and also 9 manoeuvres without
the semi-autonomous APA system) in the test area. After 9 manoeuvres each,
the subjects took a break during which they could put off the glance behaviour
measuring device.
In public traffic, the driver had to perform 9 parking manoeuvres, nearly half of
the manoeu-vres with and without the APA system each.
After each manoeuvre, the subjects were asked for their judgements concerning
workload, parking performance and (if applicable) problems with handling the
vehicle or the APA sys-tem. After each stage of the session, the subjects had to
answer a questionnaire and the in-vehicle experimenter gave judgements about
the driver’s performance in the preceding stage (e.g. concerning appropriate
reaction and safe reaction on the APA system). Additionally, there was a
detailed survey done by the experimenter at the end of the session.

Effects of ITS on drivers’ behaviour and interaction with the systems
127
The study was conducted by an in-vehicle experimenter (who sat behind the
driver and con-ducted the inquiries, counted driver’s errors etc.) and an on-road
experimenter (who realized the staged situations, counted driver’s errors etc.).
Several Can BUS data were recorded (e.g. velocity). Altogether, each subject
performed 35 parking manoeuvres which is a high number compared to
naturalistic driving. The whole session lasted approximately 4 hours.
2.5 Sample
Altogether N = 18 subjects (9 male and 9 female) between 19 and 72 years of
age (m = 40.3, sd = 21.9) participated in the study. The subjects were members
of the test driver panel of the Würzburg Institute for Traffic Sciences (WIVW).
The groups consisted of drivers, of who one half estimated their capabilities in
parallel parking as “good” and the other half respec-tively as “poor”. On the
other hand one third of the sample represented a younger age (19-20 years),
another third a middle age (25-49 years) and the last third an older age group
(65-72 years). The needed information was ascertained by online
questionnaires prior to the test. The subjects were granted an expense
allowance for their participation.
3 RESULTS
With the APA system, nearly all parking attempts are successful and the
number of parking motions is smaller than with parking without APA (see Fig.2).
These results reflect the func-tionality of the APA system. Therefore, no
learning effects for handling the APA system can be seen (number of parking
attempts with APA: inferential statistics cannot be applied as no variation is
found; number of parking motions wit APA: one-factor ANOVA for within-factor
“manoeuvre”, F(2,32) = 0.399, p = .674, eta2 = .024). Compared with this, the
drivers gain experience in parallel parking itself while parking without APA: The
number of parking at-tempts as well as the number of parking motions (one-
factor ANOVA for within-factor “ma-noeuvre”, F(3,48) = 2.460, p = .074, eta2 =
.133; F(2,51) = 2.918, p = .043, eta2 = .146, re-spectively) diminish with
growing experience in handling the vehicle during manual parking, particular in
direct comparison of the first and second manoeuvre during manual parking
with the test vehicle.

Human Centred Design for Intelligent Transport Systems
128
Fig.2. Mean number of parking attempts (left) and mean number of
parking motions (right).
Pictured is the mean with standard deviation.
The maximum velocity during the first parking motion is reduced with growing
experience in handling the APA system (one-factor ANOVA for within-factor
“manoeuvre”, F(2,34) = 3.454, p = .043, eta2 = .169; see Fig.3). Compared with
this, the distribution of velocity largely re-mains constant. As a result, there is no
learning effect on the duration of the whole parking manoeuvre.
Fig.3. Mean maximum velocity [in km/h] during the first parking motion.
Pictured is the mean with standard deviation.
According to the observations of the in-vehicle experimenter after each stage of
the experi-mental session, the drivers react more pertinently and safely to the
APA system with growing practice (one-factor ANOVAs for within-factor
“manoeuvre”; appropriate reaction: F(2,34) = 9.677, p = .000, eta2 = .363; safe
reaction: F(2,34) = 23.415, p = .000, eta2 = .579; see Fig.4). These learning
gains appear between the training stage and block 1, in particular.

Effects of ITS on drivers’ behaviour and interaction with the systems
129
Fig.4. Mean judgements of in-vehicle experimenter concerning
appropriate reaction (left) and safe reaction (right) on the APA system in
the preceding stage of the session Pictured is the mean with standard
deviation.
Additionally, the drivers were asked immediately after a parking manoeuvre to
give judge-ments on a 15-point scale concerning perceived workload and safety
while parking (see Fig.5). In the course of the session drivers’ answers on the
question “How strenuous was parking” (perceived workload, see Fig.5 on the
left) do not change neither for manual parking (one-factor ANOVA for within-
factor “manoeuvre”, F(3,51) = 1.823, p = .155, eta2 = .097) nor for assisted
parking with APA (one-factor ANOVA for within-factor “manoeuvre”, F(2,32) =
0.775, p = .469, eta2 = .046). Overall, reported workload of the driver appears
to be of me-dium height.
Similarly, drivers’ judgements concerning “How confident did you feel during
parking?” (per-ceived safety, see Fig.5 on the right) do not vary significantly
over time while parking with APA (one-factor ANOVA for within-factor
“manoeuvre”, F(2,32) = 1.389, p = .264, eta2 = .080): The drivers feel largely
confident during parking with APA. However, learning effects apply for manual
parking concerning reported safety (one-factor ANOVA for within-factor
“manoeuvre”, F(3,51) = 3.318, p = .027, eta2 = .163): Especially in the first non-
assisted parking manoeuvre drivers feel less safe compared to the following
parking manoeuvres.
Fig.5. Drivers answers on the question” How strenuous was parking”
(left) and “How confident did you feel during parking” (right). Pictured is
the mean with standard deviation.

