Content uploaded by Krishna Murthy Gurumurthy
Author content
All content in this area was uploaded by Krishna Murthy Gurumurthy on Jun 13, 2019
Content may be subject to copyright.
BENEFITS & COSTS OF RIDE-SHARING IN SHARED AUTOMATED VEHICLES
1
ACROSS AUSTIN, TEXAS: OPPORTUNITIES FOR CONGESTION PRICING
2
Krishna Murthy Gurumurthy
3
Department of Civil, Architectural and Environmental Engineering
4
The University of Texas at Austin
5
gkmurthy10@utexas.edu
6
7
Kara M. Kockelman, Ph.D., P.E.
8
(Corresponding Author)
9
Dewitt Greer Centennial Professor of Transportation Engineering
10
Department of Civil, Architectural and Environmental Engineering
11
The University of Texas at Austin – 6.9 E. Cockrell Jr. Hall
12
Austin, TX 78712-1076
13
kkockelm@mail.utexas.edu
14
15
Michele D. Simoni
16
Department of Civil, Architectural and Environmental Engineering
17
The University of Texas at Austin
18
m.simoni@utexas.edu
19
TRB Paper #: 19-04269
20
Presented at the 98th Annual Meeting of the Transportation Research Board in January 2019, and
21
published in Transportation Research Record.
22
ABSTRACT
23
A self-driving, fully automated, or “autonomous” vehicle (AV) revolution is imminent, with the
24
potential to eliminate driver costs and driver error, while ushering in an era of shared mobility.
25
Dynamic ride-sharing (or DRS, which refers to sharing rides with strangers en route) is growing,
26
with top transportation network companies (TNCs) providing such services. This work uses an
27
agent-based simulation tool called MATSim to simulate travel patterns in Austin, Texas in the
28
presence of personal AVs, and shared AVs (SAVs), with DRS and advanced pricing policies in
29
place. Fleet size, pricing, and fare level impacts are analyzed in depth to provide insight into how
30
SAVs ought to be introduced in a city. Results indicate that the cost-effectiveness of traveling with
31
strangers overcomes inconvenience and privacy issues at moderate-to-low fare levels, with high
32
fares being more detrimental than the base case. A moderately-sized Austin, Texas fleet (1 SAV
33
serving 25 people) serves nearly 30% of all trips made during the day. The average vehicle
34
occupancy (AVO) of this fleet was around 1.48 (after including the 12.7% of SAV vehicle-miles
35
traveled [VMT] that is empty/without passengers), with a 4.5% increase in VMT. This same fleet
36
performs better when advanced pricing is enforced in the peak periods (4 hours a day), moderating
37
VMT by 2%, increasing SAV demand and in turn fleet-manager revenue. SAVs are able to earn
38
around $100 per SAV per day even after paying tolls, but only at low-fare levels.
39
Keywords: Shared autonomous vehicles; dynamic ride-sharing, congestion pricing; agent-based
40
simulation; Austin, Texas
41
Gurumurthy, Kockelman and Simoni
2
INTRODUCTION
1
Smartphone technology is widely available and used by people of all ages and income brackets.
2
Internet-enabled smartphones are changing how we travel, shop and communicate in our daily
3
lives (1). In particular, on-demand services by Transportation Network Companies (TNCs), like
4
Lyft and Uber, are easily summoned and regularly arrive within 10 minutes. On-demand bikes are
5
also available in the central business districts of many major cities (2), and many parking lots are
6
becoming less relevant. A rather new service, in the form of en-route carpooling, or DRS, by TNCs
7
enables travelers to share their rides with people they have not met before, thereby reducing travel
8
costs further and making car ownership less attractive (3).
9
When AV technology eventually eliminates the driver’s role, big changes in travel behavior can
10
be expected. AVs will be expensive to own in the early years of implementation, and SAVs will
11
be the first economical alternative. DRS-enabled and fuel-efficient SAV systems will further
12
reduce operating and thus access costs, while lowering congestion, emissions, and other negative
13
externalities of driving one’s own car.
14
To reflect the details of car-sharing and ride-sharing, the multi-agent transport simulation
15
(MATSim) framework (4) is used here, to track travelers and vehicles across the City of Austin,
16
Texas (Figure 1) in 2035. MATSim uses a co-evolutionary algorithm to anticipate a convergent
17
set of dynamic traveler choices (for departure times, modes and routes) and traffic assignment to
18
the network. This analysis benefits from MATSim’s congestion feedback mechanism that uses
19
queue-based links to employ microscopic dynamic traffic assignment, as compared to static
20
assignment and aggregate zone-based and four-step demand modeling. MATSim’s agent-level
21
utility scores are maximized to seek user equilibrium in all cases. A recent DRS + SAV
22
contribution (5) to MATSim code allows for a base-case scenario consistent with that used in (6),
23
for several congestion pricing scenarios. Policies like advanced travel-time pricing (variable tolls),
24
along with different SAV+DRS fare assumptions, improve our understanding of how SAV fleets
25
and travel demand management through road pricing will impact mode choices and traffic
26
conditions, as well as traveler welfare.
