Model Contrast of Autonomous Vehicle Impacts on Traffic

Abstract

The adoption of connected and autonomous vehicles (CAVs) is in its infancy. Therefore, very little is known about their potential impacts on traffic. Meanwhile, researchers and market analysts predict a wide range of possibilities about their potential benefits and the timing of their deployments. Planners traditionally use various types of travel demand models to forecast future traffic conditions. However, such models do not yet integrate any expected impacts from CAV deployments. Consequently, many long-range transportation plans do not yet account for their eventual deployment. To address some of these uncertainties, this work modified an existing model for Madison, Wisconsin. To compare outcomes, the authors used identical parameter changes and simulation scenarios for a model of Gainesville, Florida. Both models show that with increasing levels of CAV deployment, both the vehicle miles traveled and the average congestion speed will increase. However, there are some important exceptions due to differences in the road network layout, geospatial features, sociodemographic factors, land-use, and access to transit.

Full text

Research Article
Model Contrast of Autonomous Vehicle Impacts on Traffic
Derek Hungness
1
and Raj Bridgelall
2
1
SRF Consulting Group, 6720 Frank Lloyd Wright Avenue, Suite 100, Middleton, WI 53562, USA
2
Department of Transportation, Logistics and Finance, North Dakota State University, P.O. Box 6050, Fargo, ND 58108, USA
Correspondence should be addressed to Raj Bridgelall; raj@bridgelall.com
Received 23 February 2020; Revised 9 July 2020; Accepted 28 July 2020; Published 14 August 2020
Academic Editor: Giulio E. Cantarella
Copyright ©2020 Derek Hungness and Raj Bridgelall. is is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
e adoption of connected and autonomous vehicles (CAVs) is in its infancy. erefore, very little is known about their potential
impacts on traffic. Meanwhile, researchers and market analysts predict a wide range of possibilities about their potential benefits
and the timing of their deployments. Planners traditionally use various types of travel demand models to forecast future traffic
conditions. However, such models do not yet integrate any expected impacts from CAV deployments. Consequently, many long-
range transportation plans do not yet account for their eventual deployment. To address some of these uncertainties, this work
modified an existing model for Madison, Wisconsin. To compare outcomes, the authors used identical parameter changes and
simulation scenarios for a model of Gainesville, Florida. Both models show that with increasing levels of CAV deployment, both
the vehicle miles traveled and the average congestion speed will increase. However, there are some important exceptions due to
differences in the road network layout, geospatial features, sociodemographic factors, land-use, and access to transit.
1. Introduction
Transportation demand modeling has advanced in many
ways in recent years, including activity-based modeling
and simulation modeling [1]. However, the fundamental
approach of the traditional four-step model has remained
the same over the past few decades. While various
agencies have made enhancements to submodel struc-
tures, the basic sequence of trip generation, trip distri-
bution, mode split, and traffic assignment remain
unchanged. Recent work completed by the University of
Texas Center for Transportation Research presents
modifications to a four-step model in order to accom-
modate two new features: ride-hailing services and au-
tonomous vehicles (AV) [2]. e main changes developed
in this work involved splitting households into two
groups—AV households and non-AV house-
holds—setting potentially different trip generation rates
for the new AV households; allowing for AV households
to engage in longer trips due to a reduction of value of
travel time; the inclusion of ride-hailing as new mode
choice stage; and the use of different values of passenger-
car-equivalent for AV trips to represent the gains in ca-
pacities brought by the use of AVs.
Nevertheless, a void exists between practitioners’ use of
traditional travel demand forecasting models and rapidly
evolving analysis techniques required to address new vehicle
capabilities and travel services. is paper begins by rec-
ognizing those limitations and proceeds to advance the
dialog on transportation systems modeling by analyzing
pertinent impact measures within the realm of the tradi-
tional four-step travel demand modeling structure. e
authors recognize that ongoing advancements in vehicle
automation and their wireless connectivity will funda-
mentally transform transportation and land use. Autono-
mous vehicles, also known as self-driving vehicles, can
reduce the need for urban parking spaces because they can
park themselves at remote facilities or provide another ride
on demand [3]. Connected and autonomous vehicles
(CAVs) can avoid collisions by maintaining constant
communication with all vehicles within range to monitor
their trajectories. ey can help to clear incidents quickly by
relaying information to responders and downstream traffic.
erefore, CAVs could increase the capacity of existing
Hindawi
Journal of Advanced Transportation
Volume 2020, Article ID 8935692, 10 pages
https://doi.org/10.1155/2020/8935692

highways. Further capacity enhancements are achievable by
maintaining shorter gaps while traveling faster and antici-
pating traffic conditions downstream [4]. Since CAVs can
travel along narrower paths, reduced width of lanes,
shoulders, and medians can provide even greater capacity
enhancements [5]. However, there is high uncertainty about
impacts during the transition phase of adoption where
human and robot drivers must share the road.
ere is no certainty about the timing of CAV adoption.
