Sei sulla pagina 1di 8

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/295103523

DOES PRACTICING ONE STEERING TASK


IMPROVE PERFORMANCE ON A SUBSEQUENT
STEERING TASK?

Article · October 2015

CITATIONS READS

0 59

3 authors, including:

Jacqueline Jenkins Seyed Amirhossein Hosseini


Cleveland State University Iowa State University
17 PUBLICATIONS 58 CITATIONS 2 PUBLICATIONS 0 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Seyed Amirhossein Hosseini on 19 February 2016.

The user has requested enhancement of the downloaded file.


DOES PRACTICING ONE STEERING TASK IMPROVE
PERFORMANCE ON A SUBSEQUENT STEERING TASK?

J. M. Jenkins D. M. Lewis S. A. Hosseini


Cleveland State University Cleveland State University Cleveland State University
j.m.jenkins41@csuohio.edu d.lewis88@vikes.csuohio.edu s.hosseini@vikes.csuohio.edu

ABSTRACT
A driving simulation experiment was conducted to examine whether practicing one steering task would result in
improved performance of a subsequent steering task. The three steering tasks used were lane keeping on a straight
road, a series of lane changes spaced 85 meters apart, and a series of lane changes spaced 65 meters apart. A total of
44 participants, 16 females and 28 males ranging in age from 19 to 63 years, drove one of four sequences of the
steering tasks. The lane position data for each lane change and the travel time between lane changes was used to
calculate a cost, representing a combination of the participants’ accuracy and efficiency performing the lane
changes. The cost data was then used to test whether 1) practice is better than no practice; 2) practicing a less
challenging but similar steering task is good practice for a more challenging steering task; and 3) practicing a more
challenging but similar steering task is good practice for a less challenging steering task. Ignoring any effects of the
characteristics (e.g. sex and age) of the participants who drove the different sequences of steering tasks, the results
indicate that practicing the more challenging lane changing task had a significant impact on the performance of the
subsequent, less challenging but similar task. However, when the characteristics of the participants are taken into
account, the results indicate that the performance of the subsequent steering task was not significantly affected by
practicing a previous steering task.

KEYWORDS
Practice, learning, performance improvement, lane keeping, lane changing, steering task

INTRODUCTION
When a driving simulation study is designed, it is common to include a practice scenario, driven before the
experimental scenario(s). There are many reasons to include a practice scenario, one of which is to provide
participants time to learn how the simulator responds to their steering wheel and pedal inputs and adjust their use of
the controls. This study is focused on examining how practicing one steering task can impact the performance on a
subsequent steering task.

BACKGROUND
The design of practice scenario varies from those which offer participants a fixed amount of time driving or a fixed
driving distance, to those that allow participants to drive until they report that they are comfortable. The practice
scenario may be designed to have roadway environment, roadway alignment, and traffic controls similar to that of
the subsequent experimental scenarios and participants may or may not perform similar tasks. Does the task matter?

A number of studies have shown that participants improve their control of the simulator vehicle over repeated trials
of the same task. For instance, McGehee et al. (2004) demonstrated that participants driving an uneventful section of
rural two-lane road exhibited a decrease in large steering reversals and variance in steering input and concluded that
steering behavior stabilized in 240 seconds. Sahami and Sayed (2010) analysed the average speed and lateral
standard deviation of participants driving a series of 18 curves on a two-lane road. They found that by the end of the
curves, two of the 23 participants had not been able to increase their speed. Ronen and Yair (2012) investigated the
impact of the complexity of the driving environment on the time required for participants to improve their lane
keeping performance. They analysed steering wheel deviations, lane position, longitudinal speed, and deviations
from the driving lane and found that the more complex environments (i.e. curved road and urban road) required
approximately fifteen minutes of practice and the less complex environment (i.e. straight road) required
approximately twelve minutes of practice for the performance to stabilize. All three of these studies evaluated the
performance improvement on a lane keeping task. The time needed for participants to exhibit consistent
performance was different between these studies and is likely attributed in part to the different roadway
environments and alignments used and different analysis approaches taken.