27
The remaining sections of the paper are structured as follows: a brief literature review discusses
28
the benefits of simulation and DRS, the model framework is described, and key assumptions and
29
algorithms are presented. Subsequently, travel demand and traffic congestion results across
30
different policy, fleet-size and fare-level scenarios are presented, along with recommendations for
31
transportation planning and policy futures.
32
Gurumurthy, Kockelman and Simoni
3
1
FIGURE 1 Map of the Region Modeled - Austin, Texas (Source: Google Maps)
2
MOTIVATION
3
Dynamic ride-sharing or real-time carpooling in conventional vehicles has been studied for nearly
4
two decades now, but DRS benefits are not so widely known. The case study of U.S. cities (7)
5
highlighted DRS’ operational and environmental benefits. Since then, numerous studies have
6
analyzed DRS using different approaches and for cities with diverse land use patterns. More recent
7
studies assume the use of SAVs.
8
Optimization-based studies have emphasized the potential of DRS ((8), (9), (10),(11)). While they
9
highlight the viability of a ride-sharing system, they do so without considering the endogenous
10
effects of congestion and shifting mode choices. Recent survey-based research has examined mode
11
choices and willingness to pay (WTP) for DRS services inside SAVs, to get a statistical insight
12
into how the market may play out (see, (12),(13), (14), (15),(16)). These studies provide valuable
13
insights on DRS adoption rates and profitability possibilities, giving operators a sense of future
14
fleet sizes, but they lack the system perspective needed for managing the larger transport systems
15
at play. Fortunately, several simulation studies have been pursued. Some of these use more realistic
16
travel times (like (17), (18), (19), (20)), while others employ big datasets to understand travel on
17
static networks ((21), (22)). Both sets provide some system perspective, but congestion feedbacks
18
for SAV activities has been attempted only by (23), without DRS for Berlin, Germany.
19
Gurumurthy, Kockelman and Simoni
4
Agatz et al. (8) studied an optimization-based approach to understanding the potential of DRS and
1
optimization for Atlanta. They used data from a travel demand model and assessed wait-times that
2
such fleets can operate under. Two studies ((9), (10)) used the New York City taxicabs dataset
3
and graph-based optimization approach to show that high trip-making density, like in New York
4
City, significantly reduced fleet requirements by nearly 75%. However, they did not comment on
5
the system-wide VMT benefits it may or may not have. More recently, (11) formulated an
6
optimization problem to simultaneously assign incoming trip-requests as well as those waiting in
7
queue. They used small fleet-sizes to capture half of the Chicago taxicabs dataset under fast
8
computation times but had to rely on static networks.
9
Alternatively, survey-based research has been used to study mode-choice and willingness to pay
10
(WTP) to use DRS services. A stated preference survey (12) was used to study the market group
11
which was most likely to adopt such a service. The authors remarked that young people and people
12
with multimodal travel patterns were more like to take up DRS services. Other researchers (13)
13
studied the mode-choice behavior by comparing conventional car-use to privately-owned AVs and
14
SAVs and found that even freely available SAVs would only be used by 75% of the respondents.
15
In addition, (14) found that travel times played a significant part in WTP for DRS and many long-
16
distance business trips were likely to be shared. On the contrary, (15) have data that suggests that
17
only 10% of the people are actually interested in DRS but more may be willing to adopt if the price
18
of such a service is at a 40% discount to owning and operating a personal AV. Such studies suggest
19
that the perception on DRS is still evolving and simulation-based studies must be used to take
20
advantage of experienced travel times and their role in decisions to adopt DRS.
21
Simulation studies, on the other hand, have attempted to use experienced travel-times to get more
22
realistic travel behaviors and its impact on certain fleet metrics but have not fully considered
23
induced modal demand. DRS potentials in three cities with spatially differing travel patterns were
24
studied (24), namely Ann Arbor, Michigan, Babcock Ranch, Florida and Manhattan, New York.
