Using an Activity-Based Model to Explore the Potential Impacts of Automated Vehicles

Abstract

Automated vehicles (AVs) may enter the consumer market with various stages of automation in 10 years or even sooner. Meanwhile, regional planning agencies are envisioning plans for time horizons out to 2040 and beyond. To help decision makers understand the effect of AV technology on regional plans, modeling tools should anticipate its impact on transportation networks and traveler choices. This research uses the Seattle, Washington, region's activity-based travel model to test a range of travel behavior impacts from AV technology development. The existing model was not originally designed with AVs in mind, so some modifications to the model assumptions are described in areas of roadway capacity, user values of time, and parking costs. Larger structural model changes were not yet considered. Results of four scenario tests show that improvements in roadway capacity and in the quality of the driving trip may lead to large increases in vehicle miles traveled, while a shift to per mile usage charges may counteract that trend. Travel models will need to have major improvements in the coming years, especially with regard to shared ride, taxi modes, and the effect of multitasking opportunities, to better anticipate the arrival of this technology.

Full text

USING AN ACTIVITY-BASED MODEL TO EXPLORE
POSSIBLE IMPACTS OF AUTOMATED VEHICLES
Suzanne Childress
(Corresponding Author)
Senior Modeler
Puget Sound Regional Council
1011 Western Ave., Suite #500
206-464-7090
schildress@psrc.org
Brice Nichols
Associate Modeler
Puget Sound Regional Council
1011 Western Ave., Suite #500
206-464-7090
bnichols@psrc.org
Billy Charlton
Program Manager
Puget Sound Regional Council
1011 Western Ave., Suite #500
206-464-7090
bcharlton@psrc.org
Stefan Coe
Senior GIS Analyst
Puget Sound Regional Council
1011 Western Ave., Suite #500
206-464-7090
scoe@psrc.org
August 1, 2014
WORD COUNT:
Words: 5,696
Tables: 2x250
Figures: 3x250
Total: 6,946

Childress, Nichols, Charlton, Coe 1
ABSTRACT
1
Automated vehicles (AV) may enter the consumer market with various stages of automation in
2
ten years or even sooner. Meanwhile, regional planning agencies are envisioning plans for time
3
horizons out to 2040 and beyond. To help decision-makers understand the impact of this
4
technology on regional plans, modeling tools should anticipate automated vehicles’ effect on
5
transportation networks and traveler choices. This research uses the Seattle region’s activity-
6
based travel model to test a range of travel behavior impacts from AV technology development.
7
The existing model was not originally designed with automated vehicles in mind, so some
8
modifications to the model assumptions are described in areas of roadway capacity, user values
9
of time, and parking costs. Larger structural model changes are not yet considered. Results of
10
four scenario tests show that improvements in roadway capacity and in the quality of the driving
11
trip may lead to large increases in vehicle-miles traveled, while a shift to per-mile usage charges
12
may counteract that trend. Travel models will need to have major improvements in the coming
13
years, especially with regard to shared-ride, taxi modes, and the effect of multitasking
14
opportunities, to better anticipate the arrival of this technology.
15

Childress, Nichols, Charlton, Coe 2
INTRODUCTION
16
Automated vehicles (AVs) are under development by major car manufacturers and technology
17
firms, and may enter the consumer market with various stages of automation in ten years or even
18
sooner (KPMG and CAR 2014). Meanwhile, regional planning agencies are envisioning plans
19
for time horizons out to 2040 and beyond. Within the time horizon of the plans, AVs may
20
significantly alter transportation choices, impacting regions’ planning goals. To understand
21
future travel patterns, modeling tools should anticipate automated vehicles’ impact on
22
transportation networks and traveler choices.
23
24
In the latest long-range regional plan, the Puget Sound Regional Council (PSRC) (2010)
25
established goals to guide the region toward healthy growth, including:
26
27
 improving safety and mobility,
28
 reducing greenhouse gas emissions and congestion,
29
 focusing growth in already urbanized areas to create walkable, transit oriented
30
communities,
31
 preventing urbanization of rural and resource lands, and
32
 helping people live happier and more active lives.
33
34
These goals reflect statewide legislation from Washington State’s Growth Management Act as
35
well as federal aims outlined in Moving Ahead for Progress in the 21st Century Act (MAP-21).
36
Self-driving cars could impact all these focus areas, so anticipating their adoption is imperative
37
to maintaining timely and informed regional plans.
38
39
This paper considers modeling techniques to measure the impacts of self-driving cars using an
40
activity-based model, and explores how modeling capabilities must be improved to better answer
41
questions related to this new technology. Since there is so much uncertainty around the future of
42
AVs, several model scenarios are created to consider broad impacts of self-driving vehicle
43
adoption in the Puget Sound region of Washington State. These scenarios clearly stretch current
44
model capabilities, and depend on highly uncertain inputs. However, it is still useful to test the
45
existing models in order to start a conversation with planners and decision-makers, as well as to
46
highlight shortcomings in our existing methods to modelers. The aim of this paper is not to
47
accurately predict the future impacts of automated vehicles, but rather to develop appropriate
48
ways of evaluating a range of potential impacts on regional transportation.
49
50
BACKGROUND
51
Automated vehicles could drastically change traffic flow, up-ending long-held assumptions
52
about maximum roadway capacity and volume-delay functions. Vehicle-to-vehicle coordination
53
systems allow cars to travel with much shorter headways, enabling higher volumes at high
54
speeds. If AVs also reduce collision rates, non-recurrent congestion would decrease as well.
55
FHWA (2013) estimates that 60% of all congestion is attributed to non-recurring sources such as
56
crashes and other vehicle-roadway mishaps, suggesting that a safer, more coordinated fleet could
57
reduce delay and support more consistent travel times. Even partially-autonomous vehicle
58
capabilities may increase roadway capacity. Tientrakool et al.(2011) estimate that highway
59