Market forecasts suggest some deployments of CAVs will
begin by 2025 and market penetration will be as much as
15% by 2035 [6]. One study found that depending on the
levels of willingness to pay, technology cost reduction, and
policy changes, CAV adoption by 2045 could range from
24% to 87% [7]. A survey of university employees in
Memphis, Tennessee, revealed that CAV can reach full
adoption by 2050 if prices decrease at an annual rate of 15%
or 20% [8]. Another study predicts that CAVs will provide
independent mobility for affluent nondrivers sometime
between 2020 and 2030 [5]. However, Litman notes that the
most positive impacts will probably be in the 2040s to 2050s,
and the prohibition of human-driven vehicles on certain
roadways could take even longer. Overly optimistic pre-
dictions for CAV adoption may be due to a priori as-
sumptions on market assimilation of smaller digital
innovations. Even though autonomous vehicle travel is a
reality today, significant technical problems remain and
must be addressed before CAVs can be operated under all
conditions and in all areas. Inclement weather, longer life
cycle expectancies, wireless service disruptions, affordability,
and policy conflicts must be resolved. CAV critical system
failures can be fatal. Furthermore, autonomous driving may
induce additional vehicle travel which could exacerbate
traffic problems.
ere is substantial speculation about how much CAVs
will affect future travel. A comprehensive international lit-
erature review of adoption modeling predicts that CAVs
would mostly increase vehicle miles traveled (VMT) and
reduce the use of public transit, but the results are very
sensitive to model assumptions [9]. Studies generally agree
that CAVs would provide greater accessibility for nondrivers
such as the young, elderly, and disabled [10]. Researchers
argue that increased accessibility for nondrivers will lead to a
dramatic increase in VMT [11]. One study posits that greater
accessibility for nondrivers will increase travel demand by
11% [12]. By combining a technology diffusion model and a
travel demand model, another study determined that a
reduction in the value of travel time and accessibility for new
user groups could increase VMT by 3% to 9% by 2035 [13].
CAVs could also increase the propensity to travel by more
than 4% and consequently increase car usage by more than
7% [14].
Transportation network companies (TNCs) such as Uber
and Lyft have already increased services at airports, train
stations, hotels, shopping malls, and event venues [15].
When fully automated, TNC fleets have the potential to
counter the congestion consequences of more VMT because
they can move more people in fewer vehicles. In contrast, a
mode shift towards cars, a proliferation of private CAVs, and
zero occupancy while repositioning vehicles could increase
congestion.
CAVs can influence trip frequency, trip length dis-
tributions, and mode choice. However, existing models do
not reflect the behavioral impacts of CAV deployments.
erefore, without significant CAV deployment, uncer-
tainties remain, and users have little choice but to make
certain assumptions [6]. Consequently, many planners
avoid the inclusion of CAVs in long-range transportation
plans because of the various uncertainties of their impacts
[16]. Existing models of travel demand forecasting fre-
quently rely on an assumption that past trends in travel
behavior and mode choices will continue for the next few
decades with minor alterations. e long-standing prac-
tice in transportation planning has been to calibrate
models with observed data and then validate them by
comparing the outputs to existing conditions. After cal-
ibration and validation, planners integrate anticipated
sociodemographic and economic trends to predict future
outcomes.
A recent tactic to gain insights about possible CAV
impacts is to simulate various scenarios by changing certain
model parameters [17]. For example, a modification of the
Seattle, Washington, activity-based travel model to simu-
late changes in roadway capacity, value of travel time, and
parking costs showed large increases in VMT [18]. Similar
modifications to the Charlottesville, Virginia, regional
travel demand model showed that a 32% decrease in ca-
pacity during the transition years of adoption could in-
crease vehicle hours traveled (VHT) by 46%, whereas an
increase in capacity of 100% by 2040 could decrease VHT
by 13% [19]. Agent-based modeling suggests that one
shared CAV can replace approximately 11 conventional
vehicles but will add up to 10% more VMT than compa-
rable trips in nonshared vehicles [20]. e goal of this paper
is to determine the degree to which similar modifications to
regional travel demand forecasting models agree. rough
official channels, the authors obtained the Madison,
Wisconsin, and Gainesville, Florida, models from their
respective state departments of transportation. e
Gainesville model remained unchanged after acquisition.
e contribution of this work is a comparison of forecasts
from each model after applying identical parameter
changes to forecast potential impacts from CAV deploy-
ments. It is important to emphasize here that the parameter
changes needed to be the same as those made for the model
of another city to allow for direct comparison under
identical scenarios to gain and contribute insights into the
differences in outcomes.
e organization of the rest of this paper is as follows:
Section 2 discusses the methods used to modify the existing
models, the scenarios selected for simulation, and model
output parameters. Section 3 describes the regional data
used and the specific changes made to the models. Section 4
summarizes the model outputs and provides an interpre-
tation of the differences. Section 5 discusses some limitations
of the modeling approach. Section 6 reviews the approach
and provides some concluding remarks about the general
applicability and future work.
2Journal of Advanced Transportation

2. Methods
e authors selected Madison for comparison with
Gainesville because of access and familiarity with those
models. Coincidentally, the cities are comparable because of
similarities in their sizes and college town settings.
