Exploratory simulation of pedestriancrossings at roundabouts

R. Hughes, PhD(1), N. Rouphail(2)* , PhD and Kosok Chae(2)

(1)The University of North Carolina Highway Safety Research Center

(2)North Carolina State UniversityInstitute for Transportation Research and Education

Submitted for consideration for publication in theJournal of Transportation EngineeringAmerican Society of Civil Engineers

March 2003

• Corresponding author, ITRE, NC State University Campus Box 8601, Raleigh, NC 27695-8601;phone 919-515-1154; fax 919-515-8898; rouphail@eos.ncsu.edu

1

Introduction

Despite the widespread adoption of roundabouts in continental Europe, the United

Kingdom, and Australia (NAASRA 1986; 1997; Department of Transport, United

Kingdom 1993), reaction to their installation in the United States remains mixed (ITE

Roundabout Accessibility Summit, 2002). By sharply reducing the potential for high-

speed angle crashes, roundabout designs generally result in fewer serious injury

crashes and fatalities (FHWA, June 2000).

While roundabouts appear to be safer for the operation of motorized vehicles, the

evidence for a similar safety effect in the case of pedestrians is unclear (Persaud, et al.,

2001, Brüde and Larsson, 2000). Where data are available on pedestrian exposure,

pedestrian volumes at roundabouts are generally too low to provide statistical

confidence regarding their safety performance.

While the safety of roundabouts for pedestrians will continue to be ‘inferred,’ rather than

‘observed’ (e.g. NHTSA 1999; Retting, 2002), there is presently much debate over the

issue of pedestrian access, particularly for pedestrians who are blind or function with

low vision (Guth et al. 2002; Long, et al., 2002; Access Board Bulletin 2002). Access in

this context refers to the pedestrian’s ability to (a) locate the crosswalk, (b) correctly

orient the direction of the crosswalk, (c) determine when it is safe or permissible to

cross, and (d) have sufficient time to cross. In fact, access for blind pedestrians and

those with low vision is the focus of current US Access Board draft recommendations

which propose that signalization is a necessary condition to ensure access for blind

pedestrians.

1105.6.2 Signals. A pedestrian activated traffic signal complying with 1106 shall be provided for

each segment of the crosswalk, including the splitter island. Signals shall clearly identify which

crosswalk segment the signal serves

The Access Board recommendation runs counter to the basic engineering premise of

effective roundabout operation, in which traffic should proceed uninterrupted, yielding as

needed to pedestrians at crossings, and to vehicles in the circulatory roadway. The

2

Access Board recommendation is under review and has not yet proceeded to final rule

making.

The present research was motivated by the need to address some of these critical

issues. The study was aimed at documenting, through exploratory computer modeling,

the expected nature of blind versus sighted pedestrian crossing performance at

roundabouts, and to gather preliminary data (via modeling) on alternative signalization

concepts and their potential impact on traffic performance in roundabouts.

This study is unique in that it represents a first attempt at the explicit modeling of

pedestrian vehicle interactions at roundabouts in a micro-simulation traffic environment.

The study uses observational data of gap perception behavior by blind pedestrians for

incorporation into the logic of the model. Finally, the study for the first time explores

some signalization alternatives for pedestrian crossings near roundabouts and tests

their impact on roundabout system operation.

Methodology

The methodology used in this study consisted of three steps.

1) Selection of an appropriate simulation tool

2) Selection and coding of a test roundabout incorporating observational data ofactual pedestrian gap perception behavior

3) Conduct of modeling experiments related to the differential performance ofsighted and blind pedestrians at roundabouts, and the evaluation of alternativesignalization schemes

It is important to note the limitations of the present study. While the observational data

provided for real world calibration of pedestrian gap perception parameters, most other

model parameters were kept at their default values. Of course, model input data such

as volumes, turning movements, speed limits, roundabout geometry, etc. were taken

from an actual test site, and therefore are representative of field operation. The reader

should therefore be aware of the exploratory nature of this work, and of the need for

3

formal model calibration and validation studies before definitive solutions to this problem

can be proposed and tested.

Selection of Simulation Tool

Several roundabout analytical and simulation tools were reviewed including aaSIDRA

(2002), Paramics (2000) and VISSIM (2001). Because the focus of this study was on

pedestrian vehicle interaction, it became clear very early on that VISSIM provided the

best platform to achieve the study objectives (see Rouphail, et al., 2002). Other models

either did not have the ability to explicitly model pedestrian movements, or required

extensive coding to incorporate necessary pedestrian performance attributes.