25
The study focused on trip costs and empty vehicle-miles and concluded that DRS was
26
economically viable under all three scenarios. (17) used travel demand data and static travel times
27
generated by the model for the state of New Jersey and used a grid-based approach to study the
28
efficiency of DRS over the entire state. A similar approach was taken by (18) but they used
29
dynamic travel-times updated using MATSim to show more reliable operational benefits of DRS-
30
enabled SAVs in Austin, Texas. System-wide VMT was reduced for wait-times under 10 min. A
31
cell-transmission network assignment approach was used by (19) to get accurate experienced
32
travel times. Their study showed that DRS was needed to avoid the additional VMT that SAVs
33
without DRS would bring to the system. (20) studied the economics of DRS-enabled fleets with
34
the context of electric AVs and charging-location decisions. Their DRS results showed promise in
35
curbing empty VMT as well as system-wide VMT, as battery ranges improved, but used
36
experienced travel time as observed after one run of the simulation, similar to (25). (26) also
37
studied electric AVs and assessed the benefits of DRS using a 100 mi x 100 mi grid and solved a
38
capacitated vehicle routing problem with a time window. The results suggested a high vehicle-
39
replacement ratio of 13 conventional vehicles to 1 electric SAV and nearly 50% of all occupied
40
VMT by the fleet is shared. Another research (22) also used taxi booking data in Singapore to
41
understand trip-matching and fleet-sizing using a simulation tool they developed. Among only the
42
taxi trips, nearly 30% of VMT savings including a higher rate of incoming requests could be met,
43
especially during the peak times. A study based on Orlando, Florida (21) used cellphone data and
44
static travel times from the Metroplan Orlando travel demand model to simulate SAVs with DRS.
45
Gurumurthy, Kockelman and Simoni
5
Empty VMT was maintained under 4% and nearly half of all 2.8 million trips were met with the
1
SAV fleet with wait times under 15 min. Although these studies help garner support for DRS by
2
depicting its numerous benefits (e.g., VMT reduction, low wait-times, minimal travel delay, high
3
vehicle-replacement ratios), they do not wholly acknowledge induced demand for SAVs and DRS
4
since they do not take into consideration congestion-feedback.
5
Only a few studies have tried to fill the gaps mentioned above. For example, (27) used the San
6
Francisco Bay Area's activity-based travel demand model to assess the benefits of DRS. With a
7
focus on VMT, the authors examined how road-pricing scenarios compare to business-as-usual
8
and transit-oriented scenarios, using existing trip patterns. Across scenarios, 10% to 30% VMT
9
reductions were observed, thanks to moderate to high adoption of DRS. An agent-based model
10
was used to study SAV and DRS use in Lisbon, Portugal (28) and predicted a reduction of nearly
11
30% VMT, with the average SAV traveling about 155 mi per day. Their mode choice model was
12
based on current, revealed-preference data, which can limit future-year mode splits across AV
13
technologies, and they did not allow for changes in destination choice, which will come with
14
making travel easier.
15
As alluded to above, the existing DRS literature has some important gaps. This study seeks to
16
address those by: (1) Using real-time trip-matching with the option to share rides, (2) Using
17
congested-travel-time feedbacks within MATSim for mode choice flexibility, (3) Simulating SAV
18
+ DRS fare sensitivities, and (4) Anticipating the impacts of congestion pricing, DRS, and variable
19
SAV fleet sizes.
20
MODEL FRAMEWORK
21
MATSim’s general framework was developed by (4), to simulate individual travelers’ (or
22
“agents”) movements across networks, with flexibility in mode choice, departure-time choice and
23
route choices. An initial set of activity or travel plans is fed into the system for all simulated agents
24
(e.g., 10 percent of a region’s population, to moderate computing demands), and a co-evolutionary
25
algorithm seeks to maximize the utility of all agents based on user equilibrium. Since engagement
26
in activities is a key objective for each agent, activity participation is given positive utility while
27
travel and late arrivals are given a disutility. Re-planning of activities is done, as needed, to
28
(approximately) maximize the utility of each agent. Network assignment is done using a queue-
29
based approach, and the entire iterative framework is summarized as Figure 2’s MATSim loop.
30
The authors recommend that readers look through (6) for an in-depth explanation of the MATSim
31
loop within the context of shared mobility.
32
33
FIGURE 2 MATSim Loop (Source: (4))
34
Here, travel plans for 45,000 people residing in Austin, Texas, are used as the initial demand (29).
35
This 5% sample for the City of Austin with about 900,000 people (https://www.austintexas.gov/)
36
Gurumurthy, Kockelman and Simoni
6
was obtained by extracting city-specific trips dataset to limit computational burdens in tracking
1
the shared use of large fleets of shared vehicles. The complete Austin roadway network was
2
obtained from OpenStreetMap and consists of more than 148,000 links and 63,000 nodes.
3
MATSim runs were done on the Texas Advanced Computing Center’s (TACC’s) Wrangler
4
supercomputer, to enable 24-hour completion of most scenarios.
5
SAVs and Dynamic Ride-Sharing
6
In futuristic scenarios, it is important to recognize that AVs may be privately owned, by at least a
7
small percent of the population - even with very high purchase prices. To address this, AV
8
ownership was assumed for 10% of the simulated population, as suggested by (15) for early stages
9
of U.S. adoption. A higher or lower penetration rate for AVs can impact the outcome of this study
10
but emphasis is placed here on SAVs. The benefits of AV and SAV use are realized from higher
11
value of travel time associated with travel in these modes. This is taken into account in the travel
12
disutility associated with each modes and is discussed later.