Gainesville is the largest city in the north central Florida with
a 2017 population of 132,249. Madison is the second-largest
city in Wisconsin with a 2017 population of 255,214. Col-
leges significantly influence the travel activities and economy
of cities. Gainesville is home to the University of Florida
where undergraduate enrollment in 2016 was nearly 40,000.
Madison is home to the University of Wisconsin where
undergraduate enrollment in 2016 was nearly 30,000.
Madison Metro provides public bus transit service
throughout the city and to some of its suburbs.
2.1. Current Travel Patterns. e cities of Gainesville and
Madison share similar community and transportation sys-
tem characteristics. Both cities rank within the top 20 places
in America for bicycling to work. According to the 2014
American Community Survey, 5.3% and 4.4% of Madison
and Gainesville workers, respectively, use a bicycle [21]. e
same source determined similar average travel time to work
for Madison and Gainesville residents existed, 21.6 and 20.5
minutes, respectively, while Madison’s commute times have
remained relatively stable since then; Gainesville’s have
increased to now slightly exceed Madison’s. Multiple factors
can be attributed to Gainesville’s relatively short commute
travel times, including a strong university-centric public
transit system, a compact city layout, centrally located
employment centers, and network traffic loads spread
throughout the day. A gridded street system helps to evenly
distribute traffic, and use of coordinated and active traffic
management systems also helps to avoid incidents that
would otherwise increase congestion problems [22]. Con-
versely, Madison’s concentration of employment and eco-
nomic activity is on the downtown isthmus. e isthmus is a
constrained geography lying between two lakes which
supports high-density office and residential buildings, the
state’s capitol, the University of Wisconsin campus, and
long-established single-family neighborhoods. is leads to
a key transportation challenge, the limited number of east-
west connecting arterials through the city [23].
2.2. Scenario Modeling. Modeling the impacts of CAV
technology on trip-making requires assumptions on market
adoption and penetration rates. is has led to developing
representative impacts expected to be caused by CAV in-
tegration. In travel demand forecasting models, this is done
by altering factors like roadway capacities, vehicle operating
costs, mode choice parameters, and origin and destination
access and egress times. In November 2017, the U.S. Federal
Highway Administration (FHWA) conducted a workshop
that defined six CAV scenarios that could occur by 2035
[24]. ese scenarios could serve as an early standard for
comparing the impacts of CAV adoption across the country.
e draft scenarios are as follows:
(1) Slow Roll (Slow). Minimum plausible change, adopts
currently available technologies and investments
already in motion
(2) Niche Service Growth (Niche). Proliferation of CAV
services in niche zones such as retirement com-
munities, campuses, transit corridors, urban cores,
and ports
(3) Ultimate Traveler Assist (Assist). State transportation
agencies aggressively manage congestion with con-
nected vehicle technology adopted at 85% by 2035,
but autonomous vehicle adoption stagnates
(4) Managed Lane Network (Managed). 50% to 60% of
cars and 75% of trucks platoon on lanes designated
for automated vehicles
(5) Competing Fleets (Robo-Mix). VMT doubles because
of induced demand and empty vehicle reposition as
automated TNC fleets proliferate noncooperatively
(6) Robo-Transit (Robo-PPP). Public-private-partner-
ships (PPP), cooperative data sharing, policies, and
infrastructure modification results in a proliferation
of automated TNC services
FHWA’s Policy and Strategy Analysis Team completed
the Transportation Scenario Planning for Connected and
Automated Vehicles study in 2018. e process established a
range of plausible scenarios for the deployment and
adoption of CAV technologies. Each scenario included
common assumptions such as the rapid advancement of
mobile technology, 5G cellular connectivity, accepted V2X
standards, and passage of rigorous cybersecurity tests. e
scenarios progressively intensify connectivity, automation,
and cooperation factors. e six CAV scenarios serve to
illustrate how local agencies can use similar techniques to
develop their own CAV future scenarios. Similarly, the
Florida DOT noted that while the FHWA’s six CAV future
scenarios were used in describing travel demand modifi-
cations to represent the future, they are intended to be only
guides to local modelers and do not represent the only values
or factors that can be modified to reflect CAV
implementation.
e goal of this paper is to compare, as directly as
possible, the effects of CAV technologies on the Madison
transportation system by testing the same future scenario
alterations as were published in Florida DOT’s work for
Gainesville. Identical adjustments of key parameters of both
the Gainesville and Madison traditional four-step travel
demand models were made to simulate the six scenarios.
While many other factors could be tested in the Madison
model, the authors determined that this would be the fairest
way to compare CAV outcomes and impacts between the
two regional travel demand models and cities.
e four-step model integrates submodels for trip
generation, trip distribution, mode choice, and network
assignment. Table 1 elaborates on the specifics of the
modeling methods, their submodel elements, and general
parameters of the Madison model and the Gainesville model.
