© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
DOI: 10.1109/ITSC.2011.6082924
Probabilistic Driving Style Determination by means of a Situation
Based Analysis of the Vehicle Data
Tobias Bär, Dennis Nienhüser, Ralf Kohlhaas and J. Marius Zöllner
Intelligent Systems and Production Engineering (ISPE),
FZI Forschungszentrum Informatik
Haid-und-Neu-Str. 10-14, 76131 Karlsruhe, Germany
Email: {baer, nienhueser, kohlhaas, zoellner}@fzi.de
Abstract— Today, driver assistance systems assist the driver
in manifold ways. Their acceptance and usefulness can be highly
increased by adapting them to the needs and the personality
of the driver.
In this work the driving style of the driver is determined by
means of rating the drivers actions in commonly occurring
traffic situations. Therefore, the vehicle data is evaluated and a
probabilistic affiliation of the driver being aggressive, anxious,
economical, keen, or sedate is made. The situations are chosen
to be day-to-day traffic situations, for instance approaching a
village, stopping on a stop sign, or passing through a tight bend.
Based on the determined driving style, future driving assistance
systems can be personalized to the individual driver and, thus,
get more valuable. As a showcase, we adjust our Anticipatory
Energy Saving Assistant (ANESA) to the drivers character,
which is giving driving hints how to save energy in tight curves.
By personalizing ANESA more credence is gained, resulting in
extra savings of energy as we show in the experiments made.
to drive curves with much higher speed than proposed. As
a result, they either simply ignored the advice, or took the
advice first, but accelerated ahead of the curve nevertheless
(see red line in figure 1). Neither of the behaviors is desired.
On the other hand, if tight curves are annotated with a
velocity too fast for the drivers ability, our experiments
showed that the drivers could not tap the full energy saving
potential since they braked ahead of the curve nevertheless
(see blue line in figure 1).
In figure 1 the effects of curves being annotated too fast
or too slow are visualized. The figure shows, that only if
the curve is annotated with an advisory speed individually
tailored to the driver, the energy saving potential is fully
utilized.
In this paper, the driving style of the driver is determined.
I. INTRODUCTION
1 ANESA is a result of a long term cooperation between Harman/Becker
Automotive Systems and the FZI.
Approaching a Tight Bend with Speed Hint Unadjusted
to the Driver - Aggressive vs. Anxious Driver
Velocity - Aggressive Driver
Energy Consumption - Aggressive Driver
Velocity - Anxious Driver
Energy Consumption - Anxious Driver
30
25
20
20
Aggressive Driver gets freewheeling advice
too early for his abilities
15
10
Aggressive Driver decides to accelerate, since he
is used to drive the curve faster then advised.
This, requires additional energy.
Anxious Driver brakes,
because he wants to
drive the upcoming curve slower than advised.
Energy is lost
10
5
0
700
30
25
15
5
35
Energy Consumption [ L/km ]
35
Velocity [ m/s ]
Driver Assistance Systems are an inherent part of todays
vehicle equipment. Their performance, as well as their
usability and acceptance by drivers highly depends on the
needs and the skills of the driver. First and foremost, driver
assistance systems must not confuse or distract the driver in
any way. In critical, dangerous situations for instance, the
driver does not care about energy saving advices, whereas
driving in safe and uncritical circumstances, the driver is
receptive to save fuel.
Besides the safety aspect, driver assistance systems become
more valuable if they are tailored to the personality and the
driving behavior of the driver. With ANESA1 our research
group developed an Anticipatory Energy Saving Assistance
System, helping the driver to approach upcoming velocity
restrictions freewheeling - hence, the most economic way.
ANESA also regards tight curves as velocity restrictions,
since the driver can avoid loosing energy by letting the car
freewheel ahead of the curve [1].
Test showed, that ANESA is more valuable if the advisory
speed, of which tight curves are annotated with, is individually adjusted to the personality and the driving style of the
driver. In experiments, a freewheeling advice with a low
velocity was given to aggressive drivers, which were used
0
800
900
1000
1100
1200
Road Position [ m ]
1300
1400
1500
Fig. 1. If the advised speed in tight bends is not individually adjusted
to the character of the driver, energy is lost due to unnecessary braking
(anxious drivers) or unnecessary accelerations (aggressive drivers).
