You are on page 1of 9

Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

Analysis of Human Driver Behavior in Highway Cut-in 2017-01-1402

Scenarios Published 03/28/2017

SeHwan Kim, Junmin Wang, Dennis Guenther, and Gary Heydinger


The Ohio State University

Joshua Every and M. Kamel Salaani


Transportation Research Center Inc

Frank Barickman
National Highway Traffic Safety Administration

CITATION: Kim, S., Wang, J., Guenther, D., Heydinger, G. et al., "Analysis of Human Driver Behavior in Highway Cut-in Scenarios,"
SAE Technical Paper 2017-01-1402, 2017, doi:10.4271/2017-01-1402.
Copyright © 2017 SAE International

Abstract accomplish tasks ranging from maintaining lateral position in a


marked lane to parallel parking a vehicle. In addition, there is an
The rapid development of driver assistance systems, such as
ongoing research trend in this area with the goal of developing highly
lane-departure warning (LDW) and lane-keeping support (LKS),
automated vehicles that do not even require human driver inputs.
along with widely publicized reports of automated vehicle testing,
have created the expectation for an increasing amount of vehicle
As automation systems dramatically change the role of a driver, the
automation in the near future. As these systems are being phased in,
concept of a “safe vehicle” must also be redefined. This can no longer
the coexistence of automated vehicles and human-driven vehicles on
include only the performance of the vehicle in extreme cases, such as
roadways will be inevitable and necessary. In order to develop
instability or collisions. This definition should be extended such that
automated vehicles that integrate well with those that are operated in
improving driving performance and avoiding accidents in everyday
traditional ways, an appropriate understanding of human driver
use cases are also covered. Such a demand leads to the necessity of
behavior in normal traffic situations would be beneficial.
developing new automobile driving safety criteria.
Unlike many research studies that have focused on collision-
Though some of these systems may be marketed as convenience
avoidance maneuvering, this paper analyzes the behavior of human
features, the safety of the driver and the occupants of surrounding
drivers in response to cut-in vehicles moving at similar speeds. Both
vehicles should not be compromised by their implementation. Due to
automated and human-driven vehicles are likely to encounter this
the complex nature of traffic dynamics, the actions of a vehicle
scenario in daily highway driving. This research has identified several
automation system have a direct influence on other occupants/
possible cut-in scenario configurations that can be experienced on the
vehicles nearby. Therefore, the evaluation of automated vehicle safety
highway. Data have been collected from a diverse pool of human
should not only include the direct effects of the system’s actions.
subjects using a driving simulator with preprogrammed scenarios. To
Adverse effects on safety caused by surrounding vehicles’ responses
understand each driver’s behavior in response to cut-in vehicles, a
to the automated vehicle’s behavior should also be considered.
novel means of visualization and analysis based on relative positions
Presently, several research efforts have attempted to evaluate the
is proposed in this paper. In addition, the paper provides information
function and safety of automated vehicles [2, 3], yet there exists a
on driver decision making when encountering cut-in vehicles. The
comparably small amount of research on the impacts of the individual
results could be employed as a set of guidelines for vehicle
vehicle automation features on nearby traffic/vehicles. In other
automation system behavior to ensure that they act in a manner
words, a majority of the current research on automated vehicle safety
consistent with human-driven vehicles.
does not take into consideration the safety of other vehicles driving
nearby.
Introduction
The safety of an automation system cannot be evaluated in a vacuum.
Advances in automation technologies could significantly improve the
Driving consists of continual interaction with surrounding traffic,
driving safety of automobiles by reducing the rate and severity of
therefore even if a vehicle behaves in a way that is not considered
vehicle accidents [1]. Various types of vehicle automation systems
inherently unsafe, the safety of surrounding vehicles may be affected
are currently available on the market. These systems are designed to
Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

