Assessing the Long-term Effects of Autonomous Vehicles: A Speculative Approach

Abstract

In recent years, self-driving cars have generated significant attention and discussion. While it is recognized that a number of technical and legal issues need to be solved, widespread adoption of self-driving vehicles is increasingly considered to be inevitable. However, the long-term effects of this technology are rarely considered and seldom examined in the literature. Among these potential impacts are a number of direct and indirect, positive and negative outcomes, and the net effect in terms of societal benefit or harm is far from clear. In this paper, we identify the several of these outcomes, and we explore conditions in the broader transportation system under which self-driving vehicles may be either harmful or beneficial. We investigate how autonomous operation could affect the attractiveness of traveling by car, how this in turn could affect mode choice, and how changes in mode choice would affect the broader transportation system. The paper considers three speculative scenarios, defined primarily by different behavioral responses to the availability of autonomous driving. The scenarios build on an established system dynamics model that represents the major forces involved in transportation systems. A wide range of outcomes are considered, and potential policy interventions are discussed.

Full text

Transportation Research Procedia 13 ( 2016 ) 18 – 29
2352-1465 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of Association for European Transport
doi: 10.1016/j.trpro.2016.05.003
Available online at www.sciencedirect.com
ScienceDirect
European Transport Conference 2015 – from Sept 28 to Sept-30, 2015
Assessing the Long-Term Effects of Autonomous Vehicles:
a speculative approach
Wolfgang Gruel a,
*
, Joseph M. Stanford b
aMassachusetts Institute of Technology, Cambridge, MA 02139-4307, USA | car2go GmbH, 70771 Leinfelden, Germany
bMassachusetts Institute of Technology, Cambridge, MA 02139-4307, USA
Abstract
In recent years, self-driving cars have generated significant attention and discussion. While it is recognized that a number of
technical and legal issues need to be solved, widespread adoption of self-driving vehicles is increasingly considered to be inevitable.
However, the long-term effects of this technology are rarely considered and seldom examined in the literature. Among these
potential impacts are a number of direct and indirect, positive and negative outcomes, and the net effect in terms of societal benefit
or harm is far from clear. In this paper, we identify the several of these outcomes, and we explore conditions in the broader
transportation system under which self-driving vehicles may be either harmful or beneficial. We investigate how autonomous
operation could affect the attractiveness of traveling by car, how this in turn could affect mode choice, and how changes in mode
choice would affect the broader transportation system. The paper considers three speculative scenarios, defined primarily by
different behavioral responses to the availability of autonomous driving. The scenarios build on an established system dynamics
model that represents the major forces involved in transportation systems. A wide range of outcomes are considered, and potential
policy interventions are discussed.
© 2015 The Authors. Published by Elsevier B. V.
Selection and peer-review under responsibility of Association for European Transport.
Keywords: Automated Driving; Autonomous Vehicle; Self-driving Car; System Dynamics; Long-Term Effects; Benefits; Negative Outcomes;
Scenario Planning
* Corresponding author.
E-mail address: wgruel@mit.edu
© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of Association for European Transport