The probability of the driver being aggressive, anxious,
economical, keen, or sedate is rated by means of a situational
analysis of the vehicle parameters.
Having a probabilistic description of the driving style, advanced driver assistance systems can be adjusted to the
personality of the driver, as exemplary showcased on our
Anticipatory Energy Saving Assistant (ANESA).
© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
DOI: 10.1109/ITSC.2011.6082924
II. RELATED WORK AND THE QUALIFYING OF
THE RESEARCH CONTEXT
In section II-A this work is integrated into the context of
modeling and understanding driver behavior. An overview of
related work, dealing with classification of driving behavior
or determining the driving style of the driver is given in II-B.
A. QUALIFYING THE RESEARCH CONTEXT
J.A. Michon proposed a two dimensional classification of
driving behavior models (see table I) in 1985 already [2].
Behavioral
Internal state (psychological)
Taxonomic
Task analysis
Trait models
Functional
Adaptive control models
Motivational models
TABLE I
C LASSIFICATION OF DRIVER BEHAVIOR MODELS ( ADAPTED FROM [2])
Task analysis models decompose driving into tasks and
subtasks and relate them to driver requirements and abilities. Trait models attempt to capture individual personality
characteristics, e.g. the aggressiveness or the cautiousness of
the driver, the influence of stress, or how much the driver is
prone to distraction.
Adaptive control models apply functional approach to capture behavioral changes. Motivational control models consider internal states as attitudes, subjective risk and insight
as controlling factors.
The driver profile proposed in this paper can be dedicated to
the trait models, since it attempts to estimate the probability
of the driver belonging to a certain driving style class with
the idea to have a general, probabilistic description of the
driving behavior.
Having a general description of the driving style, further
information, like the velocity the driver will choose to drive
a tight curve, can be inferred.
B. RELATED WORK
Farida Saad defines driving style as a relatively stable
description of the driver, which typifies their personal way
of driving [3]. We assent to this definition.
Taubman-Ben-Ari et. al classify drivers in eight types: [dissociative], [anxious], [risky], [angry], [high-velocity], [distressreduction], [patient], and [careful] [4]. Their aim was mainly
to find statistic relationships. For example, they found that
drivers with a higher education often developed a more
anxious driving behavior. Their data was collected by means
of a 44-item questionnaire named Multidimensional Driving
Style Inventory (MDSI).
Y. Chung et. al defined four driving styles: [reckless and
careless], [anxious], [angry and hostile], and [patient and
careful] [5]. They tried to find a connection between the
socio-demographic background of the drivers to their driving
style by means of machine learning approaches (K-means
cluster and logistic regression analysis). Their data was
gained through written a questionnaire as well.
Several attempts have been made to personalize adaptive
cruise control (e.g. [6], [7]) or collision warning systems
(e.g. [8]) to the consumers needs. Our objective is a general
probabilistic description of the driving style rather than
customized solutions for one application.
For our work, the definitions of the driving styles by
Taubman-Ben-Ari and Chung served as a model to our
defined driving styles. Contrary to these workings our determination of the driving style is based on vehicle data
measurements and is aimed to be generated whilst driving.
Furthermore, we aim on adjusting and improving driver
assistance systems, rather than to collecting links between
driving style and the social background of the driver.
III. OVERVIEW
The purpose of this work is to determine the driving style
of the driver. For this reason, five types of driving styles
were defined. These are aggressive, anxious, keen, economic,
sedate (roughly comparable to [4] and [5]). Section IV-A
gives a more detailed description of the driving style classes
and our interpretation of affinity.
The rating of the drivers affiliation to each particular driving
style class is based on their actions in five well-defined
traffic scenarios. The scenarios are chosen to be commonly
occurring, day-to-day traffic scenes. Section IV-B explains
the defined scenarios in more detail and describes their implementation in the driving simulator software. Furthermore,
the recorded vehicle data characterizing the driver in the
particular scene is explained.
To evaluate the actions of the drivers and determine the
affinity to the different driving style classes, fuzzy-rules are
applied. The rules find a detailed description in section V.