unfavorably. As a result, the safest method of driving may be between sudden deceleration and rear-end collisions, and the
comprised of first securing one’s own safety, but second controlling modeling of driver crash-avoidance behavior when meeting a
the vehicle as to not perturb nearby vehicles. stationary vehicle, is the focus of a large amount of research efforts in
this area. In fact, NHTSA has published test protocols for Forward
Collision Warning (FCW) and Automatic Emergency Braking (AEB)
Levels of Automation systems that prevent or reduce the severity of rear end collisions [16,
The Society of Automotive Engineers (SAE) defines levels of 17]. These test procedures and many other forward collision-
automation by categorizing from Level 0 to Level 5 [4, 5]. One of the avoidance testing (by NHTSA or EuroNCAP etc.) consist of lead
critical factors of SAE’s definition is the degree of necessary human vehicle behaviors such as a sudden deceleration and/or encountering
input. SAE defines Level 0 vehicle as to those where the driver a stationary vehicle.
handles the entire driving task, and defines Level 1 vehicle as those
that provide either longitudinal or lateral control. Level 2 vehicle is
defined as those that control both of the two primary degrees of Cut-in Vehicle
freedom (lateral and longitudinal). However, with vehicles at, or A sudden lead vehicle deceleration, causing a rear-end collision, and/
below, this level, the human driver is expected to monitor the system or the presence of stationary vehicles could impose significant threats
behavior, and to be ready and able to intervene whenever necessary. to driving safety. However, the presence of a cut-in vehicle, faced by
An automobile equipped with adaptive cruise control (ACC) and lane many people every day is a more common occurrence and remains a
centering is an example of a Level 2 vehicle. Of vehicles that have source of potential danger. In many cases of longitudinal interaction,
responsibility for monitoring the environment, those that require both with stopped and decelerating vehicles, drivers can maintain a
fallback control are classified as Level 3 while those that can handle proper headway through appropriate and early intervention. In cut-in
all driving tasks under specific circumstances are Level 4. Level 5 situations, drivers have no choice but to experience a small headway
vehicles are those that no longer require driver controls and operate relative to the cut-in vehicle regardless of their will. Hence,
in all domains. responding appropriately to cut-in vehicles is important for safe
driving. Due to this reduced following distance forced by the vehicle
There exist a number of classifications for many levels of vehicle cutting in, if a vehicle were to begin decelerating rapidly or were
automation. However, for consistency, The National Highway Traffic moving much slower than the subject vehicle, the ability of AEB
Safety Administration (NHTSA) has chosen to classify vehicles systems to help avoid or mitigate a collision could be substantially
automation by the SAE definition [6]. In the meantime, the reduced. Unfortunately, it is difficult to define an ideal response to
Department of Transportation (DOT) refers to vehicles above Level 2 cut-in vehicles. Therefore, analysis of human driver’s responses to
as Highly Automated Vehicles (HAV). HAVs can assume complete cut-in vehicles may provide a means to aid in developing a system
responsibility for monitoring the driving environment when in their capable of handling all situations or all cut-in vehicles ideally, as
domain, but domains and fallback responsibility can vary. Level 5 vehicles would.

As automated vehicles continue to be introduced into the fleet, the Research on a driver’s response to cut-in vehicles is complex given
co-existence of automated and traditionally-driven vehicles on the that there are numerous factors that affect a driver’ behaviors
road is inevitable. Improper behavior of automated vehicle systems in including traffic, road types, weather, and road work, to name only a
this transitional environment may lead to a collision and/or cause a few. Responses of drivers when encountering a cut-in vehicle can be
severe threat to other drivers. Therefore, in order to realize the divided into two basic categories, braking/ accelerating, and steering.
improvement in safety desired from automation, a deep This means the driver can elect to stay in the lane and allow a cut-in
understanding of human driver behaviors and identification of vehicle to become a lead vehicle, or to change a lane to the left or
different human driver characteristics are necessary. right. These two behaviors may affect the safety of other cars by
causing sudden deceleration or sudden lane changes. Of course these
Background two behaviors will differ in situations where there is a decelerating
lead vehicle or a stopped vehicle.
Research efforts on individuals’ driving characteristics have been
carried out in recent years [7, 8]. The development and use of
Figure 1 shows some possible cut-in scenarios that drivers may
accurate driver models can make a significant contribution to the
experience on a three-lane highway. Situations occurring on a
safety of automated vehicles, and can be used to evaluate the
three-lane highway are studied in this research because its data and
performance of an automated vehicle in various scenarios. To assist
scenarios could be easily applied to two-, four-, or five-lane
drivers in collision-avoidance scenarios, driver models specific to
highways. Except for rare cases such as facing a cut-in vehicle while
these cases have been developed [9, 10]. Since the models should in
changing its lane, the number of possible configurations can be
some cases be adapted to an individual, authors in [11, 12, 13]
reduced to 12 basic scenarios. When a driver is in the left lane, they
proposed dynamic driver models to identify each driver’s
face the possibility of a cut-in vehicle coming from the right side
characteristics such as driving skills. In addition, analysis of
only. Similarly, when in the right lane, they can only encounter a
collision-avoidance scenarios such as lane-changing and merging is
cut-in vehicle coming from left side assuming the driver is restricted
provided in [14].
from using the shoulder. When a driver is in the center lane, they can
meet a cut-in vehicle from either side.
Among various studies on driving behaviors, rear-end collisions are
the most predominant crash type and are also the most studied ones.
According to NHTSA [15], rear-end collisions account for
approximately 30% of all car accidents. Therefore, the relationship
Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