13
SAVs are expected to be common modes of transport in urban settings in the long term, but DRS
14
may not be popular if travelers are largely unwilling to share rides with strangers (14). Code
15
developed by (5) is used and improves our understanding of how many trips can be shared, based
16
on travelers’ utility scores in MATSim. Fares assessed for SAV users are dependent of whether or
17
not trips are shared, distance traveled and time taken to cover the distance. Trips that are not shared
18
will cost more than shared rides, but there may be a lower willingness to share rides the longer the
19
trip’s duration (as found in (14)’s extensive survey work). Here, to facilitate trip-matching, all
20
SAV trip requests are evaluated as candidates for DRS. A least-cost-path algorithm in MATSim
21
identifies a subset of these trip requests to form an aggregated trip for a nearby SAV (i.e., an SAV
22
available to the first user within a 30-min radius). An unmet trip request is assigned a high negative
23
score so that in the convergent solution, each agent has learned whether using an SAV is a viable
24
option for their trip.
25
Table 1 provides travel disutility parameter assumptions (from (6)), with 𝛽𝑚𝑜𝑑𝑒 serving as each
26
model’s alternative-specific constant and 𝛽𝑡𝑖𝑚𝑒 serving as mode-specific marginal disutility of
27
travel time (in minutes). Parking costs and transit fares are inherently modeled into the 𝛽𝑚𝑜𝑑𝑒
28
parameter. SAV fares are assessed separately and personal AVs are expected to travel empty
29
(hypothetically) and not incur significant parking costs. The disutility for using SAVs (where rides
30
may be shared) is maintained the same as that for solo travel in one’s own AV. However, the
31
opportunity cost of spending time in a shared ride is expected to reflect lower disutility in sharing.
32
Operating costs are captured for personal vehicles, whereas, fares are assessed for the shared fleet.
33
Personal conventional vehicles have a nominal operating cost of 30¢/mi (30) for fuel and
34
maintenance averaged for different car-types. Personal AVs, on the other hand, are expected to
35
run more efficiently and is unlikely to require frequent maintenance. To capture this, personal AVs
36
have an operating cost of 20¢/mi (33% discount). SAV fares have a fixed-cost, a distance-
37
dependent cost and a time-dependent cost, as shown in Table 2. The reference fare level is based
38
on (6)’s study for 10% AV penetration: Solo travel in an SAV is assumed to be priced at 50¢ per
39
trip, 40¢/mi, and 10¢/min. A shared ride in an SAV is expected to cost less and is assumed to cost
40
approximately half as much as a solo trip in an SAV, i.e., 25¢, 20¢/mi and 5¢/min. As mentioned
41
previously, shared rides will accrue higher mileage (and, consequently, time), lowering the
42
willingness to use DRS. These rates are lower than what popular TNC companies like Uber or Lyft
43
charge for their DRS option (less $1/mi, on average, for shared rides and, about, $2/mi for single-
44
Gurumurthy, Kockelman and Simoni
7
occupant rides), but with driver costs eliminated and high value of travel time (31), fares are
1
expected to be lower in the future. The impact of such lowered fares is captured in stages. Transit
2
and the non-motorized modes are not charged monetary values but the travel disutility from the
3
opportunity cost of time captures the transit fares and unavailability of a vehicle for future legs. In
4
addition, the transit network for Austin is replicated here with access thresholds of a quarter mile.
5
TABLE 1 Mode Choice Parameters (adjusted from (6))
6
Travel Mode
𝜷𝒎𝒐𝒅𝒆
𝜷𝒕𝒊𝒎𝒆
Conventional
Vehicle
-0.1
0
Public Transit
-1.7
-0.36
Walk or Bike
-0.2
0
AV
0
+0.48
SAV (with DRS)
0
+0.48
7
TABLE 2 SAV Fare Level Assumptions for Shared Rides
8
Levels
Fixed Cost
($)
Distance-based
Travel Cost
($/mi)
Time-based
Travel Cost
($/min)
Reference Fare
0.25
0.20
0.05
50% Discount
0.15
0.10
0.03
75% Discount
0.10
0.05
0.02
FLEET-SIZING AND CONGESTION PRICING
9
Most past studies replace all conventional-vehicle trips with shared-fleet trips to assess the benefits
10
of SAVs. In the early adoption stages, however, conventional vehicles will continue to exist, along
11
with public transit and non-motorized modes (biking and walking). Making motorized travel easier
12
(through AV, SAV and DRS options) can result in a lot of extra VMT and congestion, so
13
congestion pricing policies are tested here, along with changes in fleet sizes and SAV fare
14
assumptions.