By means of the driving style of the driver, the driverspecific curve advisory velocity is deduced. Amongst other
experiments, section VI compares drives undertaken with
assistance of ANESA with and without a parametrization
to the determined driving style.
IV. IMPLEMENTATION
A. Driving Style Classes
To rate the driving behavior, the driver is categorized into
five driving style types. The driving styles are not mutually
exclusive. For instance an economical driver can also be
somehow sedate or keen. Our interpretation of the driving
style is as follows:
Aggressive:
Aggressive drivers are characterized by driving at
high speeds, high degrees of accelerations and
decelerations, and generally a high risk taking
behavior. Furthermore, they tend to drive too close
to other road users and put other traffic participants
at risk.
Anxious:
Anxious, or inexperienced defensive drivers are
taking a very low level of risk. They tend to drive
slower than sign-posted. In tight curves, they do
not dare to drive too fast. Their accelerations are
very low, whereas their decelerations are sometimes
© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
DOI: 10.1109/ITSC.2011.6082924
too intense due to misjudging the vehicle braking
characteristic.
Economical:
Economical drivers are freewheeling to upcoming
traffic restrictions. They try to hold the vehicle on a
constant speed level and avoid unnecessary braking.
However, their accelerations can be quite strong
aiming to reach their desired traveling speed as
efficient as possible.
Keen:
Keen drivers are aware of the vehicle characteristic and utilize the full dynamics of their vehicle.
They are driving at the sign-posted speed limit, or
slightly over it. However, compared to the aggressive drivers they do not threaten other road users
and use their skills more responsible.
Sedate:
Sedate drivers are characterized as driving responsible and moderate. They obey the traffic rules and
tend to drive at the sign-posted velocity (or slightly
slower). Their interaction with other road users is
calm, polite, and defensive.
B. The Defined Traffic Situations and their Contribution to
the Driving Style
The driving style of the driver is calculated by means of
evaluating the following scenarios:
Approaching a Village: The driver is approaching a
village on a country road. The velocity restriction on the
country road is 100 km/h and the velocity restriction within
the village is 50 km/h. 200 m in front of the village a
with a high speed, it is a good indicator for an aggressive
or keen driving style. Lower speeds are typical for sedate
or anxious drivers.
Start of freewheeling: Economical drivers start the
freewheeling phase very early, whereas aggressive drivers
usually do not freewheel at all and tend to lower their
velocity by braking. Anxious drivers also tend to brake
rather than freewheel. They are usually not aware of the
vehicle dynamics.
Average brake pressure: Economical and experienced sedate
drivers manage to drive the village scenario without braking.
They are aware of the freewheeling-characteristic of the
car and have a good feeling when to stop accelerating
to reach the velocity signs with the sign-posted velocity.
Though keen and aggressive drivers are often aware of
the freewheeling-characteristic, they still drive the village
scene with a heavy use of the brake, resulting in a faster
travel-time. Anxious drivers also tend to brake in front of
the village. However, this is caused by their inexperience
rather than by the intention of faster travel.
Driving the Country Road: This measurement is a
10 m velocity snapshot taken on a country road without any
velocity restriction sign-posted. According to the German
traffic regulations, the allowed speed is 100 km/h. In this
scenario, the velocity is the sole indicator to inference the
driving style.
Velocity: Aggressive or keen drivers tend to drive faster than
usually allowed, whereas anxious and sedate drivers tend to
drive as advised or slightly slower. The intention of driving
economically can hardly be determined by means of this
measurement.
Driving within a Village: This measurement is a 10 m
velocity snapshot taken within a village at an allowed
speed of 50 km/h. In this scenario the same assumptions
and rules than in ’Driving the Country Road’ can be applied.
Fig. 2. Village Scenario: The car is driving on a country road and is
approaching a village. 200 m in front of the village a 70 km/h sign is posted.
The vehicle data is recorded from 800 m in front of the village to 50 m after
the village sign (blue area).
70 km/h sign is posted, which is very common in German
traffic. The measurement starts 800 m ahead of the village
and ends 50 m after entering the village (see figure 2).