Figure 1. 12 possible scenarios that driver can encounter cut-in vehicles on the three lane highway

In Scenario 1, the subject driver is in the center lane and faces a 10-DOF Vehicle Dynamic Model
cut-in vehicle coming from the right side. At this moment, there are The 10-DOF vehicle dynamic model used for this study consists of
no other vehicles around the subject driver, except for the cut-in 6-DOF ground vehicle model and a 4-DOF tire dynamics model. The
vehicle. In this situation, the driver can choose to either stay in the vehicle model and all equations are based on a suspension-
lane and possibly decelerate, or the driver may make a lane change to incorporated 6-DOF (x, y, yaw, roll, pitch, and lateral load transfer
the left or right. Scenario 2 is the same as Scenario 1, but because of equations) model in [20]. The 4-DOF tire dynamics are simulated by
traffic constraints on the third lane, the subject driver can either stay using a Magic Formula Tire Model [21]. To verify the 10-DOF
in their lane or change lanes only to the left. In Scenario 4, there are vehicle model, CarSim, a simulation software that predicts vehicle
traffic constraints on both sides, so the driver can only stay in his/her dynamics, is used. The model for a D-class sedan is replicated in the
lane. Although these scenarios seem to be similar since the subject driving simulator and the vehicle properties used for CarSim
driver is faced with a cut-in vehicle coming from the right, they are verification are shown in Table1.
all different and the required driving behaviors are all different as
well. Scenarios 7 to 12 are situations where the subject driver Table 1. Vehicle parameters
encounters a cut-in vehicle coming from the left side. Although they
seem to be exactly same as Scenarios 1 to 6 with the only difference
being the direction, drivers’ behaviors are not the same. Therefore,
Scenarios 1 to12 should not be considered as identical.

Data Collection
Driving simulators have been used in many similar studies [13, 18].
They are essential to this type of work because they make it possible
to easily and repeatedly create situations that are difficult to conduct
in real driving. This research has utilized a driving simulator that
provides for the study of participants’ behaviors with a
preprogrammed driving simulation that features a three-lane highway
and several traffic situations including cut-in vehicles.

Driving Simulator
The driving simulator is based on the Christie digital Holostage mini
CAVE system. This system provides a real-life-like driving
environment as seen in Figure 2. The simulator consists of two rear
projectors with two screens, a surround sound system and a driver’s
seat with a steering wheel, brake and accelerator pedals. Three inputs
from study participants, namely steering wheel angle, brake and
accelerator pedal positions are measured. Additionally, ultrasonic and
PIR sensors located above two pedals measure the driver’s foot
position in real time. The measured driver input data is sent into a
10-DOF ground vehicle mathematical model in Matlab/Simulink as Figure 2. The driving simulator system.
inputs. Through a TCP socket channel to Vizard [19], a virtual reality
software, the outputs (x, y, and yaw) from the vehicle dynamic
models are returned.
Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