VISSIM is a microscopic, time-step behavior-based model. It is multi-modal in scope,

comprising entities such as drivers, pedestrians, vehicles, and a road network. Model

interactions among all users can be represented, and the network performance varies

depending on user behavior, system status, and time. VISSIM also tracks each

individual vehicle type including autos, trucks, buses, rail, pedestrians, and bicyclists at

designated data collection points.

VISSIM consists of three major components –an input module, a simulator, and an

output module. The input module is a Windows-based user interface. The simulator

(processor) is used for generating and moving traffic, updating system status, and

collecting statistics. The output module typically produces animation movie files (in “avi”

format) and text output.

Incorporating Observational Data on Pedestrian Gap Selection

For the present study, gap selection attributes of blind and sighted pedestrians at

roundabouts were derived from field data collected at three operational roundabouts in

the Baltimore, Maryland metropolitan area (Towson, Annapolis, and The University of

Maryland Baltimore Campus-UMBC). The methodology by which these data were

collected is described in Guth, et al (2002). These data were collected as part of a grant

awarded by the National Eye Institute (NEI) of the National Institutes of Health (NIH).

4

The focus of the grant was on problems experienced by blind pedestrians and those

with low vision at complex intersections.

For safety reasons, experimental subjects in the work reported here did not actually

cross the street, but rather, from a stationary position on the curb, made judgments

about when they perceived it was safe to cross. In the modeling work reported here,

data on the (perceived) gap selection attributes of blind and sighted pedestrians were

taken from the single-lane UMBC roundabout data reported by Guth, et al. (2002).

Initially, estimates of the perceived critical gaps for sighted and blind pedestrians were

calculated, based on the experimental data. Each experiment reported in the work of

Guth, et al. (2002) consisted of recording, over a period of two minutes the size of the

perceived accepted and rejected gaps for individual subjects. Multiple observations

were collected from 6 blind and 4 sighted subjects who provided responses on when

they perceived it was ‘safe’ to cross. The Maximum Likelihood Estimation (MLE) method

described by Troutbeck (1992) was used to estimate the perceived critical gaps.

A closer inspection of the Guth, et al. (2002) data revealed that gap observations were

made under extremely low volume conditions. For example, the mean observed gap

size at the entry leg during the Guth, et al. experiments was 25.5 sec (equivalent to 141

vehicles per hour), while that at the exit leg was 32.5 sec (equivalent to 111 vehicles per

hour). Under such conditions, it is difficult to assert a reliable value of perceived critical

gaps given the very large size of the perceived accepted gaps. Indeed, the estimated

perceived critical gaps using the MLE method were 7.75 seconds for sighted

pedestrians, and 8.75 seconds for blind pedestrians. Compared with the required

crossing time of five seconds, this indicates a rather large safety margin, which may be

simply a reflection of the (large) size of the gaps that were available during the

experiment.

Given the difficulties in directly estimating the critical gap, an alternative approach was

then adopted. In this approach, the critical gap was estimated as the sum of latency and

crossing time. Latency was measured from the time a vehicle passed in front of the

subject to the time the subject indicated he/she thought it was safe to cross. The same

5

crossing speed of 4 ft/sec was assumed for blind and sighted pedestrians since the

literature indicates that people walking with a cane walk at about 85% the speed of a

sighted walker while people with guide dogs walk at 105 to 110% the speed of a sighted

walker (Reference will be added).

Table 1 gives the mean, median, and standard deviation of latency times recorded for

sighted and blind pedestrians at the three roundabouts. It should be cautioned that the

measure of ‘latency’ in the Guth, et al. study is not equivalent to the conventional traffic

engineering measure of ‘critical gap.’