Inputs to the trip generation model are the unique
demographic and economic data for the region. Trip
Journal of Advanced Transportation 3

distribution requires the geospatial boundaries defined for
the various transportation analysis zones (TAZ) of the re-
gion and zonal activity level data. TAZ activity levels from
households, employment, shopping, and freight movements
determine the number of trips produced or attracted. e
type of trucks is often associated with the trip type in terms
of short-haul urban versus long-haul intercity freight
movements. e distance among zones defines a travel
impedance between them. is information allows a so-
called gravity model to determine how trips will distribute
among zones based on their activity levels and the imped-
ances between them. Mode choice typically utilizes a nested
logit model to determine a probability of selecting among
available modes based on certain attributes such as travel
time, travel costs, and accessibility. Network assignment uses
the actual roads or transit service available between the
different zones to assign generated trips until the result
meets some predefined criteria such as traffic equilibrium.
2.3. Model Adjustments. e adjustment of various pa-
rameters of a four-step model could simulate potential CAV
effects. Table 2 summarizes the adjusted parameters of the
Gainesville model. e authors strictly replicated these
adjustments for the Madison model.
Friction factors are used in the model to account for the
spatial separation between trip origins and designations. It is
anticipated that with CAVs, travelers will have greater
tolerance for longer distance trips since they will have time
to do other tasks instead of operating a vehicle during their
commutes. To simulate this effect in the regional models, the
friction factors for home-based work trips were increased 2.5
or 5.0 percent, depending on the implementation intensity
of the scenario. ese amounts represent a slight increase in
longer distance work trips made by commuters using au-
tomated vehicles as well as the increased availability of
automated vehicles to influence housing located further
away from employment locations.
Both models use lookup tables to enter roadway capacities
to the network. It is assumed that as more automated and
connected vehicles appear on the highways, capacities will
increase as density balancing will occur with drivers gaining
better traveler information and avoiding congested areas and
times, as well as more existing roadway space becoming
available as CAVs will allow for smaller gaps between vehicles.
Both freeways and arterials are anticipated to benefit from
these capacity-increasing effects. ree levels of freeway ca-
pacity increases were modeled based on slow (33 percent),
moderate (50 percent), and fast (75 percent) adoption rates.
Like freeways, three levels of capacity increases were modeled
for arterials to represent slow (15 percent), moderate (35
percent), and fast (50 percent) adoption rates.
Terminal times are added to each trip’s in-vehicle time. Trip
terminal time is controlled by the type of area within which
each trip begins or ends. With the increased use of automated
vehicles, it is anticipated that terminal times will be reduced as
vehicles deliver passengers as close to their destinations as
possible. Reductions to the CBD and adjoining areas of one
minute for the Slow Ride and Ultimate Traveler Assist sce-
narios and two minutes for all other scenarios were tested.
Modifications to the vehicle trip tables were made to
accommodate anticipated shifts due to increases or de-
creases in use of CAVs in certain areas. For example, in-
creases in trips due to the use of automated vehicles by
people who currently cannot drive due to age restrictions,
Table 1: Travel model subelement comparison.
Modeling
element Madison, WI Gainesville, FL
Platform CUBE voyager (scenario manager, application manager,
data manager, keys)
CUBE voyager (scenario manager, application manager,
data manager, keys)
Time periods 3 separate AM peak hours, 3 separate PM peak hours,
midday, night Daily
Base year 2010 2010
Future year 2050 2040
Internal zones 1110 560
External zones 41 26
Trip types
Home-based work, home-based shopping, home-based
school, home-based other, non–home-based, home-based
university, single-unit trucks, combination trucks
Home-based work, home-based shopping, home-based
social rec, home-based other, home-based university,
university campus/dorm, non–home-based, 4-tire truck,
single-unit trucks, tractor-trailer
Trip
generation
Productions: workers, household size x vehicles available Productions: household size x dwelling type, autos available
Attractions: employment/school enrollment x coefficients Attractions: employment/dwelling units/school enrollment
x coefficients
Trip
distribution Gravity model: person trips by purpose Gravity model: person trips by purpose
Mode split
Nested logit: auto drive alone, auto shared ride, drive to
local bus, walk to premium bus, drive to premium bus, bike,
walk
Nested logit: auto drive alone, 2-person carpool auto, 3+
person carpool auto, walk to local bus, walk to premium
bus, drive to transit, bike, walk
Highway
assignment Equilibrium with incremental loading Equilibrium with incremental loading
4Journal of Advanced Transportation

the elderly, or the disabled are anticipated. ere may also be
a shift from transit-based trips to automobile-based trips
with the advent of CAVs.
2.4. Model Output. e four-step models for each city
produced four system-wide parameters for each scenario:
vehicle miles traveled (VMT), vehicle hours traveled (VHT),
average free flow speed, and average congested speed. e
model calculated VMT by multiplying the amount of daily
traffic recorded for each roadway segment by its length and
then summed the individual results. VMT is a measure of
demand and a proxy for the level of congestion anticipated.
Similarly, the model calculated VHT by dividing the seg-
ment VMT by its average travel speed. VHT is a measure of
the quality of service that the network provides. Free-flow
speed is the average speed that motorists travel when there is
no congestion or other adverse conditions. e model
calculates congested speed based on the reduction of free-
flow speed as a function of the traffic volume to roadway
capacity ratio.