Participants Drivers may be vigilant and very careful during the test since they
A total of 31 drivers with valid U.S. driver’s licenses participated in encounter more cut-in vehicles than usual in daily driving. To more
this research. The participating drivers are of different ages (all over accurately measure drivers’ natural behavior in response to cut-in
age 18), genders, and levels of driving experience. However, data vehicles, several vehicles in addition to the cut-in vehicles were
from two participants who claimed their lack of experience and skill arranged near the subject vehicle with constant speeds. This was done
in driving were excluded in analysis (e.g. drivers that avoid driving so the participants could not anticipate when a vehicle would cut-in.
on highway in their daily driving). Table 2 shows participants’ All cut-in vehicles have the same constant lateral speed of 0.1m/s.
information in detail. This results in the cut-in vehicles taking 3.6 seconds to complete the
lane change.
Table 2. Self-reported driver information

Results
The human driver test results are presented in this section. First,
categorization of the driver behavior in response to a cut-in vehicle is
introduced and its distributions for Scenarios 1, 2, and 4 are analyzed.
Second, an alternative analysis method for cut-in vehicle scenarios
named range-range domain is proposed to distinguish each driver’s
characteristics and to predict driver’s behavior in many different
situations. Lastly, mistakes that humans and/ or automated vehicle
systems can make when responding to a cut-in vehicle are introduced.

Categorization and Distribution of Driver Behavior


From a total dataset of 1116 runs (12 scenarios * 3 different ranges *
31 participants) collected from the 31 participants, the different
responses of drivers to cut-in vehicles may be analyzed. Even though
Driving Task drivers respond in widely varying ways, by operating the brake,
accelerator, and steering wheel, these behaviors can be largely
The test route is comprised of two 25-mile long segments on a
summarized into two different outcome-based groups. Drivers either
three-lane highway. The test takes approximately 50 minutes
stay in their lanes or change lanes.
including a three-mile introductory course. On the introductory
course, the participants became familiar with the driving simulator in
First group is the case of staying in the lane, this behavior can further
order to make them more comfortable with the simulator system prior
be categorized into two subgroups, in which: 1) the driver generates a
to testing. The speed limit on the highway for all tests is 65mph. For
moderate deceleration of less than 0.3g, with/without applying
the sake of consistency, all participants were asked to drive in the
brakes; and 2) the driver applies the brakes and generates a
range of 60 to 65 mph during the test.
deceleration of 0.3g or greater. As in lead vehicle following scenario
or cruising, in situations where there is no cut-in vehicle, almost all
Scenarios in the Simulation drivers produce deceleration of less than 0.3g. Hence, this research
The participants randomly encounter several cut-in vehicles from the has assumed decelerations of less than 0.3g as moderate deceleration.
left or right side while driving with a constant speed of around 63 to In fact, with the exception of deceleration that has been caused due to
65mph. The cut-in vehicles initiate their lane changes with the same unfamiliarity with the driving simulator, deceleration of 0.3g has only
speed as the subject vehicle and maintain that speed for the duration been detected in cut-in scenarios.
of the test. This study only analyzes constant speed cut-in vehicle
cases. That is, the moment when the subject driver encounters the Second group is the case of changing lanes. This behavior can also be
cut-in vehicle, the relative speed and range rate of the subject and further classified into three subgroups: 1) by changing lanes while
cut-in vehicles are zero. Scenario details are shown in Table 3. continuously stepping on the accelerator (to maintain speed) and
operating the steering wheel; 2) by changing lanes by releasing the
Table 3. Details in cut-in vehicle scenarios accelerator and operating the steering wheel; and 3) by steering to
change lanes while applying continuous or intermittent braking.
Some participants changed lanes after initially stepping on the brake
while several others pressed the brake pedal while changing lanes.
Both of these cases are considered as braking and steering. Like this,
drivers’ responses to cut-in vehicles could be largely divided into two
groups, and then categorized again into five more subgroups. The
distribution of 29 human subjects’ behavior (excluding the two
participants) for all scenarios with initial range of 15 m is shown in
Table 4.
Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