Performance improvement while practicing lane changes has also been studied. Jenkins and Moran (2014) found
that performance stabilized after approximately 15 lane changes when the participants were travelling at
approximately 25 miles per hour and the lane changes were spaced 40 meters apart. Jenkins and Seck (2014) had 18
participants drive at 55 miles per hour and increase their speed as they were comfortable to do so. They negotiated
ten lane changes spaced 80 meters apart followed by ten more lane changes spaced 104 meters apart. Another 18
participants were directed to travel at specified speeds ranging from 30 to 65 miles per hour and make lane changes
that were spaced to provide approximately 3.6 seconds of time between lane changes. They found that after 20 lane
changes, 17 of the 36 participants had learned to interact with the simulator. The tasks in these studies were very
similar, with participants having to make a lane change approximately every 3.6 seconds; however the speed at
which the participants drove was different. In the study by Jenkins and Seck (2014), the participants travelled at
greater speeds and required more lane changes to improve their performance compared to those in the study by
Jenkins and Moran (2014).

Sahami and Sayed (2013) took this line of inquiry a step further and examined whether the improvement of a slalom
task would impact the performance of negotiating a series of 18 curves on a two-lane road. This latter task was the
same task used in their previous study (Sahami and Sayed, 2010). They found that those who had improved to where
they were showing consistent performance on the slalom task more quickly exhibited consistency on the curve
negotiation task, as compared to those who were still improving on the slalom task. The following study builds upon
this knowledge by examining the impact of the performance improvement of one steering task on another.

EXPERIMENT
A driving simulation experiment was designed to examine the performance improvement of participants while
conducting three steering tasks: 1) lane keeping (LK) on a straight road; 2) a series of lane changes spaced 85 meters
apart; and 3) a series of lane changes spaced 65 meters apart. Making lane changes every 65 meters requires greater
steering input than making lane changes every 85 meters and is therefore a more challenging steering task. The three
tasks were presented in the following four sequences and an equal number of participants were randomly assigned to
each.
 Sequence 1, LK – 65 – 85, participants performed a lane keeping task for 7 km, followed by 30 lane
changes spaced 65 meters apart, then 30 more lane changes spaced 85 meters apart.
 Sequence 2, 65 – 85 – LK, participants began with 30 lane changes spaced 65 meters apart followed by 30
more lane changes spaced 85 meters apart and then 7 km for lane keeping.
 Sequence 3, 85 – 65- LK, participants began with 30 lane changes spaced 85 meters apart followed by 30
more lane changes spaced 65 meters apart and then 7 km for lane keeping.
 Sequence 4, LK – 85 – 65, participants started with a lane keeping task for 7 km, followed by 30 lane
changes spaced 85 meters apart, then 30 more lane changes spaced 65 meters apart.

These sequences were designed to test whether practicing one steering task would impact the performance of a
subsequent steering task. They were designed to address the following questions:
 Is practice on a different steering task better than no practice? The impact of practicing the lane keeping
task can be evaluated as the difference in performance on the subsequent lane changing tasks by comparing
the performance of those who first drove the lane keeping task to those who did not. Therefore, those who
drove Sequence 1, LK – 65 – 85, can be compared to those who drove Sequence 2, 65 – 85 – LK, to see the
impact of lane keeping on the performance of lane changes spaced 65 meters apart. Similarly, those who
drove Sequence 4, LK – 85 – 65 – LK, can be compared to those who drove Sequence 3, 85 – 65- LK, to
see the impact of lane keeping on the performance of lane changes spaced 85 meters apart. If lane keeping
is good practice then those who first drove the lane keeping task would perform better on the subsequent
steering task.
 Is practicing a less challenging but similar steering task good practice for a more challenging steering task?
The impact of practicing the less demanding lane change task can be evaluated by comparing the
performance on the more challenging lane change task between those who drove Sequence 3, 85 – 65- LK
to those who drove Sequence 2, 65 – 85 – LK. If practicing a less demanding but similar task is good
practice then those who drove Sequence 3 would exhibit better performance on the lane changes spaced 65
meters apart.
 Is practicing a more challenging but similar steering task good practice for less challenging steering task?
The impact of practicing the more demanding lane change task can be evaluated by comparing the
performance on the less challenging lane change task between those who drove Sequence 2, 65 – 85 – LK
to those who drove Sequence 3, 85 – 65- LK. If practicing a more demanding but similar task is good
practice then those who drove Sequence 2 would exhibit better performance on the lane changes spaced 85
meters apart.