The following data are extracted from the village scenario
to function as input for the fuzzy-rules:
Approaching speed: If the driver is approaching the village
Driving a Tight Bend: This measurement is taken while
the driver is approaching a tight curve on a country road.
The curve is modeled with 45 degrees at a radius of 100
meters. Data is recorded 300 meters ahead and 100 meters
after the curve. The following data are extracted and taken
into account for the fuzzy-rules:
Velocity at the angular point of the curve: While anxious or
sedate drivers usually tend to choose a lower speed in the
curve, aggressive and keen drivers choose a higher velocity
to travel the bend.
Average acceleration after the curve: Economical drivers
usually drive the whole curve scenario with an almost
constant speed. Therefore, their accelerations after the
curve, as well as their decelerations ahead of the curve
are not very strong. Measurements of aggressive or keen
drivers have huge accelerations after the angular point of
the curve. Measurements of anxious drivers have the lowest
accelerations after the curve.
Average deceleration ahead of the curve: Ahead of the curve,
© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
DOI: 10.1109/ITSC.2011.6082924
Stop Sign within a Village: In this measurement the
drivers behavior approaching and leaving a stop sign is
rated. The evaluated data are:
Minimum velocity: According to the law, the driver has to
fully stop (v = 0 km/h) in front of the stop sign. According
to our measurements, aggressive drivers do not always slow
down their car to zero.
Freewheeling distance ahead of the stop sign: As for
the ’Approaching a Village’ or ’Driving a Tight Bend’
scenarios, the same observing of the freewheeling behavior
on stop signs can be made.
Average acceleration after the stop sign: Likewise to the
tight curve scenario, keen and aggressive drivers tend to
accelerate a lot leaving the stop sign. Sedate drivers have
moderate degrees of accelerations whereas anxious drivers
have low acceleration values.
Average deceleration ahead of the stop sign: As for the
tight curve scenario, keen and aggressive drivers tend to
accelerate a lot leaving the stop sign, whereas sedate drivers
have moderate degrees of accelerations. Anxious drivers
have low acceleration values.
Number of Start Attempts: Anxious drivers often need more
than one attempt to pass a stop sign. They tend to look left
and right very carefully and leave the sign very hesitant.
V. FUZZY-RULE BASED DETERMINATION OF
THE DRIVING STYLE
With the observations and assumptions made in IV-B, 112
fuzzy-rules were set up to characterize the driver. In table II
the rules are listed for the situations ’Driving the Village’
and ’Driving the Country Road’ respectively. In table III
the fuzzy-rules for the village scenario are defined. Figure 3
shows an abridgment of measured speeds for the ’Driving the
Village’ scenario and the according fuzzy interval defined.
SPEED
BELOW
AS ADVISED
HIGH
TOO HIGH
AGGRESSIVE
NL
L (likely)
VL (very likely)
VL
ANXIOUS
VL
HL (hardly likely)
NL
NL
ECONOMICAL
HL
VL
HL
NL
KEEN
HL
VL
VL
L
SEDATE
L
VL
HL
NL
TABLE II
F UZZY RULES SET UP FOR THE ’D RIVING WITHIN A V ILLAGE ’
SITUATION AND THE ’D RIVING THE C OUNTRY ROAD ’ SITUATION .
7
6
TOO_HIGH
5
4
Velocity [m/s]
economical drivers decelerate their car by freewheeling.
Anxious, keen and aggressive drivers tend to slow down by
braking. Whereas the braking of anxious drivers is caused
by inexperience, the braking of the keen and aggressive
drivers is caused by optimizing the traveling time.
Freewheeling distance ahead of the curve: Economical
drivers start the freewheeling phase very early, whereas
aggressive drivers usually do not freewheel at all and tend
to lower their velocity by braking. Anxious drivers also tend
to brake rather than freewheel. They are usually frightened
or not aware of the vehicle dynamics.
3
HIGH
2
1
0
AS_ADV.
-1
BELOW
-2
Aggressive
Anxious
Economical
Keen
Sedate
0
0.5
1
Fig. 3. Abridgment of measured velocities in the ’Driving the Village’
scenario and the corresponding fuzzy-rule defined for the velocity. The
advised speed (50 km/h) is already subtracted.