Table 1. Perceived Latency Times (f) at three Roundabouts (sec)Towson Annapolis UMBC

Entry Exit Entry Exit Entry Exit

Blind Sighted Blind Sighted Blind Sighted Blind Sighted Blind Sighted Blind Sighted

Mean 3.40 1.37 4.56 1.97 4.99 1.36 4.43 1.96 3.37 1.58 5.44 2.03

SD 2.51 1.42 2.35 1.99 3.33 1.71 3.21 1.95 2.46 1.66 3.80 2.72

Median 3.00 1.00 4.00 1.00 4.00 1.00 4.00 1.00 3.00 1.00 4.00 1.00

Longer latencies for blind pedestrians reflect the fact that their perceptions of gaps in

traffic rely on their ability to reliably ‘hear’ the sounds of approaching and departing

vehicles, whereas the perceptions of sighted pedestrians are made primarily on the

basis of visual cues. Thus, blind pedestrians may miss those gaps which would be

sufficient in length to cross, but which cannot be reliably detected due either to their

inability to detect an (auditory) event that defines the end of the gap or their inability to

detect a ‘quiet’ period that defines a safe crossing opportunity. In addition, the results

depict significant differences in latency times observed at the entrance and exit sides of

the of the roundabout

The minimum perceived acceptable gap, assuming that the pedestrian has a latency

time ‘f’, to cross a distance of ‘w’, at nominal crossing speed ‘s’ can be computed as:

)1....(..........fswGapMinimumPerceived +=

6

Because of the large standard deviation in the observations compared to the mean

value, the research team opted for the use of the median value of ‘f’ in Table 1 for

modeling purposes, as the median value is more “resistant” to extreme values on either

side of the distribution.

Equation (1) provides a rough estimate of the critical gap. On the one hand, the actual

latency time could be slightly overestimated, since the pedestrian subjects had long

gaps to choose from during the experiments. On the other hand, the use of crossing

time without assuming any safety buffer following the completion of the crossing may

slightly underestimate the correct acceptable lag. These two errors are likely to cancel

each other in the final determination of the critical gap.

Modeling of Test Roundabout

For modeling purposes, a single lane roundabout located at the intersection of Pullen

and Stinson (PS) roads on the NC State University campus in Raleigh, NC was selected

and is shown in Figure 1. Links and connectors as defined in the VISSIM model have

been superimposed on an aerial photo of the roundabout. This particular roundabout is

classified as a compact urban roundabout (FHWA 2000) with an approach design

speed of 17 mph and an inscribed diameter of 88 ft. Operational aspects of the

roundabout in the model were based upon actual pedestrian and vehicle counts

provided by the NC Department of Transportation and its consultant. The noon peak

hour, intended to capture the maximum pedestrian flow was simulated in VISSIM.

Using Equation 1 and assuming a crossing speed of 4 ft/sec for all pedestrians (see

previous section for justification), the minimum perceived gaps for sighted and blind

pedestrians were calculated for the PS roundabout and summarized in Table 2.

7

Figure 1. VISSIM rendition of PS roundabout

Table 2. Modeled minimum crossing times and perceived gaps (sec)

* Modeled approach ** B = Blind, S = Sighted

Figure 2 depicts how the pedestrian-vehicle interaction was modeled in VISSIM for an

entry approach to the roundabout. The inclusion of the latency time was modeled as a

clear distance requirement downstream of the crossing point. The lag time was also

modeled as a clear distance time lag requirement before a pedestrian could proceed to

cross (measured upstream of the crossing point). With the time criterion, an estimate of

the approach speed must be used to locate the upstream vehicle yield line. Additional

details of how the pedestrian/vehicle interaction was modeled in VISSIM can be found

in Rouphail, et al. (2002).

Pullen North* Pullen South Stinson

Entry Exit Entry Exit Entry ExitParameter

B** S** B S B S B S B S B S

Crossing Distance 17.2 ft 15.9 ft 17.1 ft 16.7 ft 15.2 ft 15.6 ft

Crossing Time 4.30 3.98 4.28 4.18 3.80 3.90

Median Latency 3 1 4 1 3 1 4 1 3 1 4 1

Min PerceivedGap

7.3 5.3 8.0 5.0 7.3 5.3 8.2 5.2 6.8 4.8 7.9 4.9

N

N

8

Figure 2. VISSIM rendition of pedestrian-vehicle interaction at crossing

Two “Experiments”

Two sets of simulation experiments were performed in VISSIM. The first was designed

to contrast delay times for sighted and blind pedestrians as a function of approach and

exiting traffic volumes, based on their varying gap acceptance behavior. The motivation

behind this experiment was the desire to compare the relative access of sighted and

blind pedestrians to crossing opportunities at the splitter island. In the second

experiment, the effects of placing a pedestrian-activated signal alternatively at (a) the

crosswalk at the splitter island, and (b) at a location downstream from the exit lane were

tested. The signal offset distance for case (b) was approximately 100 ft, which provided

queue storage for about 4-5 vehicles.