3. Data
As previously discussed, the model files included the re-
gional data that each model needed to run the various
scenarios. Table 3 summarizes the model parameter ad-
justments used for the Gainesville model [24]. e values
shown are the changes in the various parameters defined
earlier in Table 2. e FDOT based their adjustments on
observations from research conducted throughout the state.
e scenarios posit that CAV adoption will lead to an in-
crease in capacity for highways and arterials. In general, the
order of the scenarios is from increasing levels of CAV
adoption from the “slow” scenario. e exception is the
Robo-Mix scenario because fleets are operating non-
cooperatively, which means that they are not sharing data
about traffic and road conditions to help all vehicles make
better routing and timing decisions. Terminal times decrease
more substantially as adoption increases because of door-to-
door service. e friction factors for home-based-work
(HBW) trips increase with CAV adoption to reflect a pro-
pensity to travel further for work because of a reduction in
the value of travel time. e increase in trips for nondrivers
is the same for all scenarios. e scenarios for Robo-Mix and
Robo-PPP specify a relatively large increase in car use within
CAV service zones.
4. Results
e horizon years for the original Gainesville and Madison
models were 2040 and 2045, respectively. ose remained
unchanged to preserve the calibration and validation with
their respective baseline year data. e calculated free-flow
speeds for Gainesville and Madison were 29.65 MPH and
33.14 MPH, respectively. So, the Madison free-flow speed
was approximately 12% higher. However, after running the
models for the baseline future, the average congested speed
for Madison was 19% higher than Gainesville. Table 4
compares the model outputs for the original baseline case
and the six CAV scenarios. Figure 1 provides a visualization
of the proportional changes in the three outputs from each
city’s baseline case. e baseline VMTand VHTfor Madison
was 88% and 36% higher, respectively.
e VMT trends for both models generally increased with
increasing levels of CAV adoption. e Madison model
showed a consistent VMT increase with CAV adoption
scenarios. e Gainesville model had an exception in the
Ultimate Traveler Assist (Assist) scenario where VMT de-
creased. e FDOT attributed this exception to the use of
more efficient travel routes because of better real-time in-
formation [24]. e average VHT for the Madison model
increased in all scenarios whereas it decreased for three
scenarios in the Gainesville model. e model calculates VHT
from congested speed. Both models agreed on the direction of
change in congested speed for each scenario. However, the
Gainesville model showed a much larger decrease in average
congested speed for the Competing Fleets (Robo-Mix) sce-
nario. e FDOT report attributed this to fleet vehicles
repositioning and waiting on the next fare [24].
Figures 2 and 3 show the network models for Madison
and Gainesville, respectively.
e dense network with dark lines in the center of the
map represents the central business districts (CBDs). e
differences in spatial arrangements of the two networks are
Table 2: FDOT adjustments for CAVs.
Factor CAV effect simulated Control parameter
Trip length
average
e lengths of trips such as home-based work (HBW) may
increase due to the change in value of time when relieved
from driving
Friction factor (FF)—a factor of the gravity model that sets
the travel impedance among zones. is may change the
trip distribution among zones
Roadway
capacity
Density balancing, automatic rerouting, smoother traffic
flows, fewer incidents, and higher throughput from closer
travel gaps and higher speeds
Highway capacity (HC)—lookup table adjustments to the
capacities of freeways and expressways (HC-FE) in the
central business district (CBD), and capacities of divided
arterials (HC-AD) and undivided arterials (HC-AU) in
urban areas
Accessibility Door-to-door service and self-parking will reduce the out-
of-vehicle travel time necessary to access transportation
Terminal time (TT)—lookup table for the factor reduction
in out-of-vehicle travel time in CBD and fringe areas
Mode choice Increased accessibility and door-to-door convenience for
nondrivers may spur a shift away from public transit
Auto trip table—factor increase in trips taken by the current
nondrivers (AT-D) and a factor increase for special zones
that AVs service (AT-S)
Journal of Advanced Transportation 5

evident. e route restricted land isthmus of the Madison
network separates Lake Monona and Lake Mendota and
connects major urban districts located on either side of the
CBD. e lower density links surrounding the CBD are
classified as rural or suburban area types. Table 5 summa-
rizes some of the differences observed between the two
networks. ese geospatial and network differences help to
explain the large VHT differences for some of the CAV
adoption scenarios.
CAVs are expected to increase overall route capacities.
However, differences in how large geospatial features such as
lakes fragment an area and differences in the layout of road
networks that provide access to the CBD are likely reasons
for the differences in CAV impact on mobility. For example,
additional positive VHT reductions in Madison are likely
negated because much of Madison’s employment centers are
located on a route restricted land isthmus. Unlike the grid-
like arrangement of roads in Gainesville, the isthmus is a
traffic bottleneck that has no shorter route alternative
through it. Furthermore, unlike Gainesville, there is a
beltline highway that circumnavigates Madison, and there
are very few other east-west routes through the city. is
could account for the relatively lower benefits in VHT and
congested speed reductions demonstrated by the Madison
model.
e Madison model results are further scrutinized given
the predicted increase in VHT for all the CAV adoption
scenarios. Figure 4 summarizes CAV scenario adoption
impacts for specific subareas within the Madison model,
which are defined as rural, suburban, urban, and CBD. e
results show that VHT gradually increases with CAV
adoption intensity in all rural and suburban areas but in-
creases less noticeably in the suburban and CBD areas. is
is consistent with the spatial observation that the CBD lacks
east-west connecting roads that can provide access to any
freeways or arterials where capacity increases due to CAV
Table 3: FDOT model adjustments for CAVs adoption scenarios.