Driving Simulator
The experiment was conducted in the Cleveland State University (CSU) driving simulation laboratory. The
DriveSafety RS100 used for this experiment consists of a Playseat driving seat and base, Logitech steering wheel
and pedals and a single Dell 2405 FPW 24 inch monitor to display the forward driving image. The image for the rear
view mirror is integrated into the forward driving image. There is no instrument panel on this simulator.

The driving scenarios were developed using a rural 4 lane divided freeway, without any ambient traffic. The desired
lane changes were indicated using target arrows, placed in the center of the desired travel lanes, as shown in Figure
1. Location triggers were placed at the same locations as the target arrows such that when the participant drove
through a target arrow, a location trigger would be activated. The trigger activation was captured in the data
collection file and facilitated the extraction of performance data for each lane change. All default data collection
variables were recorded at a rate of 60 times per second.

Figure 1. Target arrows indicate desired travel lanes and lane changes

Participants
A total of 44 adult participants were recruited from CSU through person-to-person contact. The 16 females and 28
males ranged in age from 19 to 63 years and each had a valid driver license. At the beginning of the drive,
participants were instructed to increase their speed as they felt comfortable to do so. They were also instructed to
change lanes to drive through the successive target arrows and to do their best to line up the center of the vehicle
with each target arrow. At the end of the session, each participant received remuneration of five dollars. Seventeen
of the 44 participants reported experiencing minor discomfort while driving. The experimental protocol (#29838-
JEN-HS) which detailed the method of recruitment, tasks to be performed by participants, possible risks and benefits
of participating, and measures taken to protect participants was approved by the CSU Institutional Review Board.
Participants were assigned to the four experimental sequences, resulting in the following distribution of sex and age
within the groups:
 Sequence 1: 8 males, 3 females, ranging from 19 to 63 years, average 30.4 years
 Sequence 2: 6 males, 5 females, ranging from 19 to 32 years, average 25.2 years
 Sequence 3: 7 males, 4 females, ranging from 21 to 57 years, average 31.5 years
 Sequence 4: 7 males, 4 females, ranging from 19 to 62 years, average 28.0 years

RESULTS AND ANALYSIS


To describe participants’ performance making lane changes, the lane position of the simulator vehicle at the moment
each target arrow was reached was used to describe the accuracy of the lane changing task and the travel time
between successive target arrows was used to describe the efficiency. Generally speaking, improved performance
would be indicated by a decrease in travel time and/or a decrease in lane position over successive lane changes.

The lane position is the distance from the center of the lane, measured in meters. A positive value indicates the
center of the vehicle was positioned to the right of the center of the lane. Conversely, a negative value indicates the
center of the vehicle was positioned to the left of the center of the lane. If the vehicle was out of the lane, the lane
position data was recorded as “-“ in the data file. This non numeric value was replaced with ±1.8 m which is the
location of the lane line.

The travel time is a measure of the efficiency of making the successive lane changes and is inversely proportional to
the average travel speed between target arrows. The faster a participant traveled the smaller the travel time.
However, when participants chose to travel too fast, control of the vehicle was risked. When the vehicle went out of
the desired travel lane additional time was spent correcting the vehicle heading and was typically accompanied by a
sharp speed drop and recovery.

Following the methodology by Sahami et al. (8), the lane position and travel time data was combined into a cost
function as given by
𝐶𝑖 = |𝐿𝑃𝑖 |(𝑡𝑖 − 𝑡𝑖−1 )
where:
Ci=the cost for lane change i
LPi = lane position of the simulator vehicle at target arrow i, meters. The sign indicates the type of error
(e.g. under steering or over steering) and is not needed in the cost calculation.
ti=time the simulator vehicle reaches target arrow i
ti-1=time the simulator vehicle reaches target arrow i-1

The cost function facilitates the evaluation of the overall performance of the participant as it describes both the
accuracy and efficiency simultaneously. Using the above formulation for the cost function, a reduction in the cost
translates to a reduction in the lane position and/or a reduction in the travel time between successive lane changes,
and therefore an improvement in performance.