15
The no-toll scenario tested here is self-explanatory and provides insight into the viability of DRS,
16
with different fleet sizes and fares. The pricing scenario is straightforward: All major network links
17
(functional class of arterial or higher) carry a toll based on travel time on the link for all road users
18
during the morning and evening peak periods (7-9 am and 5-7 pm), at $0.05/min. This advanced
19
travel-time based pricing policy was simulated and tested for welfare viability in a future with AVs
20
and single-occupant SAVs (6).
21
RESULTS
22
In order to compare future travel behaviors to current travel trends, the 5% population sample for
23
the City of Austin is simulated using parameters for current travel, along with existing transit
24
schedule loaded in the Austin network, to better understand system performance. In the base case,
25
Gurumurthy, Kockelman and Simoni
8
mode shares reflect current patterns with cars dominating the streets as observed today (88.7% car
1
users, 4.1% public transit users, and 7.2% for walking and biking (32)). The pricing and fleet
2
scenarios examined here can be compared based on various performance metrics. These include
3
total vehicle-miles traveled (VMT), empty VMT (by SAVs without occupants), and average
4
vehicle occupancy (AVO) of the SAV fleet. The daily net income of the fleet is also calculated as
5
the difference between the revenue accumulated from fares and the toll paid by the operator.
6
Figure 3 shows the impact of fares, SAV availability and pricing on mode shares that are likely to
7
be observed, and resulting changes to VMT. The reference level for SAV fares does not attract
8
strong SAV use, even when the advanced pricing strategy is applied. SAVs are simply too
9
expensive per mile, as compared to owning and operating a conventional vehicle for daily travel.
10
Even when SAVs are widely available (1 for every 10 persons), the high fare only attracts about
11
3% mode share, with the shift from transit users and non-motorized modes. Mode share for travel
12
in a personal AV remains nearly constant for all scenarios tested, and this is not surprising given
13
the lower maintenance costs and lower travel disutility associated with AVs. Moderate-to-low fare
14
levels does show a significant change in mode shares. Moderate fleet size (1 SAV for every 25
15
persons) shows the most promise capturing nearly 30% of all trips made in one day. At low-fare-
16
levels, having SAVs available in abundance (1 SAV for every 10 persons) accounts for nearly half
17
the trips made in the day, but is found to be detrimental to system VMT. This is expected as trips
18
will likely be served without the need to ride-share and this will add a lot of empty VMT.
19
Gurumurthy, Kockelman and Simoni
9
1
FIGURE 3 Mode Shares & Change in Vehicle-Miles Traveled (Note: * corresponds to results
2
for the pricing scenario)
3
Gurumurthy, Kockelman and Simoni
10
Regional VMT is predicted to increase by about 5% with the introduction of SAVs, largely arising
1
from empty VMT when SAVs are operating with little-to-no demand. It is also unlikely that a
2
pricing strategy will mitigate this increase. When the SAV service is priced 50% lower than the
3
reference, more SAV trips are made, as expected, but the demand still seems to be insufficient for
4
effective ride-sharing. Serving more trips also resulted in 6-7% increase in VMT. At a 75%
5
discount, the competitive fares nearly tripled the SAV mode share. The effectiveness of pricing
6
travel is seen when SAVs form a significant mode share. The rate of increase of VMT is reduced
7
by about 2%. However, if SAVs are in abundance (1 for every 10 people), VMT is not effectively
8
contained since more people are performing single-occupant trips.
9
The operational viability of the SAV fleet is also important to discuss and is illustrated in Figures
10
4 and 5. Small fleet sizes increase response times considerably, thus, limiting the number of people
11
that are served within an acceptable response time. High response times drive the demand for
12
SAVs down. Consequently, when 1 SAV is required to serve 100 travelers, even competitive fares
13
do not capture the market well. The increase in ridership at low fares does mean that a higher AVO
14
can be achieved, as travelers are more willing to ride longer as a tradeoff to paying lower. A
15
maximum AVO of 1.56 is observed for a moderate fleet (with 1 SAV serving 25 people) when
16
pricing is implemented and fares are provided at a 75% discount as compared to the reference fare
17
level. Low mode shares arising from high fares, as well as reduced availability, greatly reduce fleet
18
revenue. Operators are estimated to make $67 per SAV per day. Although reduced fare levels may
19
imply lower revenue, the high mode shares and ride-sharing greatly improves operator revenue.
20
Operators can make as much as $113 per SAV per day with moderate SAV availability (1 SAV
21
every 25 persons) and after paying the tolls. Increased availability of SAVs beyond this threshold
22
reduces revenue from lowered AVO. Empty VMT as a percentage of the fleet’s VMT is
23
exceptionally high for small fleets (arising from low demand), and very low for large fleets (arising
24
from SAVs staying idle). A balance is struck for moderate fleet sizes (1 SAV for every 25 persons)
25
charging competitive fares achieving a relatively low empty VMT, as well as maintaining low idle
26
hours per SAV.