19
Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
1. Introduction
Autonomous vehicles (AVs, often refered to as “self-driving cars” or “fully automated vehicles”) are widely
expected to play a major role in transforming our mobility systems in coming years.1 The expected benefits of this
technology are numerous: For example, it is predicted that autonomous systems can result in far safer operation and
improved efficiency of traffic flow. Furthermore, people who use these vehicles could use their commuting time much
more productively and pleasantly. Also, those previously unable to drive, due to age or infirmity, could gain access to
the same level of mobility as today’s car drivers (Hayes, 2011, Fagnant & Kockelman, 2013, Anderson et al., 2014).
With such promises, substantial effort has been exerted analyzing the technical and institutional prerequisites to
deployment of AVs. However, there has been very little formal research addressing potential system-wide and longer-
term effects of AVs. Most of the attention devoted to longer-term effects has taken the form of speculative, informal
discussion, usually focused on one or two issues in isolation (e.g., workforce changes, sprawl, etc.).
Our paper aims to identify some of the effects of vehicle automation at the system level. We focus in particular on
those reactions that may only become apparent when the entire system is considered and longer-term, indirect effects
are included. To grapple with the uncertainty inherent in this subject, we use a scenario-analysis approach, and employ
structured qualitative methods to develop conceptual system dynamics models. While considering the potential
benefits of AVs, we ask how behavioral changes made possible by AVs might affect traffic volume and congestion,
land use, and mode choice behavior. Building on this holistic view, we suggest policy recommendations to encourage
desired outcomes.
2. Related Work
Much of the material published on AVs to date—in both the academic and popular press—addresses specific issues,
often in isolation. For example, many studies have examined the effects of AVs on the efficiency of traffic flow (e.g.
Ioannou & Chien, 1993, Fagnant & Kockelman, 2013) or their impact on safety and car sharing (e.g., Pavone, Smith,
Frazzoli, & Rus, 2012, Spieser et al., 2014). To a lesser degree, there has been some attention paid to less-quantifiable
effects, such as behavioral changes, effects on attitudes about public transit, changes to land-use, and the impact on
urban and regional planning (see Coughlin & Yoquinto, 2015, Brustein, 2014, Madrigal, 2012, Chin, 2014).
Most papers, however, only address one or at most a few issues in relative isolation. Reports that provide a more
holistic approach are rare. Anderson et al. (2014) offer a summary of the potential benefits and perils of AVs, covering
immediate effects of the technology, including safety, mobility for underserved populations, congestion and “costs”
of congestion for users of AVs, energy use and pollution. This work also includes a brief discussion of potential effects
on land-use, including both the dispersion of destinations due to increased willingness to travel and increases in urban
density in some places due to the reduced need for parking. Fagnant & Kockelman (2013) also summarize some of
the expected benefits of AVs, including safety improvements, which they suggest could offer reductions in fatality
rates of up to 99%, and improvements in congestion, due to “shorter headways, coordinated platoons, and more
efficient route choices.”
Using analytical and simulation models, Burns, Jordan, & Scarborough (2013) explore a “new mobility system”
based on shared, driverless vehicles. The report finds significant economic, environmental, and consumer benefits,
suggesting that such an approach could “provide better mobility experiences at radically lower cost.” In similar vein,
Spieser et al. (2014) and Fagnant & Kockelman (2014) show in two case studies that a shared-vehicle mobility system
can satisfy the mobility demand of a city with significantly less vehicles while causing more travel for rebalancing the
fleet.
Townsend (2014) and Milakis et al. (2015) use scenario-approaches to examine the implications of AVs. Milakis
et al. focus more on the traffic and transportation planning implications of AVs. Their scenarios are defined along the
dimensions of technological development and policies. Townsend draws four scenarios, considering different forces
shaping the adoption of AVs, the resulting impacts on land use and transportation, financing schemes, and the role of
planning. The vastly different outcomes of these scenarios highlight how little is actually known about how AVs will
be used and how they will affect the overall system.
Research on the longer-term effects of AVs can be broken into two key categories. On the one hand side, there are
rigorous analyses based on narrow sets of data. Those typically result in well-supported outcomes, but the results are