Table 4. Distribution of driver behavior for all scenarios In the cases where the driver had a choice as to whether to stay in
their lane or change lane, more than 70% of drivers responded by
staying in their lane. In other words, most drivers stayed in their lane
when they encountered cut-in vehicles in all situations. In the cases
where the cut-in-vehicle came from the right side, many drivers
tended to change lane to the left. In addition, in the case of Scenario
1, without any constraints except for the cut-in vehicle, the rate of
lane-changes was higher than in Scenario 2. For example, in
comparing Scenarios 1 and 2, there was a common option to change
lane to the left. In case of Scenario 2, however, lane change behaviors
were seen relatively less. As the left lane usually is a passing lane
with a high average vehicle speed, drivers might not want to change
lane rapidly when there is a bit of restriction. In Scenario 4, due to
the many traffic constraints around the subject vehicle, more drivers
generated decelerations of 0.3g or greater. Thus, drivers’ behaviors
and tendencies were different depending on traffic constraints and
positions, even though they encountered cut-in vehicles from the
In reviewing the data presented in Table 4 it can be seen that a large same side.
number of similarities and differences exist between the driver’s
response to the studied scenarios. In all cases, the majority of drivers Range-Range Plot
respond by staying in their lane. In addition, in most cases as the
Even though some drivers show the same response among five
number of traffic constraints is reduced, more drivers tend to change
possible behaviors to the cut-in vehicle mentioned above, each person
lanes. In order to analyze if further differences exist in drivers’
has a different reaction rate and a different degree of feeling
control inputs applied during these types of scenarios, a subset of
threatened or safe. Hence, it is important to review the responses to
representative scenarios are selected for further review.
define such differences. Range rate or time to collision (TTC) data
provide significant insight when analyzing driver behavior in
To distinguish and analyze driving behavior, further details on driver
response to facing a decelerating, slower moving, or stationary lead
input type will be given for Scenarios 1, 2 and 4. Scenario 1 has no
vehicle. For example, the range-range rate diagram has been used for
traffic constraint, Scenario 2 has one traffic constraint, and Scenario
collision avoidance in longitudinal cases [22]. Since cut-in vehicles in
4 has two traffic constraints. In Scenario 4, drivers only have the
most cases do not generate a high deceleration or high range rate,
option of staying in their lane in response to the cut-in vehicles. The
typically these metrics may not provide much meaningful
detailed distribution of participants’ behavior for Scenarios 1, 2, and
information. For example, even moments with low range rate or
4 is shown in Table 5.
increasing TTC values could threaten drivers.
Table 5. Distribution of driver behavior in cut-in vehicle scenarios 1, 2, and 4
For an effective analysis, a different approach called range-range
domain is introduced in this paper. In the range-range domain, several
parameters including braking, acceleration, steering, and trajectory
can be easily analyzed simultaneously. Figure 3 shows the range-
range plot with longitudinal range to the cut-in vehicle on the x-axis
and lateral range to the cut-in vehicle on the y-axis. The range-range
domain consists of conflict/ non-conflict zones and the crash zone.

Figure 3. Range-range domain


Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

The conflict zone can be defined as when the cut-in vehicle starts In the range-range domain, however, it is easier to see the differences
laterally overlapping the subject vehicle in the direction of driving. between the four drivers. Since the initial range is 12m, point A in
Typically, it takes less than a second to enter into the conflict zone Figure 6 is the initial starting point and the location in the range-
when the driver encounters a cut-in vehicle. The section that is not in range domain will move toward the x-axis unless the driver changes
the conflict zone or not influenced by the cut-in vehicle due to a lanes. If the driver did not respond to the cut-in vehicle and
sufficient safe distance is defined as non-conflict zone. When the maintained speed, point A will move down to point B perpendicular
distance between the vehicle centers is smaller than the combined to x-axis and stay at the initial range value. The red line that is the
half of the length of two vehicles (approximately (L1+L2)/2) and is in final traffic constraint line shown here indicates the limit that the
the conflict zone, a collision is expected to have occurred (see Figure vehicle cannot go over due to traffic constraints after the cut-in
4). This is called the crash zone. When faced with a cut-in-vehicle, vehicle has completed the lane change. The final traffic constraints
drivers enter in the conflict zone rapidly and generally try to head for line is time-varying, but the final position is shown in Figure 6. Going
either non-conflict zone 1 or 2. When drivers maneuver by changing over the limit indicates colliding with nearby vehicles that are not the
lanes, they head for the non-conflict zone 1. When staying in their cut-in vehicle.
lane they head for non-conflict zone 2. When heading for non-conflict
zone 2, the driver will approach his or her own desired following
distance in this zone. Driver’s desired following distance changes
depending on speed. Hence, depending on the speed, the conflict zone
either contracts or expands.