SPEED
BELOW
AS ADVISED
HIGH
TOO HIGH
AGGRESSIVE
NL
L (likely)
VL (very likely)
VL
ANXIOUS
VL
HL (hardly likely)
NL
NL
ECONOMICAL
HL
VL
HL
NL
KEEN
HL
VL
VL
L
SEDATE
L
VL
HL
NL
FREEWH DISTACE
VERY SMALL
SMALL
REGULAR
HIGH
VERY HIGH
AGGRESSIVE
VL
VL
HL
HL
NL
ANXIOUS
NL
L
HL
HL
NL
ECONOMICAL
NL
NL
L
VL
VL
KEEN
L
VL
VL
L
HL
SEDATE
HL
L
VL
L
L
BRAKE FORCE
SMALL
REGULAR
HIGH
AGGRESSIVE
NL
L
VL
ANXIOUS
L
L
NL
ECONOMICAL
VL
HL
NL
KEEN
L
L
VL
SEDATE
L
VL
NL
TABLE III
F UZZY RULES SET UP FOR THE ’V ILLAGE S CENARIO ’.
Country Road’ scenario, for instance, is a good indicator
to separate sedate or anxious drivers from aggressive or
keen drivers. However, not much can be inferred about the
economical behavior of the driver by means of this speed
measurement only. The ’Stop Sign’, ’Village Scenario’,
and the ’Tight Bend’ scenario give much information
how economical the behavior of the driver is. Hence, a
weighting factor wds,s was introduced, defining how much
a situation s ∈ Situations contributes to a particular driving
style ds ∈ DrivingStyles.
∀ ds ∈ DrivingStyles :
∑
wds,s = 1
s∈Situations
Including the weighting factor, the driving style of the
whole test run is calculated as
P(ds) =
∑
wds,s · P(ds|s),
s∈Situations
with P(ds|s) being the probability of the driver behaving
with driving style ds in the given situation s. P(ds) is the
probability of the driver driving with driving style ds in the
whole test run.
VI. EXPERIMENTS
As already mentioned, not all situations serve to gain
good information for all driving styles. The ’Driving in the
To test the implemented driving style classification, a test
track of 7 km was modeled. The track contains two villages
© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
DOI: 10.1109/ITSC.2011.6082924
and two tight curves of 45◦ and 100 m radius. Furthermore,
the track contains a stop sign paced in the village. All
together, the track contains 7 measurement points, where
the vehicle data for the driving style estimation is recorded.
Our experienced test drivers managed to drive the curves
with approximately 85 km/h in our driving simulator, which
is shown in figure 4.
Conducting the tests in a simulator has the advantage of
comparable test results, because all participants are driving
exactly the same scene under the same circumstances.
The driving simulator consists of a reconstructed Smart
Fortwo car, with the signals of the steering wheel and
the pedals connected to a driving simulator software.
The simulation software simulates various data as
forces, momentums, and accelerations in real-time.
The determination of the driving style is based on those
simulated values.
In the driving simulator, there is a lack of feedback for
1
Aggr.
Anxious
Econ.
Keen
Sedate
0.8
0.6
0.4
0.2
0
Aggressive
Anxious
Economical
Keen
Sedate
Fig. 5. The affinity results of the implemented algorithm evaluated on our
test database. As expected, keen and aggressive as well as economical and
sedate are high likely to be confused.
the highest probability, 83 % of the measurements could be
classified correctly.
dsML = arg max (P(ds|M))
ds∈DS
The aggressive and keen measurements could be classified
without any mistake. The most confusion was between the
sedate and the economical drives. 14 % of the sedate test
drives were rated as economic. However, as sedate test drives
are not mutually exclusive economical, this confusion is
bearable.
B. Driving Style Rating
Fig. 4.
The driving simulator of which the tests were conducted with.
speed and dangerousness. Nevertheless, since the vehicle
spins out if the driver is driving too fast, the test persons
intent to drive responsible.
A. Evaluating the Measurement Database
Our database consists of test drives, made by students, staff
members, and visitors at the age of 20 to 55 with varying
degrees of experience. The taken measurements were labeled
as aggressive, anxious, economical, keen, or sedate. There
is a thin line between keen and aggressive measurements.