Childress, Nichols, Charlton, Coe 3
capacity could be increased by 43% using vehicle sensors and up to 273% with vehicle-to-
60
vehicle communications. Shladover et al. (2013) estimate that vehicle-to-vehicle coordination of
61
adaptive cruise control could increase capacity by 21% with 50% of all vehicles using the
62
technology, or up to 80% capacity increase with a 100% coordinated vehicle fleet, based on
63
empirical testing. Fernandes and Nunes (2012) estimate that capacity could increase as much as
64
five-fold for platoons traveling around 45 miles per hour. More efficient fleets could be
65
represented as additional roadway capacity, which can be represented very easily in existing
66
travel models.
67
68
To date, few regional-scale modeling efforts have addressed potential impacts of AVs. Gucwa
69
(2014) tested some capacity-altering assumptions on regional VMT in the San Francisco Bay
70
Area using the Metropolitan Transportation Commission’s activity-based travel model. Gucwa’s
71
results suggest that doubling capacity only increases region-wide VMT by around 1%, but does
72
significantly reduce peak congestion. Gucwa found that changing users’ values of time had much
73
more impact on VMT than capacity changes, and estimated the Bay Area’s VMT would increase
74
between 8% and 24%, depending on how automated vehicles users’ time values changed.
75
76
Gucwa’s findings suggest that changes in user behavior may have large effects on regional travel
77
as vehicle fleets become more automated. Gucwa, and many others, assume that being driven by
78
a robotic vehicle will eventually be less stressful than piloting one’s self through concentration-
79
demanding and chaotic congestion, thus making travelers less averse to in-vehicle time. Rather
80
than focusing on complicated navigation skills, travelers could spend time relaxing or working,
81
perhaps reducing the disutility placed on travel time. Since AVs are a new technology, the exact
82
influence of such attributes relative to travel time in these vehicles is unknown. However, these
83
factors are similar in nature to non-traditional transit attributes that often contribute to both mode
84
choice and route choice (Outwater et al. 2013). These attributes, such as comfort, reliability and
85
amenities like Wi-Fi, have proven difficult to explicitly represent in travel models. Instead,
86
through empirical methods, travel models can represent the utility associated with these
87
attributes through adjustments in travel time. Similarly, we can attempt to model the behavioral
88
changes that may arise from AVs by making assumptions about the equivalent perceived travel
89
time reductions that may result from ancillary factors.
90
91
Many other aspects of AV technology may affect traveler behavior as well, including costs,
92
vehicle availability and ownership, and parking price and location. Since more technical
93
infrastructure will be required to operate and manage self-driving cars, usage could more easily
94
be tracked per mile, making VMT-based taxes and pay-as-you-drive insurance policies more
95
realistic policy tools for personal vehicles. This pricing strategy could reduce overall VMT, as
96
frequently-forgotten fixed costs such as insurance, licensing, and registration fees are replaced
97
with more transparent marginal costs for every trip (Parry and Small 2005, Nichols and
98
Kockelman 2014). Shared autonomous vehicles would likely offer per-mile rates as well,
99
echoing existing business models from hired rideshare services like Uber and Lyft. Shared AVs
100
may become a popular service, since on-demand automated pickups would reduce the need to
101
own and thus store a personal vehicle. Depending on the technology’s development, many could
102
find owning a personal driverless vehicle too costly, relying on occasional pickups by shared
103
automated vehicles.
104
105