Parameter Slow Niche Assist Managed Robo-Mix Robo-PPP
HC-FE Unchanged +33% +75% +75% +50% +75%
HC-AD Unchanged +15% +35% +35% Unchanged +35%
HC-AU Unchanged Unchanged +35% +35% Unchanged +35%
TT −1 minute −2 minutes −1 minute −2 minutes −2 minutes −2 minutes
FF +2.5% +2.5% Unchanged Unchanged +2.5% +5%
AT-D +2.5% +2.5% +2.5% +2.5% +2.5% +2.5%
AT-S +5% +5% Unchanged +5% +7.5% +12.5%
Table 4: Comparison of model outputs.
Scenarios VMT (millions) VHT (thousands) Congested speed (MPH)
Gainesville Madison Gainesville Madison Gainesville Madison
Baseline 11.72 22.09 373.39 506.15 26.60 31.71
Slow 11.96 22.54 391.77 517.02 26.42 31.68
Niche 11.92 22.72 381.82 515.40 26.66 31.74
Assist 11.64 23.46 340.59 509.60 27.76 32.21
Managed 11.76 24.42 340.40 530.97 27.69 32.17
Robo-Mix 12.14 24.63 396.98 551.58 26.30 31.70
Robo-PPP 12.19 25.99 360.62 569.67 27.44 32.05
–2 2 6 10
(%)
14 18
Robo-PPP
Robo-Mix
Managed
Assist
Niche
Slow
Madison
Gainesville
(a)
(%)
–2–10 –6 2 6 10 14
Madison
Gainesville
(b)
(%)
–1 0 1 2 3 4 5
Madison
Gainesville
(c)
Figure 1: Proportional change of each scenario from the baseline. (a) VMT change. (b) VHT change. (c) Congested speed.
6Journal of Advanced Transportation

Rural
Suburban
Urba n
Dense urban
15 miles
Madison area type
Figure 2: Madison model area.
5.5 miles
CBD fringe area
Residential urbanized area
Urbanized area (CBD)
Transitioning areas
Outlying business district
Undeveloped urbanized area
Undeveloped rural
Rural
Gainesville area type
Figure 3: Gainesville model area.
Journal of Advanced Transportation 7

adoption could reduce travel times. e stable VHT changes
with CAV adoption intensity in the suburban and CBD areas
can be explained by additional trips to those areas offsetting
increases in capacity.
5. Discussion
When CAVs effectively increase the capacity of freeways and
arterials, traffic can reroute to rebalance densities. However,
the effect will be different among cities because of differences
in their geospatial characteristics, land use, and the layout of
the roadway network. is case study illustrated how sen-
sitive the forecasted CAV impacts can be to those differ-
ences. e change in average congested speed across CAV
adoption intensity scenarios was not as pronounced for
Madison as it was for Gainesville. All downtown arterials
pass through the isthmus in Madison whereas the grid-like
layout of arterials in Gainesville creates many route
alternatives through the city. erefore, Madison would be
less sensitive to highway and arterial capacity enhancements
from CAVs.
e FDOT report also stated that the results varied
among regions of Florida [24]. For example, the Robo-
Transit (Robo-PPP) scenario showed a decrease in VHT for
Gainesville but a significant increase for Central Florida. e
report attributed the increase in VHT to a larger shift in
travel on transit towards Robo-Transit as well as more in-
duced trips from nondrivers. is is plausible because of the
significant number of retirees in the area.
is study showed that there are limitations to under-
standing the potential impacts of CAVs by simply modifying
existing forecasting models. For instance, the characteristics
of specific zones based on age, income, disability, value of
travel time, and other factors may induce different rates of
CAV adoption. New models need to incorporate a factor
that accounts for the CAV impacts on changes in auto
Table 5: Comparison of model networks.
Characteristic Madison Gainesville
Network layout
e city center is on an isthmus that separates two lakes
(Mendota and Monona). Four lakes (Mendota, Monona,
Waubesa, and Wingra) fragment the urban network
e city center is a relatively small urban core located to
the east of a major interstate (I-75). No major geographic
features fragment the road network
Central business
district
Few roadways on an isolated isthmus with limited access
options to major highways
Many arterials and streets in a dense grid-like pattern
provide many access alternatives, including to I-75
Urban
environment
Distributed around two lakes (Mendota and Monona)
that bloat the spatial distribution of population and
employment
Continuous land mass that concentrates the population
and activity in a relatively small area
Urban arterial
network
Curved and diagonal routes with few through-connection
options
Grid-like and linear routes provide many interconnection
options
Suburban arterial
network
Heterogenous spatial distribution with limited direct
routes to the CBD
Many route options to the main highway (I-75) that
connects to the CBD
Rural
environment Roadways are sparsely distributed Roadways are more densely distributed
Highway access Access to the interstate system (I-39/90/94) is via few
arterials that circumnavigate the lakes Five major arterials connect to the interstate (I-75)
Base Slow Niche Assist Managed Robo-Mix Robo-PPP
–
50
100
150
200
250
VHT (thousands)
Rural
Suburban
Urba n
CBD
Figure 4: Madison 2050 VHT by area and scenario.