However, we labeled the measurements as keen, if the driver
was driving fast but nevertheless responsible. If the driver
was speeding without paying attention to any possible risk,
we labeled the measurement as aggressive.
The evaluation based on our database can be seen in figure
5. As expected, measurements rated as keen have a high
probability of being aggressive and vice versa. Furthermore,
sedate and economical test runs can hardly be distinguished.
According to the the maximum likelihood hypothesis, which
is classifying the measurement M to the driving style ds with
To verify the rules in respect to someones own assessment,
persons drove the test track and rated their behavior in a
questionnaire according to the driving style description of
section IV-A from 1 to 10 afterwards.
Figure 6 shows a comparison between the own valuation
and the valuation of our algorithms of four test persons.
Generally, more than 75 % of the test persons assigned
themselves with the highest rating to the same class then
the algorithm did.
All the test persons rated their ability to drive economical
much higher than our system. This supports our hypothesis
that most drivers do not utilize the full energy saving
potential in their driving behavior.
C. Adjusting the Advised Velocity for Curves
Further tests were made evaluating the energy saving
potential of ANESA giving speed hints adjusted to the
driving style, rather than the default hint. Taking up the
example depicted in section I, an individual advisory speed
was derived based on the driving style. Mainly, the ratings of
the driving style in terms of aggressiveness and anxiousness
were used to generate the advisory curve velocity for the
particular driver.
Figure 7 shows an anxious and an aggressive driver approaching the same curve (45 ◦ with 100 m radius). The
advisory speed for the anxious driver was dedicated to be
© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including
reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists,
or reuse of any copyrighted component of this work in other works.
DOI: 10.1109/ITSC.2011.6082924
25
0.4
0.4
0.2
0.2
5
0
0
0
Aggr.
Anx.
Econ.
Keen
Sed.
Anx.
Econ.
Keen
Sed.
(a) The driver rated himself as most (b) As the set up system, the driver
likely to be sedate. So did the rating classified himself as keen. The econof our system.
omy was overrated by the driver.
0.6
0.4
0.4
0.2
0.2
0
Anx.
Econ.
Keen
Sed.
(c) The driver rated herself to be
sedate. Likewise did our algorithm.
However, our algorithm rated her
behavior more anxious than she did.
5
0
600
700
800
Road Position [ m ]
900
1000
35
Velocity - Aggressive Driver
Energy Consumption - Aggressive Driver
Velocity - Aggressive Driver - Speed Hint Adjusted
Energy Consumption - Aggressive Driver - Speed Hint Adjusted
30
25
20
Anx.
Econ.
Keen
Sed.
(d) The driver considered himself
as economical. However, the system
rated his behavior most likely to be
aggressive.
Fig. 6. Self-Assessment vs. System Rating of the driving behavior. More
than 75 % of the drivers rated themselves to the same class than the computer
did. Everybody overrated their economical behavior.
55 km/h, whereas it was 85 km/h for the aggressive driver.
Both drivers drove the scene once with ANESA
parametrized with their individual curve speed velocity and
once with 70 km/h per default. Driving with the default velocity advice was suboptimal for both drivers. The aggressive
driver started to mistrust the speed hint and lost energy on
extra acceleration. The anxious driver lost energy due to
braking.
With the advisory speed adjusted to the determined driving
behavior, both drivers were freewheeling to the speed they
were comfortable with. This, in fact, is the optimum in terms
of energy saving and time usage.
VII. CONCLUSIONS AND FUTURE WORKS
The value of future driver assistance systems can be highly
increased by adapting them to the character of the driver. This
work presents a probabilistic determination of the drivers
driving style, based on situational evaluation of the vehicle
data. To determine the individual driving style of a person,
the behavior in day-to-day traffic situations was analyzed
and a probabilistic affiliation of the driver being aggressive,
anxious, economical, keen, or sedate was made by means of
fuzzy-rules.
Using the maximum likelihood hypothesis, 83 % of the
measurements in our database could be correctly classified.
The most confusion was between sedate and economical
drives. As a questionnaire test showed, the implemented
algorithm rated the driving style of the drivers similar to
their own ratings.