20 Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
limited to very narrow topics. Secondly, there are speculations on a broader set of issues. Those speculative discussions
and analyses rarely build on existing models or frameworks of the transportation system as we know it. The resulting
predictions, therefore, are difficult to connect to existing dynamics and mechanisms. Also, the interplay of indirect
effects, system interactions and feedback loops have not been considered yet—although they may have tremendous
impact. For example, some studies suggest that AVs might increase vehicle kilometers travelled (VKT) and this might
result in increased sprawl (Fagnant & Kockelman, 2013). However, it is not acknowledged that increased sprawl, in
turn, will also put further upward pressure on VKT, creating a “vicious cycle”.
The purpose of this paper is to identify such indirect effects, system interactions and feedback relations and thus to
improve the understanding of the potential effects of AVs especially on user behavior, traffic volume and congestion,
land use, and public transit.
3. Method
In order to identify potential effects of vehicle automation at the system level, we build our work on an established
transportation base model, which describes known relations between a broad set of variables. Building on that, we
used qualitative interviews and workshops to investigate how this base model will change in different scenarios of
AV introduction. This section describes the modelling and data collection, the scenario development and briefly
introduces the base model.
3.1. Modelling and Data Collection
As a tool to investigate effects of vehicle automation on a systems level, we have chosen to use system dynamics,
an approach used to analyze complex, dynamic systems (Forrester, 1958). At its core is a mapping process, which
identifies key variables and the causal relationships among them. By considering many variables in a system at once,
instead of focusing on relations between just a few, system dynamics can help to build “a better understanding of the
complex transportation system problems” (Abbas, 1990, for further discussion of the use of system dynamics to model
problems in transportation see Shepherd, 2014).
Because we focus on the relationships among variables and the overall structure of the system, our work is limited
to the use of causal loop diagrams (CLDs). CLDs can help to identify key variables and causal relationships. They
make perceived system structures visible—and thus open them for analysis, discussion, and opinion making
(Shepherd, 2014). One of the benefits of CLDs is that they can generate valuable insights before the relevant system
behaviors can be observed or measured. They shed light on the systemic implications of known or hypothetical
relationships by revealing the structure and potential behaviors that they entail (Forrester, 2007).
CLDs consist of only two main elements: variables, which are indicated by their names, and causal links, which
are indicated by arrows pointing from the independent variable to the dependent variable. Causal links have positive
or negative polarity, indicating the nature of the relationship—e.g., a link from variable A to B with positive polarity
means that an increase in A will cause B to be larger than it would be without A’s impact. Marks can also be added to
causal links to indicate delays. Once CLDs are assembled, “loops” will arise when the causal links from one variable
connect back to itself, after connecting to other variables (Sterman, 2000). These loops play an important role in
system dynamics modelling, so it is important to identify them and understand the role they may play in the overall
behavior of the model.
The structure of system dynamics models is largely based on qualitative data (Forrester, 1968, Luna-Reyes &
Andersen, 2003). In order to add formality to our data collection, we used the most common qualitative techniques:
interviews and group workshops (Babbie, 2014, Vennix, 1996). We conducted semi-structured interviews and
workshops with 30 individuals, with expertise in a range of fields, including transportation-planning, the automotive
industry, the sharing economy, urban planning, and government policy.
In these sessions, we presented a base model (Fig. 1) that describes key aspects of current mobility dynamics and
we asked the participants to suggest changes that might be caused by the introduction of AVs, and to identify how
they might affect the structure of the model. With each session, we analyzed the collected information and used the
results as input for further sessions (Luna-Reyes & Andersen, 2003). The summarized results are represented in the
models shown in section 4.

21
Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
3.2. Scenario Development
Given the amount of uncertainty in the potential effects of AVs, we chose to apply scenario-analysis methods
(Schwartz, 1991, Wilkinson & Kupers, 2013). Rather than trying to determine the most likely outcomes, this approach
aims to develop useful insights by exploring a range of different futures. It tries to identify the main forces that will
affect outcomes in question, followed by speculation about the development of these forces. To counter the inherent
subjectivity of this process and steer our analysis toward more objective outcomes, we used interviews and group
discussions with a diverse group of people with widely varying perspectives.
As the public discussion at the moment rarely questions if AVs will become a reality—but focuses more on when
they will be available—we assumed in our scenarios that AVs are successfully developed and fully adopted.
Technological challenges, legal obstacles, factors that may spur or hinder adoption, and the adoption process itself
were not subjects of our study.
Based on existing literature (Jonas et al., 2015, Townsend, 2014) and our interviews and workshop discussions,
two major dimensions of uncertainty emerged that seem to drive the outcomes of vehicle automation on a system
level: How will travel behavior change due to this technology and how will behavior regarding ownership change?
In other words: Are people going to use cars in the same way as they use them today or will existing behaviors change?
Will people still primarily own their own private vehicles or will mobility become a service? Based on these
dimensions, we developed three scenarios that are described in Section 4.
3.3. Baseline Model
To investigate the effects of vehicle automation, we used an established CLD of traffic and congestion developed
by John Sterman (2000). At the core of this model is the question of how to reduce road congestion. As in other
transportation models (e.g. Egilmez & Tatari, 2012, Pfaffenbichler et al., 2010), this CLD considers road capacity,
trip generation, land use, mode choice and public transit. Sterman’s model specifically considers the effects of a
common policy intervention to congestion: road building. The model makes an appropriate framework for discussion
as AVs are widely considered as a way increase road capacity, which can theoretically be accomplished due to closer
following distances possible with connected AVs.
In Sterman’s model (Fig. 1), Attractiveness of Driving plays a central role. It is a function of Travel Time, Desired
Travel Time, Public Transit Fares, and Adequacy of Public Transit. Attractiveness of Driving then plays a key role in
a number of dynamic effects that balance out the benefit achieved by road building: it increases Trips per Day, Average
Trip Length, and the Number of Cars in a specific region. All these factors tend to balance out the effect of road
building, and suggest that the system will settle out at a new equilibrium state with potentially no improvement in
congestion, but higher VKT.
As road/highway capacity expands, people are able to travel farther in less time, so the Size of the Region that can
be reached within a Desired Travel Time increases (and density of origins and destinations falls); this reduces the
Adequacy of Public Transit; this further increases the Attractiveness of Driving, which means fewer people ride transit;
fewer people riding transit means less revenue and poorer service. As a result, the number of people riding transit
decreases even further, which in turn leads to more people driving.