Figure 6. Four drivers’ behavior in range-range domain


Figure 4. Parameters in cut-in scenarios
By reviewing the information presented in Figure 6, the following
observations can be made. Driver 23 forcefully applied the brake
Analysis of Scenario 2 in Range-Range Domain generating high decelerations and quickly moved to his/her own
Figures 5 and 6 show the trajectory and range-range plots of four desired distance to the cut-in vehicle. The red bold line on driver 23’s
drivers in Scenario 2 with an initial range of 12 m. Drivers 7 and 18 profile represents the period when the driver generates a deceleration
changed lanes to the left in response to a cut-in vehicle and Drivers 4 over 0.3g. Driver 4 slowly approached the desired distance by
and 23 stayed in the lane. It may be possible to distinguish each stepping slowly and lightly on the brake. In the case of lane changing,
driver’s behavior through trajectories, but it is hard to see the Driver 7 performed a lane change faster than Driver 18. Based on the
difference between Drivers 4 and 23. drivers’ behaviors in the same situation, the characteristics of each
person could be easily understood and analyzed in the range-range
domain.

Predicting Driver Behaviors in Range-Range Domain


Figure 7 represents data from Scenario 2 with initial ranges of 12m
and 18m. It shows the locations where the driver initiates braking or
steering in the range-range domain. The circle indicates the location
when the drivers started applying on the brake. Blue circles represent
drivers that have changed lanes, and the red circles represent drivers
that stayed in their lane. If the drivers generated deceleration of 0.3g
or greater, it is indicated by red stars. Diamonds indicate the time
when the drivers started operating the steering wheel. The filled
markers indicate the responses when the initial range is 18m while
empty markers indicate initial range of 12m.

Figure 5. Trajectories of four drivers


Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

Figure 7. Distributions of brake pedal and steering wheel values of scenario 2 (with initial range of 12m and 18m) in range-range domain

As Figure 7 shows, when encountered with a cut-in vehicle, all Undesirable Responses to Cut-in Vehicles
drivers that stayed in their lane, and generated high decelerations, As previously discussed, abrupt braking due to a sudden perception
started braking in the non-conflict zone 1. In other words, these of a threat, or a delayed response time before engaging in the ideal
drivers felt urgency right at the time that the cut-in vehicle started course of action, is a natural but less-than-ideal responses to cut-in
moving into the lane and applied the brakes immediately. What is vehicles. Besides such instances, human errors may occur in
interesting is that not all drivers that applied the brakes in the situations where the cut-in vehicle has become the leading vehicle. In
non-conflict zone 1 produced a deceleration of 0.3g or greater. Two these cases the following vehicle will most often slow down to create
drivers that responded to the cut-in vehicle by changing lane and a desired separation from the leading vehicle. In such cases the driver
applying the brake in the non-conflict zone 1. In such circumstances, may decelerate abruptly to the point where the distance to the leading
drivers changed lanes with a deceleration of less than 0.3g. When the vehicle has become longer than desired. Such a situation is not
initial range is 12m, there was a tendency among the drivers to economically efficient, and could trigger a collision with following
generate a deceleration of less than 0.3g in the non-conflict zone 1 traffic. Figure 8 represents Driver 15’s response in the range-range
while changing lanes. In addition, there was a tendency among the domain after decelerating in Scenario 2 and then finding their own
drivers that stayed in their lane, to produce a deceleration of 0.3g or desired distance to the leading vehicle. Point A is Driver 15’s desired
greater, which is not an ideal braking response to the situation if there following distance to the cut-in vehicle and the range marked with an
is a following vehicle behind. Putting aside the inevitable human x is the unnecessary error that has been documented for the driver.
response time, the ideal response would be gradually decelerating
starting in the non-conflict zone 1 and then increasing the distance to
the cut-in vehicle if driver wants to stay in the lane.

For the case of 18m, due to the increase in the initial range, the
number of drivers that produced a deceleration of 0.3g or greater
decreased. Also, some of the drivers responded by braking early in
the non-conflict zone 1 without generating high decelerations. With
an increase in the initial range, it is hard to say definitively whether it
is better to brake early or comparatively later. However, the research
results show an array of responses in the range-range domain given
the same scenario with different initial ranges.