Further, the benefit of ANESA giving speed hints adjusted to
the drivers personality was evaluated. Anxious drivers saved
30
20
15
15
10
10
5
0
Aggr.
35
25
Aggressive Driver mistrusts the given hint and
decides to accelerate. Additional energy is required.
5
0
Aggr.
10
(a) With the speed hint adjusted to the behavior of the anxious driver, the anxious
driver avoids braking and saves energy therefore.
Computer Rating vs. Self-Assessment: Marcus
1
Computer Rating
Self Assessment
0.8
0.6
15
Anxious driver gets default speed hint.
Unnecessary energy is used and decomposed by braking later.
500
Velocity [ m/s ]
Computer Rating vs. Self-Assessment: Aleksandra
1
Computer Rating
Self Assessment
0.8
20
Parametrised Speed Hint. Driver
freewheels exactly
to the curve speed he
is comfortable with.
10
Aggr.
30
25
20
15
35
Energy Consumption [ L/km ]
0.6
Velocity - Anxious Driver
Energy Consumption - Anxious Driver
Velocity - Anxious Driver - Speed Hint Adjusted
Energy Consumption - Anxious Driver - Speed Hint Adjusted
30
200
Energy Consumption [ L/km ]
0.6
35
Computer Rating vs. Self-Assessment: Fabian
1
Computer Rating
Self Assessment
0.8
Velocity [ m/s ]
Computer Rating vs. Self-Assessment: Matthias
1
Computer Rating
Self Assessment
0.8
0
300
400
500
600
Road Position [ m ]
700
800
900
(b) Adapting the advisory speed to the drivers aggressive behavior, the driver
finally accepts the energy hint, saves energy and mainly time.
Fig. 7. If the advisory speed of ANESA is adjusted to the driver behavior,
the full saving potential can be utilized.
energy due to less braking was avoided, whereas for aggressive drivers additional accelerations could be mitigated.
Since this tests are based on simulation only, further work
will validate this evaluations in real traffic scenarios.
VIII. ACKNOWLEDGEMENTS
This work is funded by Harman/Becker Automotive System. ANESA is a result of a long term cooperation between
Harman/Becker Automotive Systems and the FZI. Also,
this work was supported by IPG Automotive GmbH and
the German Ministry of Education and Research BMBFEFA2014 [9].
R EFERENCES
[1] Tobias Bär and Ralf Kohlhaas and Kay-Ullrich Scholl and J. Marius
Zöllner : Anticipatory Driving Assistance for Energy Efficient Driving
, FISTS - IEEE Forum on Integrated and Sustainable Transportation
Systems, 2011
[2] Michon J.A. : A critical view of driver behaviour models: What do
we know, what should we do? , Human behaviour an traffic safety,
Plenum New York , 1085
[3] Frida Saad : Dealing with Behavioural Adaptions to Advanced Driver
Support Systems , Modelling Driver Behaviour in Automotive Environments, 147ff, 2007
[4] O. Taubman Ben Ari and M. Mikulincer and O. Gillath : The multidimensional driving style inventory – scale construct and validation ,
Accident Analysis and Prevention, Vol. 36, No. 3, 323-332, 2004
[5] Yi-Shih Chung and Jinn-Tsai Wong : Investigating Driving Styles and
Their Connections to Speeding and Accident Experience , Journal of
the Eastern Asia Society for Transportation Studies, Vol. 8, 2010
[6] M Hoedemaeker : Driving behaviour with ACC and the acceptance
by individual drivers, IEEE Intelligent Transportation Systems Proceedings , 2000
[7] M. Canale and S. Malan and V. Murdocco : Personalization Of ACC
Stop and Go Task Based on Human Driver Behaviour Analysis , IFAC
15th Triennial World Congress, Bacelona, Spain , 2002
[8] P. Fancher and Z. Bareket and R. Ervin : Human-Centered Design of
an Acc-With-Braking and Forward-Crash-Warning System , Vehicle
System Dynamics, Vol. 36, 203 - 223 , 2001
[9] EFA2014 : Press information from the partners of the “Energy Efficient
Driving 2014”(EFA2014) research project. , German Ministry of
Education and Research , 2010