22 Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
Fig. 1. The baseline model (Sterman, 2000)

23
Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
4. Introducing Autonomous driving: Effects in Three Scenarios
In the following sections, we discuss three scenarios for a future mobility system with AVs (see section 3.2). In
each section, we first describe the scenario. We then explain the major changes that we made to Sterman’s model to
reflect the assumptions of each scenario. All of our scenarios assume complete adoption of AVs, including the benefits
of connected-vehicle technologies.
4.1. Scenario 1: Technology Changes, But We Don’t
Scenario description: This scenario assumes no change in behavior or in ownership. AVs are used in the same way
as cars are used today and vehicles are privately owned. Expected improvements (e.g., our interview subjects
presumed that “driving will be so much safer,” or “my commute will be so much more pleasant”) do not lead to any
behavioral changes.
Modifications to existing model: The fundamental assumption of this scenario is that behavior regarding travel
choices will not change with the introduction of AVs—i.e. the number of trips, distances travelled, and mode choice
stay the same. To reflect that assumption, we do not consider any additional effects of Attractiveness of Travelling by
Car in our model. We do, however, allow increases in Trips per Day per Car and Cars per Person, caused by the
increase in Mobility for those Unable to Drive. We consider AVs as a way to increase road efficiency. Thus, Pressure
to Increase Efficiency of Traffic Flow replaces Pressure to Reduce Congestion and Efficiency of Vehicle Operation
replaces Road Construction. We also renamed and inserted some variables and links for clarification (Travel Time as
inverse of speed, Attractiveness of Travelling by Car instead of Attractiveness of Driving; introduction of Congestion,
Dispersion of Origins/Destinations; link between: Dispersion of Origins/Destinations and Adequacy of Public
Transit).
We further introduce expected effects of AVs: Efficiency of Vehicle Operation, Safety of Vehicle Operation,
Mobility for Those Unable to Drive are assumed to increase while Attention Needed for Driving is going to decrease
(Fagnant & Kockelman, 2013, Anderson et al., 2014). These effects have further consequences: Increased Efficiency
of Vehicle Operation increases Highway Capacity and will reduce Energy Consumption. This, in turn, will decrease
Pollution and Energy Expenses. Lower expenses will increase the Attractiveness of Travelling by Car. Increased Safety
of Vehicle Operation, i.e. less accidents, will not only reduce congestion but also make travelling by car more
attractive. Improved Mobility for Those Unable to Drive leads to more Trips per Car per Day and to an increased
number of cars.
Expected system-level outcomes: In the case that AVs do not cause a change in people’s behavior, a lot of benefits
on a system level can be expected compared to our current mobility system. Travelling by car will be safer, cheaper,
less energy consuming, more environmentally friendly, and a lot of time spent in the car can be used in a better way.
Also, the situation of millions of people who currently have limited access to mobility would improve. Although these
new traffic participants will increase traffic volumes, it can be assumed that this effect could be offset by the increased
efficiency of vehicle operation and traffic flow.
4.2. Scenario 2: New Technology Drives New Behavior
Scenario description: In contrast to Scenario 1, this scenario assumes major changes in behavior related to travel
and use of vehicles, but it assumes no change in ownership choices. Individuals and businesses discover new uses for
cars and people are willing to take more and longer trips than they did in their conventional cars. For example, our
interview subjects observed: “I could work in my car, so I wouldn’t have to stay so long at the office,” and “I could
live farther away”.