Childress, Nichols, Charlton, Coe 4
AVs may reduce the need for close-by parking as vehicles could conceivably self-park in
106
cheaper, more distance parking locations (Fagnant and Kockelman 2013). This behavior could
107
alter fixed costs at trip ends, reducing driving costs that lead to mode shifts or more automobile
108
travel to areas with high parking cost. Aside from altering destination choices and mode choice,
109
this behavior may also increase VMT as empty vehicles are sent for pickup and parking by
110
owners or users in a shared system. Some of these impacts can be easily modeled by simply
111
reducing parking costs in all zones, but accounting for increased VMT requires more knowledge
112
on parking cost, location, and trip tour timing.
113
114
VMT will likely increase as new users and more (perhaps longer) trips are induced from more
115
efficiently-operated roadways. Baseline demand consistently increases after congestion is
116
reduced with capacity expansion or operational improvements (see Cervero 2001 and Litman
117
2014b for meta-analyses of induced travel studies). Additionally, as in-vehicle time is less
118
stressful, travelers may be willing to tolerate slower travel times and longer travel distances,
119
adding more congestion still.
120
121
Fully autonomous vehicles may provide new mobility opportunities to those unable or unwilling
122
to drive a vehicle themselves, especially unlicensed young people, the physically impaired, and
123
some senior citizens. These user groups may be able to make more trips, access more
124
destinations, and rely on modes other than shared rides, public transit, and taxi. The amount of
125
additional mobility provided by AVs depends on mode shifts for non-drivers. Affordable,
126
competitive trips provided by a personal or shared AV would likely improve the opportunities a
127
non-driver could access, especially in more suburban, automobile-oriented contexts. Recognizing
128
how different groups are affected by AV developments is important to understanding regional
129
mobility and accessibility to jobs and resources.
130
131
Altogether, impacts of autonomous vehicles are highly speculative. Future impacts depend on
132
technological development, market reactions, and regulatory actions, making it challenging to
133
confidently predict impacts to regional transportation systems. With so many unknown and
134
potential effects of AVs, it is challenging to anticipate long-term effects with certainty. However,
135
some of these impacts should be considered early on, to understand model sensitivity and
136
develop feasible analysis boundaries. With these analyses, agencies can prepare more dynamic
137
long-range plans, by defining and evaluating some rational futures and exploring most likely
138
scenarios as technologies appear and mature.
139
140
MODEL SCENARIOS
141
To model potential impacts from automated vehicles in the Puget Sound region, four scenarios
142
are considered. The following sections explore ways to model some of the impacts mentioned
143
above and to provide guidance for other regions interested in planning for automated vehicle
144
futures.
145
146
PSRC’s activity-based travel model, called SoundCast, was applied to test the possible impacts
147
of automated vehicles. SoundCast includes a travel demand component written in the Daysim
148
software. SoundCast simulates individual travel choices across a typical day (PSRC 2014). These
149

Childress, Nichols, Charlton, Coe 5
choices include long-term choices like work location and auto-ownership, as well as shorter-term
150
choices like mode choice and route choice. Inputs to the model include parcel-based locations of
151
households and jobs, and highway and transit networks.
152
153
The scenarios have all been modeled using the base year of 2010, to avoid forecasting market
154
penetration scenarios or speculation on business models or technology development over time.
155
Using the most recent base year also helps focus the analysis directly on AVs, and avoids
156
uncertainties in future growth and changes to the transportation system. This isolation is useful to
157
understand some model behaviors and helps develop basic guidelines for evaluating automated
158
vehicles. As these analyses mature, future years should be evaluated for more comprehensive
159
case studies.
160
161
These scenarios explore how driverless cars can influence demand through changes in capacity,
162
perceived travel time, parking cost, and operating cost. They are described separately below.
163
164
Scenario 1: Increased Capacity
165
166
“AVs use existing facilities more efficiently.”
167
168
The first scenario reflects operational improvements from full or partial vehicle automation. This
169
scenario is modeled by increasing the hourly capacity coded on roadway network links and
170
captures one major impact of AVs on a roadway network. While it’s currently unclear what
171
magnitude of capacity increase is likely, based on cited research a 30% increase represents a
172
modest result from AV adoption. All freeway and major arterial capacities are increased by 30%.
173
174
Scenario 2: Increased Capacity and Value of Time Changes
175
176
“Important trips are in AVs.”
177
178
Scenario 2 builds upon the first scenario by assuming that, along with capacity improvements
179
from AV use, individuals using the AVs will perceive the time in them less negatively than time
180
spent driving in regular vehicles. The scenario envisions the point in time that AVs have only
181
been partially adopted, and only by higher income households. As with many new technologies,
182
the initial price will most likely only be attractive to higher income households. Considering that
183
the current cost of self-driving GPS technology alone is around $70,000, (KPMG and CAR
184
2012) adoption may be among high-income households for some time to come. This assumption
185
follows existing adoption patterns of more expensive cutting-edge vehicles such as hybrid and
186
electric vehicles. For example, Hjorthal, (2013) showed that early adopters of electric vehicles
187
were households with high income, owning more than one car, and used mainly to complement a
188
conventional car for commutes. Petersen and Vovsha (2005) found that higher income house-
189
holds tend to utilize newer vehicles, and among household members, the new vehicles are
190
allocated to workers at a higher rate than retirees and school children of driving age. A similar
191
trend might initially occur with AVs adoption. High income households might purchase one of
192