For each sequence, the cost for each participant and the average cost of the participants within the group were
calculated for each lane change. The results are plotted in Figure 2 with the average cost shown as a thick line. For
each sequence, the average cost shows a general negative trend over successive lane changes. The linear trends are
as follows:

 Sequence 1: 𝑐𝑜𝑠𝑡 = −0.0105(𝑙𝑎𝑛𝑒 𝑐ℎ𝑎𝑛𝑔𝑒) + 1.58 (𝑅2 = 0.23)


 Sequence 2: 𝑐𝑜𝑠𝑡 = −0.0109(𝑙𝑎𝑛𝑒 𝑐ℎ𝑎𝑛𝑔𝑒) + 1.77 (𝑅2 = 0.16)
 Sequence 2: 𝑐𝑜𝑠𝑡 = −0.0102(𝑙𝑎𝑛𝑒 𝑐ℎ𝑎𝑛𝑔𝑒) + 2.00 (𝑅2 = 0.09)
 Sequence 2: 𝑐𝑜𝑠𝑡 = −0.0109(𝑙𝑎𝑛𝑒 𝑐ℎ𝑎𝑛𝑔𝑒) + 2.02 (𝑅2 = 0.08)

Each of the trend lines has a slope of -0.01, indicating that the groups of drivers in each of the sequences were
improving their lane changing performance. However there are large variations in the individual responses. The fit
of the linear trend to the average cost among participants within each sequence are very weak, as indicated by low
R2 values.
Figure 2. Costs for lane changes for the four experimental sequences

Comparisons of Mean Cost


The first analysis looked strictly at the differences in the mean cost to test whether 1) practice is better than no
practice; 2) practicing a less challenging but similar steering task is good practice for a more challenging steering
task; and 3) practicing a more challenging but similar steering task is good practice for a less challenging steering
task.

To test whether practice on a different steering task is better than no practice, the mean cost for the lane changes
spaced at 65 meters was compared between Sequences 1 and 2 (𝑥̅1 = 1.4, s1 = 0.393, n1 = 29, 𝑥̅2 = 1.59, s2 = 0.501,
n2 = 29). The difference was not significant at 90% (t = 1.09, p = 0.14). This means that the participants driving the
first lane change task of Sequence 1 did not perform better than the participants driving the first lane change task of
Sequence 2. Therefore, the lane keeping task driven at the beginning of Sequence 1 had no significant impact on the
performance of the subsequent lane changing task.

Similarly, the mean cost for the lane changes spaced at 85 meters was compared between Sequences 4 and 3 (𝑥̅4 =
1.812, s4 = 0.679, n4 = 29, 𝑥̅3 = 1.774, s3 = 0.453, n3 = 29). The difference was not significant at 90% (t = 0.25, p =
0.40). This means that the participants driving the first lane change task of Sequence 4 did not perform better than
the participants driving the first lane change task of Sequence 3 and therefore the lane keeping task had no
significant impact on the performance of the subsequent lane changing task.

To test whether a less challenging but similar steering task is good practice for a more challenging task, the mean
cost for the lane changes spaced 65 meters apart was compared between Sequences 3 and 2 (𝑥̅3 = 1.591, s3 = 0.679,
n3 = 29, 𝑥̅2 = 1.529, s2 = 0.501, n2 = 29). The difference was not significant at 90% (t = 0.40, p = 0.35). This means
that participants driving the second set of lane changes in Sequence 3 did not perform any better than those driving
the first set of lane changes in Sequence 2. Therefore the impact of driving the first set of lane changes spaced 85
meters apart did not have a significant impact on the participants’ performance on the next set of lane changes
spaced 65 meters apart.

To test whether a more challenging but similar steering task is good practice for a less challenging task, the mean
costs for the lane changes spaced 85 meters apart was compared between Sequences 2 and 3 (𝑥̅2 = 1.337, s2 =
0.428, n2 = 29, 𝑥̅3 = 1.774, s3 = 0.453, n3 = 29). The difference was significant at 95% (t=3.78, p=0.00). This means
that those who first drove the lane changes spaced 65 meter apart performed better when driving the lane changes
spaced 85 meter apart.

Effects of Participant Characteristics and Sequence of Steering Tasks


One of the inherent difficulties of studying the improvement in driving performance is that every task completed by
a participant has the potential to influence the performance of the subsequent task(s). Therefore, in this experiment,
participants were assigned to one of the four experimental sequences. The resulting four groups of participants vary
in their characteristic make-up. The previous comparisons of mean cost analysis neglects these differences. To
account for the individual responses of the participants within a sequence, a regression analysis was performed to
evaluate the effect of the participant sex and age, as well as the effects of the sequence of steering tasks on the
average cost data.