LatencyLag

9

RESULTS

Experiment I

Starting with the base volume model with pedestrian volumes of 40 per hour on the

Southbound Pullen approach, a sensitivity analysis was conducted to investigate the

relationship between pedestrian delay for blind and sighted pedestrians and vehicle

volume. This analysis was performed for a range of vehicle volumes from 60% to 140%

of the base noon counts (which are shown in Table 3). The largest vehicle volumes

result in a v/c ratio of about 70% for the tested approach. In VISSIM, delay is incurred

whenever the speed of the vehicle or pedestrian is below the desired speed value. To

extract pedestrian delay, four entry areas and four exit areas were defined in VISSIM.

Each area included half the crosswalk from the near curb to the splitter island and from

the splitter island to the far curb. A delay counter started when a pedestrian arrived at

the curb or splitter island, and was stopped when the pedestrian began crossing to the

splitter island, or the far curb.

Table 3. Base vehicle OD counts and conflicting volumes with pedestrians

Conflict veh volumesO/D NB Pullen SB Pullen EB Park WB Stinson Entry ExitSB Pullen* 0 535 5 48 588 617NB Pullen 567 0 4 101 672 583WB Park 5 4 0 1 10 10

EB Stinson 45 44 1 0 90 150* Shaded approach used for testing the interaction between pedestrians and vehicles.

Trend lines and pedestrian delay estimates generated from VISSIM are shown in

Figures 3, 4, and 5. Figure 3 shows the full crossing delay from curb to curb for blind

and sighted pedestrians, respectively. The delay to blind pedestrians is, as expected,

higher than that for sighted pedestrians. Figure 4 and 5 show the results for the entry

and exit legs respectively. The maximum delay was estimated at 60 sec for blind

pedestrians at the exit lane. Compared with the corresponding sighted pedestrian delay

value of 20 sec, delay increased by 200%. As a result of their smaller accepted gaps, it

is not surprising to note that delays for sighted pedestrians were less than those

experienced by blind/low vision pedestrians. Moreover, there appear to be small

10

differences in pedestrian delay between the exit and entry legs for sighted pedestrians

(since the difference of their minimum gaps was smaller (0.3 sec)), but a larger

difference for blind/low vision pedestrians at the two locations (where minimum gaps

differed by 0.7 seconds). Vehicle delay was found to be largely unaffected by

pedestrians in our VISSIM simulations. Vehicle delay values are not shown in the figure,

but the results indicate that for the given (low) pedestrian volumes, the largest

determinant of vehicle delay was that incurred at the roundabout yield line.

Figure 3. Simulated full crossing (curb to curb) delaysfor sighted and blind pedestrians

Pedestrian Delay

0

20

40

60

80

100

120

140

300 400 500 600 700 800 900 1000

Conflicting Vehicle Volume, VEN or VEX(vph)

Del

ay(s

ec/P

ed)

BlindSighted

Base Volume

140% Base

60% Base

11

Figure 4. Simulated blind pedestrian crossing delaysat exit and entry legs

Figure 5. Simulated sighted pedestrian crossing delaysat exit and entry legs

Blind Pedestrian Delay

0

10

20

30

40

50

60

70

80

300 400 500 600 700 800 900 1000

Conflicting Vehicle Volume, VEN or VEX(vph)

Del

ay(s

ec/P

ed)

Blind-ExitBlind-EntryTrend-line(Blind_Ex)Trend-line(Blind_En)

Base Volume

140% Base

60% Base

Sighted Pedestrian Delay

0

10

20

30

40

50

60

70

80

300 400 500 600 700 800 900 1000

Conflicting Vehicle Volume, VEN or VEX(vph)

Del

ay(s

ec/P

ed)

Sighted-ExitSighted-EntryTrend-line(Sighted_Ex)Trend-line(Sighted_En)