Childress, Nichols, Charlton, Coe 6
these vehicles, where it would be used for work and other important trips, while regular vehicles
193
would supplement for other, less important uses.
194
195
To test this scenario, modeled travel time was changed. In assignment, trip-based VOTs are
196
reduced by 65% for highest-income households, from $24 to $15.60/hour. Then in the demand
197
models, the automobile travel time was directly modified to be 65% of skimmed travel time in
198
the skims for the high value of time trips. In other words, a weight of 0.65 was applied to travel
199
time for auto trips with a high value of time. This travel time reduction reflects empirical results
200
from the Puget Sound, comparing preference for commuter rail lines versus local bus options,
201
where bus trips offer similar or shorter trips times, yet travelers opt for commuter rail, perhaps
202
for a more comfortable ride, consistent scheduling, or some other un-modeled biases. The
203
existing model accurately predicts commuter rail ridership when perceived time on commuter
204
rail is set at 65% of time on public bus. This scenario represents a similar but not equivalent
205
situation, in which travel time is perceived as less onerous between urban driving and sitting in a
206
self-driving vehicle. This behavior, of course, has not been revealed or even stated by drivers and
207
at this point is speculation based on other modes of transport.
208
209
Reduction in travel time has implications throughout the modeling chain. Travel time is a
210
variable in the following models: daily activity pattern, mode choice, destination choice, and
211
time of day choice. Because travel times are perceived as shorter, people will be willing to travel
212
further distances to work and school. They will also be willing to travel in more congested
213
conditions at peak hours, and may take more trips to do non-mandatory activities like eating
214
meals and shopping.
215
216
Scenario 3: Increased Capacity, Value of Time Changes, and Reduced Parking Costs
217
218
“All cars are self-driving, and none are shared.”
219
220
The third scenario uses assumptions similar to the previous scenario, but takes them a step
221
further to assume that all cars are self-driving. The scenario envisions the progression of the AVs
222
transitioning from high-income early adopters to total market penetration. This progression
223
would be similar to many new technologies like cell phones or the Internet that became
224
affordable through innovation and economies of scale. Since everyone is assumed to use an AV
225
in this scenario, travel time is reduced to 65% of skimmed travel time, for all trips, not just high-
226
VOT trips as in Scenario 2. In this scenario, all travelers, for all trip purposes, enjoy the benefits
227
of robot chauffeurs. As in the previous scenarios, freeway and major arterial capacity is
228
increased by 30%.
229
230
A third adjustment is also made for this scenario; parking costs are reduced by half to reflect
231
AVs self-parking in cheaper locations or better utilizing existing space (e.g., parking capacity
232
can be increased on existing lots since no room for driver access is required, thus increasing
233
supply of spaces and reducing costs). This change is made only in zonal parking costs and does
234
not capture VMT generated from vehicles seeking distance parking spaces or even roaming the
235
streets waiting for pickup commands. More detailed models could be developed to capture this
236
behavior and could form an independent research topic.
237

Childress, Nichols, Charlton, Coe 7
Scenario 4: Per-mile Auto Costs Increased
238
239
“"All autos are automated, with all costs of auto use passed onto the user.”
240
241
The final scenario considers a counterpoint situation in which AVs have become so common,
242
and shared AVs systems so effective, that personal AV ownership is no longer necessary.
243
Mobility is perhaps treated as a public utility, where all trips are provided by a taxi-like system at
244
a set rate. In this scenario, vehicles and road use are priced by a combination of industry and
245
government to cover infrastructure operation and maintenance costs. The scenario assumes that
246
all costs of driving are passed on to the user. The cost components that would be included under
247
such a scenario are: vehicle parking, vehicle and infrastructure maintenance, accidents, road
248
construction, vehicle cost, and negative externalities like congestion, air pollution, and global
249
warming. It is assumed that the system provides the same service as a personal automobile, but
250
comes at a higher per-mile rate. A rate of $1.65/mile was chosen to reflect total user and system
251
auto per mile costs and current ride-sharing taxi services. Litman (2007) estimated that the cost
252
per auto mile in urban area during the peak period was about $1.51 per mi. Furthermore 2014
253
per-mile price from Uber (2014) in Seattle was $1.65. The per-mile costs are a large increase
254
from current total costs of around 60 cents/mile (AAA 2013) and much larger than marginal
255
driving costs of 15 cents in PSRC’s model.
256
257
No capacity increase is assumed in this scenario, to reflect a worst-case scenario in which the
258
AVs provide no additional capacity (perhaps due to regulations preventing close car following,
259
for example). Table 1 summarizes these four scenarios for quick reference.
260
261
Table 1. Scenario Definitions.
262
263
264
265
266
Scenario 1 Scenario 2 Scenario 3 Scenario 4
"AVs increase network
capacity."
"Important trips are in
AVs"
"Everyone who owns a car
owns an AV."
“"All autos are automated,
with all costs of auto use
passed onto the user.”
30% capacity increase on
freeways, major arterials
30% capacity increase on
freeways, major arterials
30% capacity increase on
freeways, major arterials
Travel time is perceived at
65% of actual travel time
for high value of time
household trips (>$24/hr.)
Travel time is perceived at
65% of actual travel time
for all trips
50% parking cost reduction
Cost per mile is $1.65

