Advancement of The Car Following Model of Wiedemann On PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Transportation Research Part F xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Transportation Research Part F


journal homepage: www.elsevier.com/locate/trf

Advancement of the car following model of Wiedemann on


lower velocity ranges for urban traffic simulation
Marcus Mai a,⇑, Lei Wang b, Günther Prokop a
a
Dresden University of Technology, Dresden Institute of Automobile Engineering – IAD, Chair of Automobile Engineering, Dresden DE 01062, Germany
b
BMW Group, Knorrstraße 147, Munich DE 80788, Germany

a r t i c l e i n f o a b s t r a c t

Article history: In the present paper, a theory-based advancement of the car following model of
Received 23 December 2016 Wiedemann is introduced. It is suitable for a more realistic simulation of human longitu-
Received in revised form 25 May 2017 dinal guidance behaviour in the lower velocity range, like in urban traffic. The algorithm
Accepted 29 August 2017
was identified and developed by means of a driving simulator study, which is introduced
Available online xxxx
as well. While the decision making process in recent car following models based on
Wiedemann is solely determined by the time headway to the leading vehicle, the results
Keywords:
of the study show that information about time to collision is also used in the lower velocity
Driver behaviour modelling
Car following behaviour
range. This approach delivers a more realistic simulation of human car following beha-
Tau-theory viour, which can be integrated into driver models for microscopic traffic simulations and
for the control of surrounding traffic in driving simulators.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Algorithms for the simulation of human car following behaviour are essential parts of driver models, to control the major
part of longitudinal traffic flow in microscopic traffic simulations. Those simulations are also applicable for the control of the
surrounding traffic within driving simulators. For instance, PELOPS (Programm zur Entwicklung längsdynamischer,
mikroskopischer Prozesse in systemrelevanter Umgebung – Program for the development of longitudinal dynamic, micro-
scopic processes within system relevant environment; Christen & Huang, 2008) was developed as a microscopic traffic sim-
ulation tool and contains a driver model, which was applied for this purpose by Hochstädter, Zahn, and Breuer (2000).
Many of the present algorithms of human car following behaviour are based on the car following model of Wiedemann
(1974), which was developed and validated for motorway traffic and is therefore only applicable in a limited way for lower
velocities like in urban traffic. In most cases, a new parametrization of the existing model has to be accomplished.
In the course of the development of a new driver behaviour model for the evaluation of automated driving functions by
means of stochastic traffic simulation (Mai, Wang, Helmer, & Prokop, 2015), the phenomenological approach of the advanced
Wiedemann model from Christen and Huang (2008) was remodelled. A new approach for the decision making process in the
lower velocity range was established, based on the results of a driving simulator study. This approach is well anchored in
human physiology of perception and has been proved to be more stable and universal. It will be introduced and discussed
in the present paper.

⇑ Corresponding author.
E-mail address: marcus.mai@tu-dresden.de (M. Mai).

http://dx.doi.org/10.1016/j.trf.2017.08.014
1369-8478/Ó 2017 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
2 M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx

2. Research question

Within the car-following model of Christen and Huang (2008), the decisions about the longitudinal acceleration of the
own vehicle (Index 2) depend on the absolute velocity v2, the relative distance Dx (x1 – x2), and the relative velocity Dv
(v2 – v1), regarding the leading vehicle (Index 1). A sectional view of this three dimensional decision diagram (neglecting
the velocity v2 for the sake of an easier understanding) is presented in Fig. 1. The grey curve illustrates the approaching
and following process regarding the leading vehicle, which is explained subsequently:
The driver approaches the leading vehicle by a constant relative velocity Dv until the influence distance is reached (a). He
then releases the accelerator and uses the powertrain drag torque to decelerate until the minimum following distance is
reached (b). This initiates the actuation of the brake and the relative velocity Dv is reduced to zero. Due to perception thresh-
olds and movement delays, the relative velocity Dv is further reduced, so the relative distance Dx increases again (the driver
falls back behind the leading vehicle; c). Before the maximum following distance is reached again, the driver accelerates (d)
and the following task is initiated (e), which consists of regulating the relative distance Dx between maximum and minimum
following distance.
As Fig. 1 shows, the driver’s decisions, whether he feels influenced by the leading vehicle, or by what distances he follows
this vehicle, are independent from the relative velocity Dv, as long as Dv can be perceived by the driver (the curve under (a),
which divides the uninfluenced field from the approaching field is considered to be a perception threshold for the relative
velocity Dv). The complete decision diagram from Christen and Huang (2008) is presented in Fig. 2. It shows that the char-
acteristic distances are solely dependent from the absolute velocity v2 of the driver’s own vehicle. The influence distance,
maximum following distance, and minimum following distance can therefore be described in terms of the time headway
thea:
Dx
thea ¼ ð1Þ
v2
According to these definitions, drivers must show specific reactions, when the time headway thea meets specific thresh-
olds, independently from the relative velocity Dv. The transferability of this mechanism to real human drivers was investi-
gated in a driving simulator study, to find possible solutions for the lower velocity range.

3. Methodology

The data used for research was collected by Rehbein (2013), performing a driving simulator study of approaches on lead-
ing vehicles. He used the static driving simulator of the Chair of Traffic and Transportation Psychology at Dresden University
of Technology (see left side of Fig. 3). The visualization of the driving scene was projected on three white walls, describing a
120°-field of view. All data was gathered with a measuring frequency of 10 Hz. Information about the human subject sample
is given in Table 1.
The independent variables were the driver’s own velocity v2 (50, 100, and 130 km/h) and the relative velocity Dv (20 and
40 km/h) at the approach onset. As stated in Section 2, the characteristic distances are only independent from the relative
velocity Dv as long as Dv is perceivable by the driver in a suitable manner. Therefore, the relative velocities Dv investigated

Fig. 1. Sectional view of the decision diagram of the car following model from Christen and Huang (2008), v2 = constant.

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx 3

Fig. 2. Overall decision diagram of the car following model from Christen and Huang (2008).

Fig. 3. Left: Driving simulator of the Chair of Traffic and Transportation Psychology; Right: Visualisation of the simulated scenario from Rehbein (2013),
motorway condition, v2 = 130 km/h.

Table 1
Characteristics of the subject sample from Rehbein (2013).

Number of participants 53 (17 female/36 male)


Age distribution 24.1 a ± 2.3 a
Frequency of driving a passenger car
<1/month 1/month 1/week >1/week daily
20.8 % 11.3 % 28.3 % 30.2 % 9.4 %
Mean mileage within the last three month
1124 km

in this study were rather high. The influence distance was operationalized by the release of the accelerator and the minimum
following distance was operationalized by the first actuation of the brake. The possibility of overtaking the leading vehicle
was prevented by closing the adjacent lane, as presented in the right side of Fig. 3.
Besides the time headway thea, also the time to collision TTC was evaluated:
Dx
TTC ¼ ð2Þ
Dv
According to the s-Theory of Lee (1976), human drivers are most likely able to perceive these time related values through
the change in retinal projections of the leading vehicle (TTC) or the visual flow field around the leading vehicle (thea).
To verify, if decisions were made independently from the relative velocity Dv and based on one of the two time related
values, Mann-Whitney-U-Tests were executed between the respective Dv conditions (20 and 40 km/h). This test was
selected, because the gathered data cannot be regarded as normally distributed, as can be seen in Figs. 7 to 12. The following
hypothesis was formulated:
‘‘If no significant difference between the two relative velocity conditions exists, the decisions are independent from the
relative velocity Dv and based on the time related value under consideration.”

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
4 M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx

According to Bortz and Schuster (2010, p. 128), the level of significance was set to 0.100 to prevent beta errors. The results
of this analysis are reported subsequently.