Base Volume

140% Base

60% Base

12

Experiment II

The results of Experiment I suggest that if equal access is to be granted to blind and

sighted pedestrians at roundabouts, some means needs to be created that serves to

more closely equate the quality of service for blind and sighted pedestrians. This can be

done one of two ways: (a) augment the cue environment of the blind pedestrian to the

point where crossable gaps are more reliably detected, or (b) force the occurrence of

more crossable gaps for the blind pedestrian. The Access Board recommendation that

roundabouts be signalized would serve to both ‘force’ a gap in the traffic stream (upon

pedestrian activation of the signal) and provide a reliable cue (e.g., through an

accessible pedestrian signal or other device) that such a gap is present. Two

signalization options are modeled in this experiment. First is the option of locating the

pedestrian signal at the splitter island. The second option is to install the signal

downstream of the roundabout, similar to the current practice in the UK. In our model,

only a single signal installation was carried out, although in reality multiple signals at the

various approaches may be considered. Two levels of pedestrian demand were

modeled. The low demand of six pedestrian actuations per hour mimics the case where

only blind pedestrians could activate the signal. The higher demand of fifty pedestrians

per hour assumes that all pedestrians could actuate the signal. Vehicle demand was

modeled at 140 percent of the base volume.

The results summarized in Table 4 show that pedestrian delay increases as a function

of pedestrian demand for signals at both the splitter island and the mid-block crosswalk

locations. Data on total pedestrian travel time are also of interest in that the use of the

mid-block crosswalk would require the additional time to walk to/from the mid-block

location from the area of the splitter island. The differences in delay between scenarios

1 versus 2 (and 3 versus 4) are caused by the fact that when actuated by pedestrians,

the vehicle signal phase must first cycle over its minimum green time, before it can

service the pedestrian phase. With only six calls per hour, it is likely that the pedestrian

phase can be served almost immediately after each call.

13

Total travel time is clearly extended under the assumption that all pedestrians would

have to walk to/from the mid-block location to effectively cross the street. However, it is

interesting to note that from a delay perspective, a blind pedestrian crossing at a

dedicated mid-block signal (under Scenario 3) experiences a delay of about 15 sec,

which is actually smaller than what a sighted pedestrian would experience at the

unsignalized crossing at the splitter island (20-35 seconds as shown in Figure 5 for the

140% Base case). With the added safety of crossing at a signal, this treatment may,

under the right set of circumstances, achieve the equal access requirement.

Table 4. Pedestrian delay and travel time for two roundabout signalization options

Perhaps a more compelling reason for considering the availability of a mid-block

crossing option is shown in a graphical comparison of the situations depicted in Figures

6 and 7. Figure 6 illustrates the effect on traffic of a pedestrian-activated signal placed

at the splitter island for just one approach at the roundabout. Figure 7 shows the effect

on traffic when the signal is placed at an upstream/downstream ‘mid-block’ location.

These images are from the ‘avi’ graphic output file of VISSIM. The model shows that

when a pedestrian-activated signal is placed at the splitter island location, a queue

forms behind vehicles attempting to exit, causing delay not only for those vehicles in the

queue but for other vehicles whose paths are blocked by vehicles in the queue. A

slightly better situation is seen in Figure 7 where a queue still forms at the signal, but

depending upon the location of the signal and the length of the queue which forms,

operations in the roundabout may not be affected by the queue.

Scenario 1 2 3 4Splitter Is. 6 50 0 0Ped

VolumeCrossingLocation Mid-Block 0 0 6 50

Entry 17.0 26.4 15.7 26.3Ped Delay(sec / ped) Exit 15.4 25.3 15.0 26.8

Entry 33.2 42.4 84.5 96.8Ped Travel Time (sec / ped) Exit 32.7 42.7 88.7 103.2

14

Figure 6. VISSIM rendition of pedestriansignal at the splitter island

Figure 7. VISSIM rendition of pedestriansignal at a mid-block location

15

Conclusions

The present investigation has shown that modeling (in this case, using VISSIM) has the

potential for enabling traffic engineers to consider the range of issues involved in

accommodating pedestrian crossings at roundabouts, and in particular, the unique

crossing requirements of those pedestrians who are blind or functioning with low vision.

Using estimates of the gap selection attributes of blind and sighted pedestrians

gathered under actual operational roundabout conditions, the output of the model

reflects the problems (in terms of pedestrian delay, or lack of access) that can be

expected by blind pedestrians. Much more work is needed to construct realistic

estimates of the pedestrian critical gaps, using observations of actual crossings by

sighted and blind pedestrians including rejected and accepted vehicular gaps under a

range of traffic volume conditions. The critical gap parameter is paramount to the

development and evaluation of the effectiveness of unsignalized pedestrian treatments

at roundabouts.