27
Gurumurthy, Kockelman and Simoni
11
1
FIGURE 4 Empty VMT, Average SAV Idle Hours in a Day, and Average Vehicle Occupancy
2
(AVO) [Note: * corresponds to results for the pricing scenario]
3
Gurumurthy, Kockelman and Simoni
12
1
FIGURE 5 Revenue, Tolls and Average Trips Served (Note: * corresponds to results for the
2
pricing scenario)
3
Gurumurthy, Kockelman and Simoni
13
CONCLUSION
1
AVs and SAVs are imminent, according to several experts and auto manufacturers. Smartphone
2
technology is ushering in an era of easy-to-use shared mobility, making it important to understand
3
well. DRS is picking up steam and this study utilizes agent-based tools to address some key gaps.
4
The study reveals that the use of SAVs with DRS is beneficial to the system when available
5
reasonably and priced competitively, especially with pricing schemes applied. Pricing improves
6
shared-use uptake and helps moderate rising congestion by reducing added VMT as compared to
7
not pricing. The service was analyzed for three fare levels for DRS, and it was seen that lowering
8
the fares thoughtfully increases the fleet’s revenue and demand. Even though SAVs are tolled, the
9
fleet continues to generate more than $100 a day per SAV in the best setting. This revenue is
10
comparable to how much existing fleets with human drivers make minus driver-related costs.
11
Lower prices increases riders’ willingness to share rides and produces a higher AVO. However,
12
large fleets promote single-use and diminishes the value of DRS. The revenue generated by these
13
large fleets also fall per SAV.
14
In the future, there is a good chance that transport is replaced by SAVs with rides more likely to
15
be shared. Future fleet operators must set fares such that the number of people served is high, but
16
maintain a moderate fleet of SAVs (on the order of 1 SAV for every 25 persons) to ensure a high
17
rate of DRS. This in turn provides system-wide benefits such as moderating the rising congestion
18
and shrinks the U.S. car fleet size. Congestion pricing will help resolve travel delays further, but
19
it must be set in accordance with SAV fleets in use. Federal dollars reserved for an underutilized
20
transit service can be used to provide an end-to-end service with SAVs ensuring that mobility is
21
affordable, convenient, and delay-free. By introducing private fleet operators, it is then also
22
possible to redirect accumulated tolls from fleets to individual drivers, thereby maintaining a
23
revenue-neutral pricing policy.
24
There are some minor drawbacks in this study. One of them is not modeling induced demand by
25
the elderly, children, and those without drivers’ licenses. Computational limitations restricted
26
simulations to a 5% population sample. Even if not explicitly considered, the results of this study
27
will hold true for induced demand as well. Higher reliance on fully-automated modes will increase
28
the viability of the service as demand served is key to a profitable service. Another aspect to note
29
is that popular TNCs charge about $2/mi for single-occupant trips and about $1/mi for shared rides.
30
Contrary to mode shares observed for SAVs here, TNCs enjoy nearly 5% mode share even at these
31
high prices in places like Seattle. This stems primarily from modelers’ assumption that an average
32
conventional vehicles cost 30¢/mi, whereas, the majority of conventional vehicles cost upwards of
33
60¢/mi nowadays. They would then be readily willing to let go of their vehicles and share rides,
34
thus, reinforcing the benefits SAVs and DRS have to offer.
35
AUTHOR CONTRIBUTION STATEMENT
36
The authors confirm the contribution to the paper as follows: study conception and design:
37
Gurumurthy, K.M and Kockelman, K., Simoni, M; Data analysis and interpretation of results:
38
Gurumurthy. K.M and Kockelman, K., and Simoni, M.; Draft manuscript preparation:
39
Gurumurthy. K.M., Kockelman, K. All authors reviewed the results and approved the final version
40
of the manuscript.
41
ACKNOWLEDGEMENTS
42
Gurumurthy, Kockelman and Simoni
14
We would like to thank the Texas Department of Transportation for funding this project in part
1
under project 0-6838, and Scott Schauer-West for helping with language edits. We would also like
2
to thank Sebastian Hörl for his support that enabled the authors to extend his contribution.
3
REFERENCES
4
1. Shaheen, S. A., A. Coehn, I. Zohdy, and B. Kock. Smartphone Applications to Influence
5
Travel Choices: Practices and Policies. Publication FHWA-HOP-16-023. FHWA.
6
2. Shaheen, S. A., A. Coehn, and I. Zohdy. Shared Mobility: Current Practices and Guiding
7
Principles. Publication FHWA-HOP-16-022. FHWA, 2016.