Figure 8. Driver inappropriate responses in range-range domain


Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

As shown, one could evaluate a system of maintaining an ideal 4. SAE International Surface Vehicle Information Report,
distance to the leading vehicle in the most efficient way. One may “Taxonomy and Definitions for Terms Related to On-Road
also analyze each driver’s undesirable responses in the range-range Motor Vehicle Automated Driving Systems,” SAE Standard
domain. Figuring out drivers’ desired distance to the leading vehicle J3016, Rev. Jan. 2014.
and how to account for the human errors are useful for the 5. SAE International Surface Vehicle Information Report,
development of automated vehicles that may reduce the chance of “Guidelines for Safe On-Road Testing of SAE Level 3, 4, and 5
causing threats to surrounding vehicles. Prototype Automated Driving Systems (ADS),” SAE Standard
J3018, Rev. Mar. 2015.
Limitations 6. NHTSA, “Federal Automated Vehicle Policy - Accelerating the
Each driver exhibits his or her unique responses or driving habits Next Revolution in Roadway Safety,” Sep. 2016.
while driving with their vehicles. Due to the virtual nature of the 7. MacAdam, C. C., "Understanding and Modeling the Human
driving simulator, the data obtained from the simulator may be Driver," Veh. Syst. Dyn. Int. J. Veh. Mech. Mobility, vol. 40, nos.
inherently different from data collected in real driving situations. 1-3, pp. 101-134, 2003, doi:10.1076/vesd.40.1.101.15875.
Also, as tests have been carried out with under a research assumption 8. Plöchl, M., and Edelmann, J., “Driver Models in Automobile
that the initial range rate between the subject vehicle and the cut-in Dynamics Application,” Veh. Syst. Dyn. Int. J. Veh.
vehicle is zero, there are limits to accounting for all real-life cut-in Mech. Mobility, vol. 45, nos. 7-8, pp. 699-741, 2007,
vehicle situations. Comparative research against real daily data, and doi:10.1080/00423110701432482.
research on other factors including driver age, driving experience are
9. Jurecki, R., Stariczyk. T.L., “Driver Model for the
being carried out.
Analysis of Pre-Accident Situations,” Veh. Syst. Dyn.
Int. J. Veh. Mech. Mobility, vol. 47, pp. 589-612, 2009,
Summary/Conclusions doi:10.1080/00423110802276028.
With the continuously increasing amount of vehicle automation 10. Hong, T., Kwon, J., Park, K., Lee, K. et al., "Development of a
systems and development of driver assistance systems, there are both Driver's Intention Determining Algorithm for a Steering System
expectations and concerns. As these systems are introduced, driving Based Collision Avoidance System," SAE Technical Paper
safety is a primary factor, and it would be desirable for a system to 2013-01-0054, 2013, doi:10.4271/2013-01-0054.
address every situation without inducing new hazards. By utilizing 11. Zhang, Y., Lin, W.C., and Chin, Y.K.S., "A Pattern-Recognition
data collected from a driving simulator, this study defined and Approach for Driving Skill Characterization," IEEE
analyzed a set of cut-in vehicle scenarios that are commonly Transactions on Intelligent Transportation Systems, vol. 11, no.
experienced in real-world driving situations with a potential risk of 4, pp. 905-916, 2010, doi:10.1109/TITS.2010.2055239.
accidents. In this paper, range-range domain is proposed for the 12. Wang, C., Zhang, X., Guo, K., Ma, F. et al., "Application
analysis of cut-in vehicle scenario. This research has identified an of Stochastic Model Predictive Control to Modeling Driver
array of drivers’ responses to cut-in vehicles in each of the given Steering Skills," SAE Int. J. Passeng. Cars - Mech. Syst.
situations with different ranges. Also, using the range-range domain, 9(1):116-123, 2016, doi:10.4271/2016-01-0462.
each driver’s behaviors were classified and their characteristics were
13. Schnelle, S., Wang, J., Su, H., and Jagacinski, R., “A Driver
differentiated. Predictions for drivers’ tendencies and behaviors in
Steering Model with Personalized Desired Path Generation,”
each situation have also been made. Lastly, undesirable reactions
IEEE Transactions on Systems, Man and Cybernetics:
generated from drivers have been discussed. For the development of
Systems, Vol. 47, No. 1, pp. 111-120, 2017, doi:10.1109/
automated vehicles that operate safely in mixed traffic environments,
TSMC.2016.2529582.
it is essential to research human driver’s behaviors. Future work
should include driver modeling for the cut-in vehicles. 14. Jula, H., Kosmatopoulos, E.B., and Ioannou, P.A., "Collision
Avoidance Analysis for Lane Changing and Merging," IEEE
Transactions on Vehicular Technology, vol. 49, no. 6, pp. 2295-
References 2308, 2000, doi:10.1109/25.901899.
1. NTSB, “The Use of Forward Collision Avoidance Systems to 15. NHTSA, “Traffic Safety Facts 2012: A Compilation of Motor
Prevent and Mitigate Rear-End Crashes,” Special Investigation Vehicle Crash Data from the Fatality Analysis Reporting
Report, PB2015-104098, 2015. System and the General Estimates System,” DOT HS 812
2. Rau, P., Yanagisawa, M, and Najm, W., “Target Crash 032, 2012. URL:https://crashstats.nhtsa.dot.gov/Api/Public/
Population of Automated Vehicles,” presented at 24th ViewPublication/812032.
International Technical Conference on the Enhanced Safety of 16. NHTSA, “Forward Collision Warning System Confirmation
Vehicles, Sweden, June 8-11, 2015. Test,” Feb. 2013.
3. Gibson, M., Lee, J., Venkatraman, V., Price, M. et al., "Situation 17. Forkenbrock, G. J., and Snyder, A. S., “NHTSA’s 2014
Awareness, Scenarios, and Secondary Tasks: Measuring Automatic Emergency Braking Test Track Evaluations,”
Driver Performance and Safety Margins in Highly Automated NHTSA, DOT HS 812 166, June 2014.
Vehicles," SAE Int. J. Passeng. Cars - Electron. Electr. Syst. 18. Gaspar, J., Brown, T., Schwarz, C., Chrysler, S. et al., "Driver
9(1):2016, doi:10.4271/2016-01-0145. Behavior in Forward Collision and Lane Departure Scenarios,"
SAE Technical Paper 2016-01-1455, 2016, doi:10.4271/2016-
01-1455.
Downloaded from SAE International by University of Leeds, Monday, August 13, 2018