The present study’s evaluation of a hypothetical pedestrian-activated signal at the

splitter island approximates what might be the most obvious signalization treatment

implemented in response to the Access Board’s pending recommendation as to how to

improve pedestrian access. While such a treatment would always guarantee a

crossable gap for the pedestrian, it is clear that it could have a very disruptive effect on

traffic operations within the roundabout both in terms of traffic efficiency as well as in

terms of a possible increase in certain classes of collisions (e.g., rear end collisions,

sideswipes, etc.). Unfortunately, there is very little guidance in the literature on the

effectiveness of this treatment.

Use of an upstream/downstream (mid-block) pedestrian-activated signal and crosswalk

would appear to be a good compromise, inasmuch as it would guarantee a crossable

gap while minimizing any negative impact that queues formed by the signal would have

on operations in the roundabout, per se. Clearly, more work needs to be done to define

the limits of effective implementation of such a signalization concept (e.g., ped and

vehicle volumes, distance removed from the roundabout, nature of the signal and signal

16

characteristics employed, etc.). From more of a ‘policy’ standpoint, it needs to be

considered whether or not such a upstream/downstream ‘mid-block’ crossing location

would/should be the only location where it was permissible for pedestrians to cross, or

whether it should be provided as a voluntary ”alternative’ to the crosswalk located at the

splitter island.

The effective use of computer modeling in the present case suggests that modeling may

represent a viable alternative to traditional field data collection methods where subjects

are placed at risk for the sake of treatment evaluation. While modeling does not rule out

the need for eventual evaluation of effects ‘in the field,’ it does permit one to approach

operational field evaluations with the knowledge (from the model) that the treatments

being evaluated have been shown to have a high probability of success.

17

REFERENCES

aaSIDRA User Guide. (2002). Akcelik & Associates Pty Ltd.

Architectural and Transportation Barriers Compliance Board. 2002. Draft guidelines foraccessible public rights-of-way. Washington, DC. Available on line at: http://access-board.gov/rowdraft.htm#DRAFT.

Department of Transport (United Kingdom). (1993). Geometric Design of Roundabouts.TD 16/93.

Draft Guidelines for the Design of Roundabouts, (1997). Transport and TechnologyDivision of the Department of Main Roads, Queensland, Australia

Federal Highway Administration (FHWA). (2000) Roundabouts: An Information Guide.

Guth, D., Long, R., Ponchilla, P., Ashmean, D, and Wall, R. Non-visual gap detection atroundabouts by pedestrianss who are blind; A summary of the Baltimore roundaboutsstudy (submitted for publication, 2002). Available on line at: http://www.access-board.gov/publications/roundabouts/research-summary.htm

Long, R., Ponchillia, P., Guth, D., Ashmead, D., & Wall, R. (2002). Roundabouts andpedestrians with blindness and visual impairments: An issue of Information access.Association for Education and Rehabilitation of the Blind and Visually Impaired. Toronto.

Pedestrian Access to Modern Roundabouts: Design and Operational Issues forPedestrians who are Blind. US Access Board On-Line Bulletin, http://www.access-board.gov/publications/roundabouts/bulletin.htm, 2002.

Persaud, B.N., Retting, R.A., Garder, P.E., and Lord, D. Safety effects of roundaboutconversions in the United States: Empirical Bayes observational before-after study.Transportation Research Record 1751, 108, Washington, DC, Transportation ResearchBoard.

PTV AG. (2000). “VISSIM Traffic flow simulation Technical Description.” December.

QUADSTONE, (2000). Paramics-online v3

Retting, R. Insurance Institute for Highway Safety. October 23, 2002 letter toArchitectural and Transportation Barriers Compliance Board, Washington, DC. Availableon line at: http://www.hwysafety.org/fed/otis_rar_102302.pdf.

Roundabouts: A Design Guide, National Association of Australian State RoadAuthorities (1986).

18

Rouphail, N., Wan, B., Chae, K., Hughes, R., and Harkey, D. (2002). Evaluation andApplication of Pedestrian Modeling Capabilities Using Computer Simulation. Universityof North Carolina at Chapel Hill and NC State University, Prepared for WesternMichigan University and National Institutes of Health/National Eye Institute.

Troutbeck, R.J. (1992), Estimating the critical acceptance gap from traffic movements,Physical Infrastructure Centre Research Report 92-5, Queensland University ofTechnology, Brisbane Australia

Ulf Brüde and Jörgen Larsson, (2000). What roundabout design provides the highestpossible safety? Swedish National Road & Transport Research Institute, Nordic Road& Transport Research Report No. 2 2000.