8
3. Chan, N. D., and S. A. Shaheen. Ridesharing in North America: Past, Present, and Future.
9
Transport Reviews, Vol. 32, No. 1, 2012, pp. 93–112.
10
https://doi.org/10.1080/01441647.2011.621557.
11
4. Horni, A., K. Nagel, and K. W. Axhausen, Eds. The Multi-Agent Transport Simulation
12
MATSim. Ubiquity Press, 2016.
13
5. Hörl, S. Agent-Based Simulation of Autonomous Taxi Services with Dynamic Demand
14
Responses. Procedia Computer Science, Vol. 109, 2017, pp. 899–904.
15
https://doi.org/10.1016/j.procs.2017.05.418.
16
6. Simoni, M. D., K. M. Kockelman, K. M. Gurumurthy, and J. Bischoff. Congestion Pricing
17
in a World of Self-Driving Vehicles: An Analysis of Different Strategies in Alternative
18
Future Scenarios. Transportation Research Part C: Emerging Technologies, Vol. 98, 2019,
19
pp. 167–185. https://doi.org/10.1016/j.trc.2018.11.002.
20
7. Levofsky, A., and A. Greenberg. Organized Dynamic Ride Sharing: The Potential
21
Environmental Benefits and the Opportunity for Advancing the Concept. Presented at the
22
80th Annual Meeting of the Transportation Research Board, Washington, D.C., 2001.
23
8. Agatz, N., A. L. Erera, M. W. P. Savelsbergh, and X. Wang. Dynamic Ride-Sharing: A
24
Simulation Study in Metro Atlanta. Procedia - Social and Behavioral Sciences, Vol. 17,
25
2011, pp. 532–550. https://doi.org/10.1016/j.sbspro.2011.04.530.
26
9. Santi, P., G. Resta, M. Szell, S. Sobolevsky, S. H. Strogatz, and C. Ratti. Quantifying the
27
Benefits of Vehicle Pooling with Shareability Networks. Proceedings of the National
28
Academy of Sciences, Vol. 111, No. 37, 2014, pp. 13290–13294.
29
https://doi.org/10.1073/pnas.1403657111.
30
10. Alonso-Mora, J., S. Samaranayake, A. Wallar, E. Frazzoli, and D. Rus. On-Demand High-
31
Capacity Ride-Sharing via Dynamic Trip-Vehicle Assignment. Proceedings of the National
32
Academy of Sciences, Vol. 114, No. 3, 2017, pp. 462–467.
33
https://doi.org/10.1073/pnas.1611675114.
34
11. Hyland, M., and H. S. Mahmassani. Dynamic Autonomous Vehicle Fleet Operations:
35
Optimization-Based Strategies to Assign AVs to Immediate Traveler Demand Requests.
36
Transportation Research Part C: Emerging Technologies, Vol. 92, 2018, pp. 278–297.
37
https://doi.org/10.1016/j.trc.2018.05.003.
38
12. Krueger, R., T. H. Rashidi, and J. M. Rose. Preferences for Shared Autonomous Vehicles.
39
Transportation Research Part C: Emerging Technologies, Vol. 69, 2016, pp. 343–355.
40
https://doi.org/10.1016/j.trc.2016.06.015.
41
13. Haboucha, C. J., R. Ishaq, and Y. Shiftan. User Preferences Regarding Autonomous Vehicles.
42
Transportation Research Part C: Emerging Technologies, Vol. 78, 2017, pp. 37–49.
43
https://doi.org/10.1016/j.trc.2017.01.010.
44
Gurumurthy, Kockelman and Simoni
15
14. Gurumurthy, K. M., and K. M. Kockelman. Modeling Americans’ Autonomous Vehicle
1
Preferences: A Focus on Dynamic Ride-Sharing, Privacy & Long-Distance Mode Choices.
2
Presented at the 98th Annual Meeting of the Transportation Research Board, Washington,
3
D.C., 2019.
4
15. Quarles, N. T., and K. M. Kockelman. Americans’ Plans for Acquiring and Using Electric,
5
Shared and Self-Driving Vehicles.
6
http://www.caee.utexas.edu/prof/kockelman/public_html/trb19americanFleetSurvey.pdf.
7
Accessed Jun. 26, 2018.
8
16. Quarles, N. T., and K. M. Kockelman. America’s Fleet Evolution in an Automated Future.
9
https://www.caee.utexas.edu/prof/kockelman/public_html/TRB19USfleetevolutionSAEVs.
10
pdf. Accessed Jun. 26, 2018.
11
17. Brownell, C., and A. Kornhauser. A Driverless Alternative: Fleet Size and Cost Requirements
12
for a Statewide Autonomous Taxi Network in New Jersey. Transportation Research Record,
13
Vol. 2416, No. 1, 2014, pp. 73–81. https://doi.org/10.3141/2416-09.