24 Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
Fig. 2. CLD showing changes to the baseline model resulting from assumptions of Scenario 1
Modifications to existing model: Based on the assumption that people’s behavior will change due to the
introduction of AVs, we now consider effects caused by an increased Attractiveness of Travelling by Car: In our

25
Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
model, this increase will lead to more Trips per Car per Day and higher Average Trip Length, as well as lower levels
of Public Transit Ridership.
Fig. 3. CLD showing changes to the baseline model resulting from assumptions of Scenario 2

26 Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
We assume that increased Comfort & Utility of Time in Car leads to a higher Acceptable Travel Time. We also
include variables to represent potential new behaviors: Trips with empty vehicles will be possible. This might not only
lead to less parking traffic, but also to additional trips in the form of Empty Vehicle Trips to Avoid Parking. Overall,
this will make driving by car more attractive. New Uses for Cars become possible with AV technology, such as
sending your car to run errands for you, so it can be expected that the number of Trips per Day per Car and the Cars
in Region will increase.
Expected system-level outcomes: AVs will increase the attractiveness of travelling by car, causing a major shift in
travel behavior. As discussed in section 3.3, increasing Attractiveness of Travelling by Car triggers dynamics within
the system that will lead to a new equilibrium with higher traffic volumes at a similar level of congestion and with a
decreased public transit ridership. There will also be upward pressure on traffic volumes due to the growth in Trips
per Day per Car and Cars in Region that can be expected to result from New Uses for Cars. Apart from that, these
dynamics will drive the effects on land use and public transit discussed in Section 3.3: people will be willing to drive
longer distances, which will increase sprawl and make it more difficult to provide adequate transit services. Overall,
the structure of these dynamics suggests potentially much higher levels of VKT with its consequences and a risk of
devastating effects on land use and public transit.
4.3. Scenario 3: New Technology Drives New Ownership Models
Scenario description: This scenario builds on the behavioral changes of Scenario 2, but also assumes complete
change in ownership: It examines a case where all vehicles in operation are shared AVs. Multiple reasons could cause
the shift to shared vehicles—e.g., increased appeal of vehicle sharing as vehicles can drive where they are needed,
lower prices due to higher vehicle utilization, or enforcement by cities that ban private cars from certain areas.
Modifications to existing model: As many people share one car, the adoption of shared AVs will increase the
Utilization of Vehicles, and thus the Trips per Day per Car. It will reduce the number of Cars per Person and due to
higher utilization reduce the Fixed-Costs per Trip. This cost reduction will make travelling by car more attractive. On
the other hand, vehicle-sharing leads to increased Price Transparency as its per-trip-payment exposes many of the
hidden costs of driving that most people usually do not consider. This transparency has the potential to reduce the
Attractiveness of Travelling by Car. The AVs capability to drive while empty allows for redistribution of vehicles to
better match supply and demand, and ensures that cars are available where and when they are needed. Rebalancing
will increase the number of Trips per Day per Car.
Expected system-level outcomes: While vehicle sharing reduces the number of cars in the region, it also leads to
higher utilization of cars and thus more trips per car per day. Overall, it can be expected that traffic volumes and VKT
are going to increase compared with the current state, due to rebalancing trips and new behaviors. Benefits could result
from parameters not modeled here—e.g., newer vehicles on the road, increased fuel efficiency of AVs, and more
appropriate vehicle choices for different uses.
But the vehicle sharing scenario comes with more interesting effects. The cost of driving not only depends on the
individual user, but also on the structure of demand and supply of the vehicle sharing system. In other words, the
greater the imbalance between demand and supply in an area, the more empty rebalancing trips are needed, and the
more vehicles are required in a fleet to fulfill all travel requests. The number of empty trips and idle vehicles will
influence the per-mile cost for all users and thus influence the Attractiveness of Travelling by Car. This means per-
mile travel costs in small, dense areas will be lower than in sprawled-out areas. This would fundamentally change the
current cost-structure of driving in which driving becomes cheaper per mile the more someone drives. Travel cost also
influence the Size of the Region within Acceptable Travel Cost. Vehicle sharing could lead to a new balancing effect:
as other factors in the model encourage sprawl, the demand/supply imbalance in the region grows, which increases
per-mile costs and total trip costs, which reduces the size of the region within acceptable travel costs, which puts
downward pressure on sprawl. This could then drive some of the reinforcing effects of public transit: if vehicle sharing
reverses sprawl to some degree, then public transit may even gain attractiveness, resulting in higher ridership, which
over time translates to better service, and fewer trips taken by car.