To test whether practice on a different steering task is better than no practice, a regression analysis was performed
on the average cost data for the lane changes spaced at 65 meters from Sequences 1 and 2. The model was not
significant (F=0.96) nor were the effects of sex (t=0.35, p=0.74), age (t=0.94, p=0.36), and sequence (t=1.61,
p=0.13). Similarly, the average cost data for the lane changes spaced at 85 meters from Sequences 4 and 3 was
analyzed. The regression model was not significant (F=1.95) nor were the effects of age (t=0.74, p=0.47) and
sequence (t=0.35, p=0.73). The effect of sex was significant (t=-1.91, p=0.07).

To test whether a less challenging but similar steering task is good practice for a more challenging task, a regression
analysis was conducted on the average cost data for the lane changes spaced 65 meters apart from Sequences 3 and
2. The model (F=0.69) and the effects of sex (t=-0.46, p=0.65), age (t=1.20, p=0.25) and sequence (t=-0.29, p=0.77)
were not significant.

To test whether a more challenging but similar steering task is good practice for a less challenging task, a regression
analysis was conducted on the average cost data for the lane changes spaced 85 meters apart from Sequences 2 and
3. The model (F=1.16) and the effects of the sex (t=0.12, p=0.91), age (t=0.43, p=0.67), and sequence (t=1.53,
p=0.15) were not significant.

CONCLUSIONS
The performance improvement for the lane changing tasks was modeled using a cost function, formulated as the
product of the absolute lane position and the travel time. The average cost for each lane change within each driving
sequence was then calculated. The cost trends within groups, shown in Figure 2, decreases over successive lane
changes. What is interesting is that the general trend is very similar for each of the sequences. This is important
finding as it illustrates the need to control for the effects of performance improvement that occurs over repeated
trials of the same or even similar task.

The average cost data was used to conduct two analyses. The first analysis compared mean cost between sequences
to address whether practicing a lane keeping task improves performance on subsequent lane changing tasks, and
whether one lane changing task is good practice for another. The results lend themselves to the notion that practicing
a more demanding lane changing task will positively impact the performance on a subsequent, less challenging lane
changing task. However, this naïve analysis approach neglects the differences within the participant groups.

The second analysis examined the effects of the differences between the participants groups which drove the
different sequences. A series of regression analyses were performed to examine the effects of sex, age and sequence.
The results were not remarkable. The effect of the sequence of the steering tasks was not significant, therefore there
is no evidence to support that practicing one steering task will impact the performance on a subsequent steering task.

ACKNOWLEDGMENTS
The authors thank the Office of Research at Cleveland State University for funding the 2014 Undergraduate
Summer Research Award entitled “Is learning to drive a simulator task dependent?”.

REFERENCES
Jenkins, J. M. and Moran, B. (2014) “Scenario for Practicing Steering in a Driving Simulator” Advances in
Transportation Studies: an international journal. 2014 Special Issue, Vol. 2, pp. 47-58.
Jenkins, J. M. and Seck, N. (2014) “Comparing two driving simulation scenarios for steering and speed control”
Advances in Transportation Studies: an international journal. Vol. XXXIII, pp. 21-32.
McGehee, D. V., Lee, J. D., Rizzo, M., Dawson, J. and Bateman, K. (2004) “Quantitative Analysis of Steering
Adaptation on a High Performance Fixed-base Driving Simulator” Transportation Research Part F, Vol. 7, pp. 181-
196.
Ronen, A., and Yair, N. (2012) “The adaptation period to a driving simulator” Transportation Research Part F, Vol.
18, pp. 94-106.
Sahami, S. and Sayed, T. (2010) “Insight into Steering Adaptation Patterns in a Driving Simulator” Transportation
Research Record: Journal of the Transportation Research Board, No. 2185, Transportation Research Board of the
National Academies, Washington, D.C., pp. 33-39.
Sahami, S. and Sayed, T. (2013) “How drivers adapt to drive in driving simulator, and what is the impact of practice
scenario on the research?” Transportation Research Part F, Vol. 16, p. 41-52.

View publication stats

Potrebbero piacerti anche