8Journal of Advanced Transportation

ownership and sharing. e integration of smart connected
corridors could affect the sensitivity to CAV-related changes
in capacity.
6. Conclusions
e impacts of connected and autonomous vehicles on trip
making and on the roadway system will be significant. e
research community generally agrees on how the adoption
of CAVs will affect traffic, but they can only speculate on the
level of those impacts since there are no significant de-
ployments yet. With such uncertainty, it is nearly impossible
to create and validate new travel demand models that in-
corporate factors such as the elasticity of auto ownership, the
propensity to travel, mode choice, and highway capacity.
erefore, the emerging tactic is to adjust the parameters of
existing travel forecasting models to simulate the anticipated
effects of CAV deployments. is work made identical
changes to parameters of the Madison, Wisconsin, model
and the Gainesville, Florida, model to compare outcomes.
In order to compare the results of each model run,
Vehicle Miles of Travel (VMT), Vehicle Hours of Travel
(VHT), and Network Congested Speeds were reported from
six plausible future scenarios for Madison and Gainesville.
As expected, increases occurred in VMTas more CAVs enter
the system. Percent changes from the base condition are
much more evident in Madison. VHT measures exhibit
greater difference between the Madison and Gainesville
models. In Madison, VHT increases for all scenarios, al-
though the Central Business District does not fluctuate as
much when individual area types are examined. Increases in
average congested speeds were exhibited in both models for
three scenarios: Robo-PP, Managed and Assist. Robo-Mix,
Niche, and Slow Growth scenarios remained relatively
unchanged or decreased slightly in average congested net-
work speed. e results from the Gainesville model appear to
be more prominent than for the Madison model in terms of
congested speed variance between scenarios.
e results of this study are limited to only two tradi-
tional four-step models chosen not only for similarities in
community and transportation system composition but also
for model availability and convenience. Identical adjust-
ments to inputs, constrained by the originally selected
scenarios, were made in order to compare their respective
network outputs with increasing CAV adoption. Both
models showed that at maximum adoption CAVs will in-
crease VMT, which agrees with the general results from
other research studies. e models showed that although
VMT would generally increase with CAV adoption levels;
some of the scenarios predict exceptions when there are
large differences in road network layout, geospatial features,
land use, sociodemographic factors, and accessibility to
transit. e models also reflect a corresponding increase in
the VHT because workers would tend to travel longer
distances for home-based work trips when the value of travel
time decreases. Although both VMT and VHT increased
with higher levels of CAV adoption, the models showed that
congested speeds also increased. is is an expected outcome
because CAVs can effectively enhance the capacity of a road
network. ey accomplish this by vehicles following more
closely and at higher speeds. eir ability to synchronize
movements can smooth out traffic flows. CAVs can also
select the most efficient routes in real time.
Planners in other parts of the country can replicate this
work and compare results. However, mathematical for-
mulations that relate each step of existing travel demand
models can hide details about underlying travel behaviors
and CAV effects. Future work will investigate sensitivities of
the congested speed for a range of parameter values to
discover trends that spatial regression and spatial autocor-
relation methods may better explain. e authors intend to
expand their work in this area to include further exploration
in the sensitivities of four-step models to CAV imple-
mentation by introducing additional scenarios and making
subsequent representative model adjustments and network
testing.
Data Availability
e data are to be requested directly from the Wisconsin
Department of Transportation and the Florida Department
of Transportation.
Conflicts of Interest
e authors declare that they have no conflicts of interest
regarding the publication of this paper.
Authors’ Contributions
Study conception and design were carried out by
D. Hungness and R. Bridgelall. Data collection was prepared
by D. Hungness and R. Bridgelall. Analysis and interpre-
tation of results were contributed by R. Bridgelall and
D. Hungness. Draft manuscript preparation was conducted
by R. Bridgelall and D. Hungness. All authors reviewed the
results and approved the final version of the manuscript.
Acknowledgments
e authors are grateful to the Wisconsin Department of
Transportation for providing the Dane County Travel De-
mand Forecasting model expressly and exclusively for this
research. ey are also grateful to the Florida Department of
Transportation for providing the Gainesville Travel Demand
Forecasting model and supporting documentation.
References
[1] B. Kaplin and S. Peterson, Connected and Autonomous Ve-
hicles and the Boston MPO – A First Look, Boston Region
Metropolitan Planning Organization, Boston, MA, USA,
2017.