27
Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
Fig. 4. CLD showing changes to the baseline model resulting from assumptions of Scenario 3

28 Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
5. Conclusion
The scenarios examined here illustrate a variety of potential outcomes from the adoption of AVs. They are not
meant to describe an exact picture of the future but rather show a range of possible futures. In all three scenarios,
driving will become safer, time spent in the car can be used differently, and the situation of people with limited access
to mobility would improve. Furthermore, the per-mile-cost and energy consumption would decrease. However, our
models show that not all the effects would be desirable: In all three scenarios, VKT is likely to increase, leading to a
potential increase in energy consumption and emissions. The level of increase, however, appears to differ significantly
among the scenarios.
Scenario 1 shows that if the people’s behavior does not change, AVs will most likely lead to better outcomes than
our current mobility system. If, however, behavior changes occur as depicted in Scenario 2, AVs increases the
attractiveness of travelling by car and, as a consequence, traffic volumes will rise significantly. This leads to a higher
equilibrium-level of congestion, with substantially higher levels of VKT. Also, sprawl would be further encouraged,
and public transit would become less and less adequate. In this scenario, AVs will not contribute to a more sustainable
mobility system but will likely put additional stress on traffic systems that in many areas are already performing at
their limits.
The results of Scenario 3 are less clear: in a purely shared AV scenario, the number of cars would decrease, which
can create positive outcomes as less parking space is needed. At the same time, however, the traffic volume and VKT
are even higher than in Scenario 2 due to rebalancing trips. This scenario suggests additional ambiguities in terms of
cost, which again affect the attractiveness of travelling by car. Costs are likely to decrease due to higher vehicle
utilization, but prices will become more transparent, which will make travelling by car appear more expensive.
Furthermore, the current cost-mechanics of driving would change, to favor short trips in dense areas, eliminating the
current per-mile cost-advantage of long car trips. The cost structure of vehicle sharing could also counteract sprawl
and thus make public transit a more adequate mobility option. This phenomenon could be supported by additional
factors that we have not considered in our models: shared AVs could have a role in strengthening transit by acting as
feeders providing first- and last-mile connectivity to trunk lines, supplementing existing service, and potentially
replacing high-cost, underutilized routes. This could improve the accessibility and appeal of public transit. We also
did not address the potential role of ride-sharing in shared AVs, which could have a substantial effect in very dense
areas, where users may only have to slightly alter their travel plans to share rides with other travellers.
The analysis of our three scenarios provides strong indication that the introduction of AVs alone cannot solve the
problems our current mobility systems are facing. When aiming for a greener, more sustainable future for
transportation, we have to consider this new technology and its broad impact on a system level. The models we
developed suggest several starting points for policy interventions that could help to steer the system towards more
desirable outcomes.
To reach those desired outcomes, it may be essential to develop ways that reduce the Attractiveness of Traveling
by Car, increase the Attractiveness of Public Transit, discourage urban sprawl, limit the amount of driving that people
can do (a form of rationing), or some combination of these. With regard to our model, this approach can be broken
down into several categories, many of which are already quite familiar in transportation planning circles: (1) increasing
financial costs—e.g. through road use charges or tiered road-pricing mechanisms that would allow a base level of
low-price road use, followed by increasing rates with higher levels of driving; (2) increasing travel time—e.g. by new
speed limits or lower road throughput; (3) limiting the comfort and utility of time in car—e.g. by requiring that at least
one passenger pays full attention to the road; (4) making transit and other modes more attractive and/or encourage use
of AVs to connect to transit; (5) regulating land use to discourage long commutes and encourage other modes; (6)
restricting driving—e.g. restrict the amount of driving each individual can do, prohibit private empty car trips, limit
to certain uses or only allow rides that connect to or from transit stops.
Finally, as we consider potential research and policy interventions, we believe it is important to keep in mind that
taking a “wait and see” approach with AVs can come with serious consequences. By the time there is a critical mass
of AVs in use, it may be too late to impose corrective measures. On the other hand, pro-active action can help to shape
a framework for AVs to become an integral part of a more sustainable transportation system we envision. For example,
it may be possible to get AVs widely accepted as shared-use vehicles and public transit tools, before they become
entrenched as a new form of private personal mobility. Our models and scenarios have identified some powerful forces