Human Centred Design for Intelligent Transport Systems
130
Learning effects are visible in the glance behaviour during the first parking
motion with APA (see Fig.6): The more experience the driver has with handling
the APA system, the less vis-ual attention is used to monitor the system’s
display. For other “areas-of-interest” (e.g. wind-shield, windows, backwards,
mirrors) no systematic learning effects can be proven. For manual parking, no
systematic changes over the course of the session can be shown except for
looking backwards.
Fig.6. Relative frequency of glances [in percentage] into AOI during the
first parking motion of manoeuvres with and without the APA system,
respectively.
Surprisingly, with growing practice more drivers look at least once into the
system’s display while parking without APA, even though this display is
deactivated while parking without APA (seeFig.7). During assisted parking,
nearly all subjects look at least once into the display during the first parking
motion. In this case, no learning effect can be shown: Using the APA system
needs at least one glance into the system’s display.
Fig.7. Relative frequency of drivers who look at least once into the
system’s display [in percentage] during the first parking motion with and
without the APA system, respectively

Effects of ITS on drivers’ behaviour and interaction with the systems
131
4 CONCLUSIONS
To sum up, the lower numbers of parking attempts and parking motions as well
as the positive judgements of the drivers about the APA system indicate that the
usage of the parking assistant facilitates parking. These results are consistent
with published studies which have shown positive effects of APA systems
respecting operational demands for the driver during parallel parking or
regarding an improved parking performance (e.g. [2] – [4]). Nevertheless, the
expansion of automation (for instance the launch of full-autonomous APA
systems) should be handled with care: Higher automation levels of APA
systems might go along with a decrease of the driver’s attention to the
environment, as, for instance, the increased probability of dangerous traffic
situations [4]. These dangerous situations might mainly occur when the system
reaches its limits (e.g. moving obstacles like passers-by or objects hard to
detect due to size or design like posts or bicycles).
Learning effects appear in particular in glance behaviour and maximum velocity
during the first parking motion as well as in drivers’ reaction to the APA system:
The more experience the driver has with handling the system, the less visual
attention is used to monitor the system’s display, the lower is maximum velocity
and the drivers react more appropriately and safely to the APA system. No
systematic learning effects can be found in drivers’ judgements concerning their
perceived workload and perceived safety. Moreover, the results largely con-firm
the assumptions made by the “power law of practice” [5]: If learning effects
appear, these are dominant in the first part of the session. Therefore, this law
can be applied to learning effects while handling an APA system (as an
example of an ADAS).
However, the learning effects in handling the APA system might even influence
the glance behaviour during parallel parking without a parking assistant: The
more manoeuvres are done with the APA system, the more often the drivers
look at least once into the display while parking without the APA system, even
though this display is deactivated. Some drivers seem to have introduced the
system’s display into their gaze pattern during parallel parking even in manual
parking. It is questionable whether this carry-over effect will hold on after a
larger number of manoeuvres without an APA system. Nevertheless,
unintended carry-over effects respecting glance behaviour have to be
considered in further studies.
5 REFERENCES
[1] Lambert, G., Kirchner, A., and Hüger, P.: ‘Parkassistenzsysteme.
Technologien von heute und morgen’, in VDI-Gesellschaft Fahrzeug- und
Verkehrstechnik (Ed.): ‘Integrierte Sicherheit und Fahrerassistenzsysteme‘
(VDI-Verlag, 2008, VDI-Berichte Nr. 2048)
[2] Doisl, C.: ’Systemergonomische Analyse von Anzeige- und
Bedienkonzepten zur Unter-stützung des Parkvorgangs’. PhD thesis, TU
München, 2007
[3] Lee, W.C.: ‘Beiträge zur Entwicklung eines Fahrerassistenz-Systems für
Einpark-vorgänge‘. PhD thesis, TU Illmenau, 2006

Human Centred Design for Intelligent Transport Systems
132
[4] Totzke, I., Mühlbacher, D., and Krüger, H.-P.: ‘Semi-autonomous
Advanced Parking As-sist – a source of drivers’ distraction?’, in D. de
Waard, A. Axelsson, M. Berglund, B. Peters, and C. Weikert (Eds.):
‘Human Factors: A system view of human, technology and organisation)
(Shaker Publishing, 2010), pp. 1-10
[5] Newell, A., and Rosenbloom, A.: ‘Mechanisms of skill acquisition and the
Law of Practice’, in Anderson, J.R. (Ed.): ‘Cognitive Skills and their
Acquisition’ (Erlbaum, 1981), pp. 1-55
[6] Jahn, G., Oehme, A., Rösler, D., and Krems, J. F.: ‘Kompetenzerwerb im
Umgang mit Fahrerinformationssystemen‘, in Bundesanstalt für
Straßenwesen (Ed.): ‘Berichte der Bundesanstalt für Straßenwesen‘
(Wirtschaftsverlag NW, 2004, Fahrzeugtechnik Heft F 47)
[7] Totzke, I., Krüger, H.-P., Hofmann, M., Meilinger, T., Rauch, N., and
Schmidt, G.: ‘Kom-petenzerwerb für Informationssysteme - Einfluss des
Lernprozesses auf die Interaktion mit Fahrerinformationssystemen‘
(Berthold Druck, 2004, FAT-Schriftenreihe Band 184)