Childress, Nichols, Charlton, Coe 8
RESULTS
267
The model outputs from Scenarios 1-4 are compared to the 2010 baseline to investigate the
268
potential impacts of the new technology. Table 2 shows the scenario results for typical measures
269
output by travel models. All the scenarios with a capacity increase indicate increased vehicle
270
miles travelled (VMT), ranging from around 4 % to 20%. However, only one of the three
271
capacity-increase scenarios showed an increase in vehicle hours traveled (VHT). In the first two
272
scenarios, the additional network capacity offsets the additional vehicle miles by allowing
273
vehicles to travel at a faster speed. In the third scenario, however, the reduction in perceived
274
travel time on all trips to 65% of the actual time, along with reduced parking costs induced so
275
much additional demand that the benefits from increase in capacity was offset.
276
Table 2. Scenario Results, Base Year 2010, Summaries by Average Travel Day.
277
278
279
280
Note that in all three of the capacity-increase scenarios the average network speed is higher than
281
the base scenario by about one or two miles per hour. The vehicle-hours of delay are reduced by
282
Measure Value Base 1 2 3 4
VMT Total Daily 78.7 M 81.5 M 82.6 M 94.1 M 50.8 M
% Change --- 3.6% 5.0% 19.6% -35.4%
(Versus Base)
VHT Total Daily 2.82 M 2.72 M 2.76 M 3.31 M 1.67 M
% Change --- -3.9% -2.1% 17.3% -40.9%
Trips Trips/Person 4.1 4.2 4.2 4.3 4.1
Distance Average Trip Length 6.9 7 7.2 7.9 5.8
(miles) Work Trips 12.4 12.9 12.9 20 11.5
School Trips 5.8 5.8 5.8 6.7 4.7
Delay Daily Average 846.0 700.0 727.2 996.1 350.2
(1000 hours) Freeways 288.1 201.2 218.3 338.7 56.4
Arterials 557.9 498.8 508.9 657.5 293.8
Speed Daily Average 27.9 30 29.9 28.4 30.4
(mph) Freeways 40 44.7 44.2 40.8 49.2
Arterials 22.5 23.2 23.1 22.3 24.3
Mode SOV Share 43.7 43.7 42.7 44.8 28.7
(%) Transit Share 2.6 2.7 2.7 2.4 6.2
Walk Share 8.6 8.6 8.4 6.8 13.1

Childress, Nichols, Charlton, Coe 9
about 150,000 vehicle hours in the first scenario and 100,000 vehicle hours in the second
283
scenario, but VHT and delay are both increased in Scenario 3 as VMT increases nearly 20%.
284
This surge in VMT corresponds to about 150,000 hours extra delay and about 17% more vehicle
285
hours. The increase in delay reflects the system-wide assumption of reduced perceived travel
286
time, where people are less averse to delay and thus more willing to tolerate congestion.
287
288
The additional vehicle miles result mostly from an increase in the number of trips and an
289
increase in the length of the trips. SoundCast includes sensitivity to travel time in the daily
290
activity pattern, exact number of tours, and intermediate stop models that predict the number of
291
trips people take. As perceived and actual travel time is reduced, the number of trips people will
292
take will increase because of a negative coefficient on travel time. For trip lengths, the
293
destination choice models have a negative coefficient on travel time, so users will travel farther if
294
the perceived travel time is reduced. In Scenario 3, average distance to work increases
295
dramatically to 20.0 miles, from a base of 12.4 miles. Much of this increase may be due to some
296
curious geographical quirks of our region: with less onerous drive time, some drivers may be
297
choosing to follow a circuitous path around Puget Sound instead of utilizing the shorter car-ferry
298
option across the Sound into downtown Seattle. In this scenario, total vehicle miles also increase
299
as travelers switch modes from transit and walking to single occupancy vehicles; transit shares
300
decrease around 9% and walk shares decline 21%.
301
302
Scenario 4 serves as counterpoint to Scenarios 1-3, to explore other ways in which AV could
303
affect regional transportation. This scenario is optimistic towards AV adoption and use; shared
304
AVs make owning a vehicle unnecessary, but travel is priced rather high (up to $1.65 per mile
305
versus 15 cents in the base), such that many trips are suppressed or trip lengths shortened.
306
Pessimism is assumed for operational benefits; AVs are thought to be used so widely in this
307
scenario that operational benefits are saturated, and no capacity increases are realized. If
308
increased per-mile costs were applied to all trips, model results suggest VMT may be reduced as
309
much as 35% versus the base. Vehicle-hours are similarly reduced by over 40%. Though
310
numbers of trips per person are very similar across all scenarios, Scenario 4 indicates travelers
311
will generally opt for shorter trips – average trip lengths are down 15% versus the base and over
312
25% less than Scenario 3, where average trip lengths are the longest of all scenarios. Scenario 4
313
results also suggest taxi-like pricing would cut drive-alone mode shares by a third, while transit
314
and walk modes might increase by 140% and 50%, respectively. Though some travel could be
315
suppressed in this scenario, the overall network performs better than the base or any other
316
scenario. Delay is less than half that in the baseline, and freeway speeds are nearly 10 mph faster
317
than the base.
318
319
Further analysis of tour lengths was performed. Table 3 shows the percent difference in tour
320
lengths by purpose when comparing Scenario 1 to Scenarios 3 and 4. Escort tours had the
321
greatest sensitivity to the modeled time and cost changes among the scenarios. When comparing
322
average escort tour lengths from the base scenario, Scenario 3 showed a 24% increase and
323
Scenario 4 had a 51% decrease. Because escort tours involve more travelers in the same vehicle,
324
they have a higher value of time than other tours. This higher value of time translates to a
325
greater sensitivity to travel time changes, and thus shorter tour lengths in Scenario 3. Scenario
326
4’s decrease comes from sensitivity to cost per mile of these tours.
327
328