19. WorldViz Virtual Reality Software, “Vizard Virtual Reality 22. Burgett, A., Carter, A., Miller, R., Najm, W., and Smith, D.,
Software,” http://www.worldviz.com/virtual-reality-software- “A Collision Warning Algorithm for Rear-End Collisions,”
downloads, accessed Nov. 2015. presented at 16th International Technical Conference on
20. Berntorp, K., “Derivation of a Six Degrees-of-Freedom Ground- Enhanced Safety of Vehicles, Washington, D.C., May 1998.
Vehicle Model for Automotive Applications,” Department of
Automotive Control, Lund University, Sweden, Feb. 2013. Contact Information
URL: http://lup.lub.lu.se/record/3459785. SeHwan Kim can be reached at
21. Pacejka, H. B. and Bakker, E., “The Magic Formula Tyre kim.5228@osu.edu
Model,” Proceedings 1st Int. Colloquium on Tyre Models for
Vehicle Dynamics Analysis, vol. 21, pp. 1-18, Delft, Oct. 1991. Acknowledgments
Funding for this research and future work is provided by the National
Highway Traffic Safety Administration (NHTSA)

The Engineering Meetings Board has approved this paper for publication. It has successfully completed SAE's peer review process under the supervision of the session organizer. This process
requires a minimum of three (3) reviews by industry experts.

This is a work of a Government and is not subject to copyright protection. Foreign copyrights may apply. The Government under which this paper was written assumes no liability or
responsibility for the contents of this paper or the use of this paper, nor is it endorsing any manufacturers, products, or services cited herein and any trade name that may appear in the paper has
been included only because it is essential to the contents of the paper.

Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE International. The author is solely responsible for the content of the paper.

ISSN 0148-7191

http://papers.sae.org/2017-01-1402

You might also like