4. Results

In the upper velocity range (v2  100 km/h), the decisions on releasing the accelerator (influence distance) and on acti-
vating the brake (minimum following distance) seem to become independent from the relative velocity Dv, if the time head-
way thea is regarded. This is illustrated in Figs. 4 and 5.
However, the Mann-Whitney-U-Tests show highly significant differences for time headway thea over relative velocity Dv
in all four conditions shown in Figs. 4 and 5 (p < 0.001 in each case). Therefore, the Wiedemann mechanisms explained above
cannot be confirmed for the upper velocity range. Regarding the comparatively small standard deviations of the time head-
way thea, it seems very likely that participants indeed made their decisions based on this value, but not independently from
Dv.
In the lower velocity range (v2 = 50 km/h), neither TTC nor time headway thea were able to reliably represent decision
thresholds, which were independent from the relative velocity Dv. This is illustrated in Fig. 6. However, an interesting mech-
anism emerges, if the mean value (defined as smean) of time headway thea and TTC is considered:
TTC þ thea
smean ¼ ð3Þ
2
As it can be seen, smean emerges as a decision threshold for accelerator release (influence distance) and brake activation
(minimum following distance), independently from the relative velocity Dv. The Mann-Whitney-U-Tests confirm that by
indicating almost no differences between the two Dv conditions (p = 0.330 for influence distance and p = 0.476 for minimum
following distance).
For a deeper insight into the study’s results, the distributions of the six behavioural variables under all investigated con-
ditions are given in Figs. 7–12 in conclusion.

5. Discussion

The results of the driving simulator study from Rehbein (2013) suggest that human drivers use information about own
motion (time headway thea) as well as relative motion (TTC) in equal parts, while driving with lower velocity. As velocity
increases, decisions tend to be exclusively made based on information about own motion (time headway thea). There are
two possible, overlapped mechanisms, which could account for this phenomenon:

1. As velocity v2 increases, the stimulus intensity of the visual flow field increases as well, whereas the stimulus intensity of
the relative motion remains constant. Hence, the parallel perception of both stimuli is impeded to the eye and the better
available information is used solely.

Fig. 4. Sample mean values for influence and minimum following distances from Rehbein (2013), v2 = 130 km/h.

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx 5

Fig. 5. Sample mean values for influence and minimum following distances from Rehbein (2013), v2 = 100 km/h.

Fig. 6. Sample mean values for influence and minimum following distances from Rehbein (2013), v2 = 50 km/h.

2. As velocity v2 increases and time headways thea don’t decrease equally, the relative distances Dx also increase (see Eq. (1)).
According to Lee (1976), the changes in retinal projection of the leading vehicle become more difficult to perceive, so the
dominance of visual flow is further amplified.

That the decisions were most likely based solely on the time headway thea in the upper velocity range, but were not inde-
pendent from the relative velocity Dv, is not necessarily contradicting the Wiedemann mechanisms from Christen and
Huang (2008).
On the one hand, the perception based theory mentioned above describes a transitional process, whereby smean as deci-
sion threshold is increasingly replaced by time headway thea, because the TTC as part of smean becomes further difficult to
perceive with increasing velocity v2. This does not necessarily mean that the TTC is not considered anymore at 100 or
130 km/h. Due to a rather rough discretisation of the velocity v2 in the study of Rehbein (2013), the boundary of the velocity
v2 cannot be designated, were decisions become solely possible based on time headway thea.
On the other hand, the study set up itself possibly inflicted some difficulties in perception at higher velocities, as the study
was accomplished by means of a driving simulator. It is well known that visual perception can be quite different in a driving
simulator than in real world driving, due to spatial resolution, time delays, aliasing, and brightness. For instance, the Curry,

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
6 M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx

Fig. 7. All distributions of the TTC for influence distance from Rehbein (2013).

Fig. 8. All distributions of the time headway thea for influence distance from Rehbein (2013).

Fig. 9. All distributions of smean for influence distance from Rehbein (2013).