Childress, Nichols, Charlton, Coe 10
Figure 1. Percent Difference in Tour Lengths: Scenario 1 Compared to Scenarios 3 and 4
329
330
331
332
Geographic Distribution of Results
333
Aside from aggregate system performance, model results were used to provide insight into the
334
spatial distribution of possible effects from AV. Figure 2 visualizes geographic distribution
335
results of the most “aggressive” automated car future, Scenario 3. In this analysis, an
336
accessibility metric called “aggregate tour mode-destination logsums,” or simply “aggregate
337
logsums,” is used. Aggregate logsums are household-based measures of accessibility, calculated
338
as the sum of the expectation across all possible locations, across all modes (Bowman and
339
Bradley, 2006). The aggregate logsums are calculated separately for households grouped by
340
income, vehicle availability, and transit accessibility, and separately by purpose. A fairly typical
341
household type was selected for analyses in Figures 2: a medium-income household located
342
within ¼ - ½ mile of transit, owning some vehicles, but fewer vehicles than adults. Aggregate
343
logsums are an index measure, and do not have much meaning by themselves, but can be used to
344
compare the differences between two scenarios.
345
346
Figure 2 shows that with capacity increases and a reduction in the perception of travel time as in
347
Scenario 3, perceived accessibility would be higher for most households, but especially higher
348
for more remote, rural households. Note that perceived accessibility increases for all households,
349
but especially for households in less urban areas. Two groups were selected to analyze how
350
different income groups would be impacted: one low income group and one high income group.
351
For the low income group, the percent change in aggregate logsums was 8.5% between the base
352
scenario and Scenario 3. For the high income group, the percent change in aggregate logsums
353
was about the same at 8.9% between the base scenario and scenario 3. The modest difference
354
between the low and high income groups difference in logsums indicates that with the scenario
355
as designed, low income populations experience nearly the same increase in accessibility as
356
higher income groups.
357
358

Childress, Nichols, Charlton, Coe 11
Figure 2. Accessibility Increase: Scenario 3 minus Base.
359
360
361
This result suggests that AVs, as modeled with assumptions in Scenario 3, would not reduce
362
access for any specific group and would actively increase accessibility in regions away from the
363
typically highly-accessible urban core. Scenario 3 assumes that driving is easier (increased
364
capacity), cheaper (lower parking costs), and more enjoyable (perceived travel time decreases)
365
for all users, leading to a jump in accessibility benefits directly through personal vehicle trips.
366
Though many Puget Sound residents would enjoy higher accessibility in this scenario, total VMT
367
climbs nearly 20%, possibly compromising the region’s goals of reducing greenhouse gas
368
emissions and containing growth into existing urban areas. Figure 3 shows how these VMT
369
increases are dispersed across the region.
370
371

Childress, Nichols, Charlton, Coe 12
372
Figure 3. Scenario 3, Estimated Changes in Average Daily VMT per Person over base
373
374
375
This result indicates that average VMT per person in nearly all zones would increase, with the
376
most extreme increases occurring in outlying areas, and even in some core zones of central
377
Seattle and Bellevue. Zones decreasing in VMT are generally sparsely-populated with few
378
samples to properly estimate. Improving regional mobility is one of PSRC’s goals, but such
379
improvements made through increased personal vehicle trips may be conflicting with
380
environmental and land-use targets.
381
DISCUSSION and RECOMMENDATIONS
382
Planning Implications
383
These results imply that AVs could both help and hinder PSRC’s policy goals. Speed and
384
capacity increases may improve regional mobility, but they also could induce additional demand,
385
leading to more VMT, and hence greater greenhouse gas emissions. If, on the other hand, a
386
greater share of AVs are electric than would have been otherwise, greenhouse gas emissions
387
could possibly fall. Reducing perceived travel time may provide a more enjoyable traveling
388
experience, but could facilitate longer trips and more VMT. The model runs show that
389

Childress, Nichols, Charlton, Coe 13
improvements in vehicle hours of delay from capacity expansion can easily be offset by the
390
reduction in perceived time. The amount of additional network capacity this technology can
391
provide is unknown, as are behavioral reactions of travelers. These analyses simply show that a
392
range of reasonable assumptions about AV adoption could have quite different impacts on
393
regional transportation. For example, if self-driving cars are priced per mile, both vehicle miles
394
travelled and vehicle hours travelled could be greatly reduced, by as much as 20 and 30%,
395
respectively, with SOV shares declining 40% and transit shares almost doubling. Conversely,
396
model assumptions in the first three scenarios indicate potential for much more VMT and delay,
397
with more people carried in SOVs, generally worse or equivalent network performance, but
398
higher mobility overall.
399
400
Self-driving vehicle adoption impacts are addressed in this paper from the perspective of PSRC’s
401
long-range plan goals of mobility, accessibility, and congestion impacts, but future research
402
should explore potential safety, emissions, and land use changes. Many simplifying assumptions
403
were used to isolate and test network and behavioral changes potentially associated with
404
automated technology development. However, if AV use does dramatically change regional
405
VMT, trip lengths, and mode shifts, it follows that land uses may shift dramatically as well.
406
Understanding these built environment changes will be very important in planning for future
407
impacts of AV technology.
408
409
Modeling Implications
410
Some improvements to this study’s methodology are achievable now, such as testing future-year
411
settings and linking the travel and land use models. These are perhaps the next logical next steps
412
in more detailed AV analyses, since changes in accessibility may be quite large and those
413
accessibility changes would impact land use development patterns.
414
415
More importantly, existing tools are demonstrably not sufficient for expressing the full range of
416
possibilities that automated vehicles may present. This study makes oversimplifications, such as
417
using a present-year land use assumptions and assuming broad AV costs and user values of time.
418
The model was estimated and calibrated against data that represents today’s network reality,
419
which is far outside of the reality that may exist with wide AV adoption. The challenges faced in
420
modeling AV scenarios highlight limitations of today’s tools in addressing this technology.
421
Many modeling improvements are considered below.
422
423
424
The future business model for shared AVs is entirely opaque. At a minimum, this could be
425
represented more directly with a top-tier taxi mode, which SoundCast currently lacks. Most
426
recent travel surveys indicate growing shares for taxi and taxi-like trips from ridesharing
427
services. Including a taxi mode would allow modelers to tweak performance and prices specific
428
to shared AVs. This would go a long way toward preparing our model for outcomes where many
429
of us may have robotic chauffeurs.
430
431
In activity-based models, household-owned AVs could be represented as a separate mode from
432
non-automated vehicles with their own modal constants and variables. Representing AVs as a
433
separate mode may be necessary if policy makers would like to consider separated lanes for
434