[2] G. S. Nair, F. F. Dias, N. Ruiz-Juri, J. Kuhr, and C. R. Bhat,
Travel Modeling in an Era of Connected and Automated
Transportation Systems: An Investigation in the Dallas-Fort
Worth Area, Transportation Research Board, Washington,
DC, USA, 2020.
Journal of Advanced Transportation 9

[3] J. Hawkins and K. Nurul Habib, “Integrated models of land
use and transportation for the autonomous vehicle revolu-
tion,” Transport Reviews, vol. 39, no. 1, pp. 66–83, 2019.
[4] H. Raymond, B. V. Arerm, and S. Hoogendoom, “Automated
driving, traffic flow efficiency, and human factors: literature
review,” Transportation Research Record, vol. 2422, no. 1,
pp. 113–120, 2014.
[5] T. Litman, Autonomous Vehicle Implementation Predictions:
Implications for Transport Planning, British Columbia: Vic-
toria Transport Policy Institute, Victoria, Canada, 2019.
[6] T. Williams, Transportation Planning Implications of Auto-
mated/connected Vehicles on Texas Highways, Texas A&M
Transportation Institute, College Station, TX, USA, 2017.
[7] P. Bansal and K. M. Kockelman, “Forecasting Americans’
long-term adoption of connected and autonomous vehicle
technologies,” Transportation Research Part A: Policy and
Practice, vol. 95, pp. 49–63, 2017.
[8] A. Talebian and S. Mishra, “Predicting the adoption of
connected autonomous vehicles: a new approach based on the
theory of diffusion of innovations,” Transportation Research
Part C: Emerging Technologies, vol. 95, pp. 363–380, 2018.
[9] A. Soteropoulos, M. Berger, and F. Ciari, “Impacts of auto-
mated vehicles on travel behaviour and land use: an inter-
national review of modelling studies,” Transport Reviews,
vol. 39, no. 1, pp. 29–49, 2019.
[10] K. Kockelman, S. Boyles, P. Stone et al., An Assessment of
Autonomous Vehicles: Traffic Impacts and Infrastructure
Needs, University of Texas at Austin, Austin, TX, USA, 2017.
[11] J. B. Greenblatt and S. Shaheen, “Automated vehicles, on-
demand mobility, and environmental impacts,” Current
Sustainable/renewable Energy Reports, vol. 2, no. 3, pp. 74–81,
2015.
[12] M. Sivak and B. Schoettle, Influence of Current Nondrivers on
the Amount of Travel and Trip Patterns with Self-Driving
Vehicles, e University of Michigan Transport Research
Institute, Ann Arbor, MI, USA, 2015.
[13] S. Trommer, V. Kolarova, E. Fraedrich et al., Autonomous
Driving-e Impact of Vehicle Automation on Mobility Be-
haviour, Institute for Mobility Research (BMW Group),
Munich, Germany, 2016.
[14] L. T. Truong, C. De Gruyter, G. Currie, and A. Delbosc,
“Estimating the trip generation impacts of autonomous ve-
hicles on car travel in Victoria, Australia,” Transportation,
vol. 44, no. 6, pp. 1279–1292, 2017.
[15] S. Shaheen, A. Cohen, and I. Zohdy, Shared Mobility: Current
Practices and Guiding Principles, U.S. Department of Trans-
portation, Washington, D.C, 2016.
[16] E. Guerra, “Planning for cars that drive themselves,” Journal
of Planning Education and Research, vol. 36, no. 2, pp. 210–
224, 2016.
[17] J. Zmud, Updating Regional Transportation Planning and
Modeling Tools to Address Impacts of Connected and Auto-
mated Vehicles, e National Academies of Science, Wash-
ington, DC, USA, 2018.
[18] S. Childress, B. Nichols, B. Charlton, and S. Coe, “Using an
activity-based model to explore the potential impacts of au-
tomated vehicles,” Transportation Research Record: Journal of
the Transportation Research Board, vol. 2493, no. 1, pp. 99–
106, 2015.
[19] J. S. Miller and D. Kang, Ways to Consider Driverless Vehicles
in Virginia Long Range Travel Demand Models, Virginia
Transportation Research Council, Charlottesville, VA, USA,
2019.
[20] D. J. Fagnant and K. M. Kockelman, “e travel and envi-
ronmental implications of shared autonomous vehicles, using
agent-based model scenarios,” Transportation Research Part
C: Emerging Technologies, vol. 40, pp. 1–13, 2014.
[21] K. McLeod, Where We Ride: Analysis of Bicycle Communiting
in American Cities, e League of American Bicyclists,
Washington, DC, USA, 2014.
[22] M. Watkins, For Commuters, Gainesville is Best in Florida, e
Gainsesville Sun, Gainesville, FL, USA, 2015.
[23] N. Erdman, Madison in Motion, City of Madison, Madison,
WI, USA, 2017.
[24] FDOT, Guidance for Assessing Planning Impacts and Op-
portunities of Automated, Connected, Electric and Shared-Use
Vehicles, Florida Department of Transportation (FDOT),
Tallahassee, FL, USA, 2018.
10 Journal of Advanced Transportation