Greenberg, and Kiefer (2005) compared drivers’ decision making in critical situations between real world driving experi-
ments and a driving simulator study. One result was that the perception of smaller relative velocities Dv is challenging in
driving simulators. Compared to the absolute velocity v2, the investigated relative velocities Dv of 20 and 40 km/h are smal-

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx 7

Fig. 10. All distributions of the TTC for minimum following distance from Rehbein (2013).

Fig. 11. All distributions of the time headway thea for minimum following distance from Rehbein (2013).

Fig. 12. All distributions of smean for minimum following distance from Rehbein (2013).

ler in the upper velocity range (v2  100 km/h) than in the lower velocity range (v2 = 50 km/h). This might have led to mis-
judgements in the way that the approach on the leading vehicle itself was systematically recognized to late, especially at
relative velocities Dv of 20 km/h.

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014
8 M. Mai et al. / Transportation Research Part F xxx (2017) xxx–xxx

Both of these suggestions are only confirmable within a real world driving experiment, using a more detailed discretisa-
tion of the independent variables. The execution of such a study is still pending.

6. Conclusion

The presented approach constitutes a substantial advancement for the decision making process of the Wiedemann model
from Christen and Huang (2008). Thereby, the simulation of car following behaviour becomes possible along a wide range of
velocity, with no need of reparameterizing the underlying algorithms. The approach can be integrated into driver models for
microscopic traffic simulations and for the control of surrounding traffic in driving simulators.
As the presented survey was a driving simulator study, which may inflict some complications on the visual perception of
time related values as time headway thea and TTC at higher velocities, a real world study will be performed medium-termly.
The objective of this study is the verification of the Wiedemann decision mechanism according to Christen and Huang (2008)
for higher velocities, because no final assessment of that matter is possible at the moment. Also, the postulated theory about
smean as a decision threshold on lower velocities is to be verified in this real world study, so the developed mechanism can be
validated completely.

Acknowledgements

The student research project from Rehbein (2013) was accomplished and supervised by the Chair of Traffic and Trans-
portation Psychology of the Dresden University of Technology. The authors would like to thank particularly Bernhard Schlag,
Gert Weller, Fanny Schreiber, and of cause Paul Rehbein for their close and very productive cooperation.

References

Bortz, J., & Schuster, C. (2010). Statistics for human and social scientists. Heidelberg: Springer (in German).
Christen, F., & Huang, Q. (2008). The driver model of the traffic flow simulation program PELOPS – modelling and possibilities of application.. In Proceedings
of the second berlin conference on driver modelling, Berlin, Germany (pp. 128–144) (in German).
Curry, R. C., Greenberg, J. A., & Kiefer, R. J. (2005). NADS versus CAMP closed-course comparison examining ‘‘Last-Second” braking and steering maneuvers
under various kinematic conditions. NHTSA Research report (DOT HS 809 925). Washington, D.C.
Hochstädter, A., Zahn, P., & Breuer, K. (2000). A universal driver model with the application examples traffic simulation and driving simulator.. In Proceedings
of the ninth Aachen colloquium on vehicle and engine technology, Aachen, Germany (pp. 1343–1360) (in German).
Lee, D. N. (1976). A theory of visual control of braking based on information about time-to-collision. Perception, 5, 437–459.
Mai, M., Wang, L., Helmer, T., & Prokop, G. (2015). Numerical driver behaviour model for stochastic traffic simulation for the evaluation of driver assistance
systems and automated driving functions.. In Proceedings of the seventh conference on driver assistance, Munich, Germany. (in German).
Rehbein, P. (2013). Braking behaviour of passenger car drivers in a driving simulator study. Student research project, Chair of Traffic and Transportation
Psychology, Dresden University of Technology, Germany (in German).
Wiedemann, R. (1974). Simulation of traffic flow. Habilitation. Germany: University of Karlsruhe (in German).

Please cite this article in press as: Mai, M., et al. Advancement of the car following model of Wiedemann on lower velocity ranges for urban
traffic simulation. Transportation Research Part F (2017), http://dx.doi.org/10.1016/j.trf.2017.08.014

You might also like