Childress, Nichols, Charlton, Coe 14
AVs. As with high-occupancy vehicles and toll links, AVs may need to be modeled using a
435
separate set of user classes with unique values of time and network link attributes.
436
437
The reduction in perceived travel time in AVs would be better modeled by attributing the
438
improvement in experience of travel time to actual measurable variables, as has been researched
439
with transit (Outwater, 2013). In mode and destination choice models, the stages of automation
440
could be a set of zero-one variables for the AV mode; assuming that the AV mode would
441
become more attractive with more automation and that with more automation, travel impedance
442
variables would have lower coefficients.
443
444
Currently, modelers lack the evidence to add AV-related alternatives and variables into travel
445
demand models. Because these vehicles do not yet exist but modelers need to incorporate their
446
possible impacts on travel demand, the most straightforward way to understand behavior would
447
be to conduct a stated preference survey.
448
449
A stated preference survey about travel behavior using AVs should try to answer some of the
450
following questions:
451
452
 How much would different types of people be willing to purchase different levels of
453
automation and for what price?
454
 Who would prefer to use the AVs as a shared service, and forgo car ownership?
455
 How will people perceive and value their time differently in AVs?
456
 Would people anticipate moving farther away from work?
457
 Would businesses choose to locate farther from the city center?
458
 What aspects of the AVs would cause people change their behavior most such as ability
459
to work, avoiding congestion, or safety?
460
461
Stepping further back and thinking about more than just variables and their coefficients, there are
462
some real shifts in how people perceive travel even today that our models simply don’t capture.
463
Multitasking (e.g. reading/emailing on a smartphone while on the bus), the effect of ingrained
464
habits and “lifestyle choices” (e.g., a person who really loves driving their luxury car, or another
465
person who would never consider driving to work even if it had free parking) need to be
466
incorporated in the next generation of models. Those types of high-level differences will be
467
amplified when a disruptive technology like AVs are introduced.
468
469
Closing Remarks
470
For modelers and policymakers alike it’s important to remember that, when presented with
471
automated vehicle technology, people are still going to behave based on the options available to
472
them and on the constraints they face in their daily lives. We have tried to lay out reasonable (or
473
at least conceivable) scenarios in this study, but the future may be more different than we’ve
474
envisioned. That possibility makes it even more critical that we improve our tools now to support
475
the policymakers and planners who will have to grapple with this new technology.
476
477
This research is just a starting point. We hope to continue the discussion as we sharpen our
478
predictive tools in the coming years.
479

Childress, Nichols, Charlton, Coe 15
REFERENCES
480
AAA (2013) Your Driving Costs. https://exchange.aaa.com/wp-content/uploads/2013/04/Your-
481
Driving-Costs-2013.pdf
482
483
Anderson, J., N. Kalra, K. Stanley, P. Sorensen, C. Samaras, O. Oluwatola (2014) Autonomous
484
Vehicle Technology: A Guide for Policymakers. RAND Institute.
485
http://www.rand.org/content/dam/rand/pubs/research_reports/RR400/RR443-1/RAND_RR443-
486
1.pdf.
487
488
Bowman, John L. and Mark Bradley (2008), Activity-based models: approaches used to acheive
489
integration among trips and tours throughout the day. European Transport Conference,
490
Leeuwenhorst, The Netherlands, October, 2008.
491
http://www.jbowman.net/papers/2008.Bowman_Bradley.Approaches_to_integration_ETC.pdf
492
493
Burns, L., W. Jordan, B. Scarborough (2013) Transforming Personal Mobility. The Earth
494
Institute, Columbia University.
495
http://sustainablemobility.ei.columbia.edu/files/2012/12/Transforming-Personal-Mobility-Jan-
496
27-20132.pdf.
497
498
Cervero, R. (2001) Induced Demand: An Urban and Metropolitan Perspective. University of
499
California, Berkeley. http://www.uctc.net/papers/648.pdf.
500
501
Fagnant, D., K. Kockelman (2013) Preparing a Nation for Autonomous Vehicles: Opportunities,
502
Barriers, and Policy Recommendations for Capitalizing on Self-Driven Vehicles. Eno Center for
503
Transportation. http://www.enotrans.org/wp-content/uploads/wpsc/downloadables/AV-
504
paper.pdf.
505
506
Fernandes, P., U. Nunes (2012) Platooning with IVC-Enabled Autonomous Vehicles: Strategies
507
to Mitigate Communication Delays, Improve Safety and Traffic Flow. IEEE Transactions on
508
Intelligent Transportation Systems 13(1): 91-106.
509
510
FHWA (2013) Traffic Incident Management. U.S. Department of Transportation, Federal
511
Highway Administration. http://ops.fhwa.dot.gov/aboutus/one_pagers/tim.htm.
512
513
Gucwa, M. (2014) The Mobility and Energy Impacts of Automated Cars. Dissertation, Stanford
514
University.
515
516
Hjorthal, R. (2013) Attitudes, ownership and use of Electric Vehicles - a review of literature.
517
Institute of Transport Economics. Report 1261/2013. 4/2013.
518
https://www.toi.no/getfile.php/Publikasjoner/T%C3%98I%20rapporter/2013/1261-2013/1261-
519
hele%20rapporten%20nett.pdf
520
521
Litman, T. (2007) Transportation Cost and Benefit Analysis II Cost Summary and Analysis.
522
Victoria Transport Policy Institute. http://www.vtpi.org/tca/tca06.pdf
523
524