14
18. Fagnant, D. J., and K. M. Kockelman. Dynamic Ride-Sharing and Fleet Sizing for a System
15
of Shared Autonomous Vehicles in Austin, Texas. Transportation, Vol. 45, No. 1, 2018, pp.
16
143–158. https://doi.org/10.1007/s11116-016-9729-z.
17
19. Levin, M. W., K. M. Kockelman, S. D. Boyles, and T. Li. A General Framework for
18
Modeling Shared Autonomous Vehicles with Dynamic Network-Loading and Dynamic
19
Ride-Sharing Application. Computers, Environment and Urban Systems, Vol. 64, 2017, pp.
20
373–383. https://doi.org/10.1016/j.compenvurbsys.2017.04.006.
21
20. Loeb, B., and K. M. Kockelman. Fleet Performance and Cost Evaluation of a Shared
22
Autonomous Electric Vehicle (SAEV) Fleet: A Case Study for Austin, Texas. Transportation
23
Research Part A: Policy and Practice, Vol. 121, 2019, pp. 374–385.
24
https://doi.org/10.1016/j.tra.2019.01.025.
25
21. Gurumurthy, K. M., and K. M. Kockelman. Analyzing the Dynamic Ride-Sharing Potential
26
for Shared Autonomous Vehicle Fleets Using Cellphone Data from Orlando, Florida.
27
Computers, Environment and Urban Systems, Vol. 71, 2018, pp. 177–185.
28
https://doi.org/10.1016/j.compenvurbsys.2018.05.008.
29
22. Wang, Y., B. Zheng, and E.-P. Lim. Understanding the Effects of Taxi Ride-Sharing — A
30
Case Study of Singapore. Computers, Environment and Urban Systems, Vol. 69, 2018, pp.
31
124–132. https://doi.org/10.1016/j.compenvurbsys.2018.01.006.
32
23. Bischoff, J., and M. Maciejewski. Simulation of City-Wide Replacement of Private Cars with
33
Autonomous Taxis in Berlin. Procedia Computer Science, Vol. 83, 2016, pp. 237–244.
34
https://doi.org/10.1016/j.procs.2016.04.121.
35
24. Burns, L. D., W. C. Jordan, and B. A. Scarborough. Transforming Personal Mobility. p. 42.
36
25. Boesch, P. M., F. Ciari, and K. W. Axhausen. Autonomous Vehicle Fleet Sizes Required to
37
Serve Different Levels of Demand. Transportation Research Record, Vol. 2542, No. 1, 2016,
38
pp. 111–119. https://doi.org/10.3141/2542-13.
39
26. Farhan, J., and T. D. Chen. Impact of Ridesharing on Operational Efficiency of Shared
40
Autonomous Electric Vehicle Fleet. Transportation Research Part C: Emerging
41
Technologies, Vol. 93, 2018, pp. 310–321. https://doi.org/10.1016/j.trc.2018.04.022.
42
27. Rodier, C., F. Alemi, and D. Smith. Dynamic Ridesharing: Exploration of Potential for
43
Reduction in Vehicle Miles Traveled. Transportation Research Record, Vol. 2542, No. 1,
44
2016, pp. 120–126. https://doi.org/10.3141/2542-15.
45
Gurumurthy, Kockelman and Simoni
16
28. Martinez, L. M., and J. M. Viegas. Assessing the Impacts of Deploying a Shared Self-Driving
1
Urban Mobility System: An Agent-Based Model Applied to the City of Lisbon, Portugal.
2
International Journal of Transportation Science and Technology, Vol. 6, No. 1, 2017, pp.
3
13–27. https://doi.org/10.1016/j.ijtst.2017.05.005.
4
29. Liu, J., K. M. Kockelman, P. M. Boesch, and F. Ciari. Tracking a System of Shared
5
Autonomous Vehicles across the Austin, Texas Network Using Agent-Based Simulation.
6
Transportation, Vol. 44, No. 6, 2017, pp. 1261–1278. https://doi.org/10.1007/s11116-017-
7
9811-1.
8
30. American Automobile Association. Your Driving Costs: How Much Are You Really Paying
9
to Drive? 2017.
10
31. Fagnant, D. J., and K. Kockelman. Preparing a Nation for Autonomous Vehicles:
11
Opportunities, Barriers and Policy Recommendations. Transportation Research Part A:
12
Policy and Practice, Vol. 77, 2015, pp. 167–181. https://doi.org/10.1016/j.tra.2015.04.003.
13
32. Austin Chamber of Commerce. Austin Chamber 2016 Mobility Report.
14
https://www.austinchamber.com/upload/files/2016-mobility-report.pdf. Accessed Oct. 26,
15
2018.
16
17