Childress, Nichols, Charlton, Coe 16
Litman, T. (2014) Autonomous Vehicle Implementation Predictions: Implications for Transport
525
Planning. Victoria Transport Policy Institute. http://www.vtpi.org/avip.pdf.
526
527
Litman, T. (2014b) Generated Traffic and Induced Travel. Victoria Policy Institute.
528
http://www.vtpi.org/gentraf.pdf.
529
530
Lyons, G., J. Jain, D. Holley (2007) The use of travel time by rail passengers in Great Britain.
531
Transportation Research Part A: Policy and Practice 41(1): 107-120.
532
533
Nichols, B. and K. Kockelman (2014) Pay-As-You-Drive Insurance: It’s Impact on Household
534
Driving and Welfare. Transportation Research Record.
535
536
Outwater, M., B. Sana, N. Ferdous, W. Woodford, C. Bhat, R. Sidharthan, R. Pendyala, S. Hess
537
(2013) Characteristics of Premium Transit Services That Affect Mode Choice: Key Findings and
538
Results. TCRP H-37. Resources Systems Group, University of Texas at Austin, Arizona State
539
University, and University of Leeds.
540
541
Parry, I. and K. Small (2005) Does Britain or the United States Have the Right Gasoline Tax?
542
American Economic Review 94(4): 1276-1289.
543
http://www.econ.wisc.edu/~scholz/Teaching_742/Parry-Small.pdf.
544
545
Pinjari, A., B. Augustin, N. Menon (2013) Highway Capacity Impacts of Autonomous Vehicles:
546
An Assessment. Center for Urban Transportation Research. University of South Florida.
547
http://www.usfav.com/publications/TAVI_8-CapacityPinjari.pdf.
548
549
Puget Sound Regional Council (2010) Transportation 2040.
550
http://www.psrc.org/assets/4847/T2040FinalPlan.pdf.
551
552
Puget Sound Regional Council (2014) Activity-Based Travel Model.
553
http://www.psrc.org/data/models/abmodel/.
554
555
Shladover, S. E. (2009) Cooperative (Rather than Autonomous) Vehicle-Highway
556
Automation Systems (pp. 10–19). Berkeley.
557
558
Shladover, S., D. Su, X. Lu (2013) Impacts of Cooperative Adaptive Cruise Control on Freeway
559
Traffic Flow. Transportation Research Record 2324 (pp. 63-70).
560
http://trb.metapress.com/content/c7x847k3647888n1/.
561
562
Tientrakool, P., Y. Ho, N. F. Maxemchuk (2011) Highway Capacity Benefits from Using
563
Vehicle-to-Vehicle Communication and Sensors for Collision Avoidance. Vehicular Technology
564
Conference , 2011 IEEE. 10.1109/VETECF.2011.6093130.
565
566
Van Arem, B., C. van Driel, R.Visser (2006) The Impact of Cooperative Adaptive Cruise
567
Control on Traffic-Flow Characteristics. IEEE Transactions on Intelligent Transportation
568
Systems 7(4), December 2006. http://doc.utwente.nl/58157/1/Arem06impact.pdf.
569
570

Childress, Nichols, Charlton, Coe 17
Petersen, E, P. Vovsha (2005) Intra-Household Car Type Choice for Different Travel
571
Needs. Association for European Transport and contributors.
572
http://abstracts.aetransport.org/paper/download/id/2119.
573
574
Uber (2014) UberX Rates. https://www.uber.com/cities/seattle.
575
576
Wood, S., J. Chang, T. Healy, J. Wood (2012) The Potential Regulatory Challenges of
577
Increasingly Autonomous Motor Vehicles. Santa Clara Law Review 52(4): 9.
578
http://digitalcommons.law.scu.edu/cgi/viewcontent.cgi?article=2734&context=lawreview.
579