29
Wolfgang Gruel and Joseph M. Stanford / Transportation Research Procedia 13 ( 2016 ) 18 – 29
that may come to play on the dynamics of our transportation system–and thus provide a valuable starting point for
future policy discussions. AVs are a promising technology, but to leverage their potential, we have to launch the public
discussion about the desired–and undesired–long-term outcomes of their use right now.
References
Abbas, K.A., 1990. The Use of System Dynamics in Modelling Transportation Systems with Respect to New Cities in Egypt. System Dynamics
Society Conference Proceedings.
Anderson, J.M., Kalra, N., Stanley, K.D., Sorensen, P., Samaras, C., Oluwatola., O.A., 2014. Autonomous Vehicle Technology: A Guide for
Policymakers. RAND Coorperation, Santa Monica, CA.
Babbie, E.R., 2014. The basics of social research. Wadsworth Publishing, Belmont, CA.
Brustein, J., 2014. Self-Driving Cars Will Mean More Traffic. Bloomberg. http://www.bloomberg.com/bw/articles/2014-07-21/self-driving-cars-
will-mean-more-traffic.
Burns, L., Jordan, W., Scarborough, B., 2013. Transforming Personal Mobility. Earth Institute, Columbia University.
Chin, R., 2014. Driverless cars - the future of transport in cities? http://www.theguardian.com/sustainable-business/driverless-vehicles-future-car-
sharing.
Coughlin, J., Yoquinto, L., 2015. The Long Road Home: Autonomous cars could lead people to live almost 200 miles from their workplaces. Slate.
http://www.slate.com/articles/technology/future_tense/2015/05/autonomous_cars_and_the_future_of_the_commute.html.
Egilmez, G., Tatari, O., 2012. A dynamic modeling approach to highway sustainability: Strategies to reduce overall impact. Transportation Research
Part A 46, 1086 –1096.
Fagnant, D., Kockelman, K.M., 2013. Preparing a Nation for Autonomous Vehicles. Eno Center for Transportation, Washington, DC.
Fagnant, D., Kockelman, K.M., 2014. The Travel and Environmental Implications of Shared Autonomous Vehicles Using an Agent-Based Model.
Transportation Research Part C 40, 1–13.
Forrester, J., 1958. Industrial Dynamics: A Major Breakthrough for Decision Makers. Harvard Business Review 36 (4), 37–66.
Forrester, J.W., 1968. Industrial Dynamics—A Response to Ansoff and Slevin: Management Science. Management Science 14 (9), 601–618.
Forrester, J.W., 2007. System dynamics—the next fifty years. System Dynamics Review 23 (2/3), 359–370.
Hayes, B., 2011. Leave the Driving to It. American Scientist 99 (5), 362.
Ioannou, P.A., Chien, C.C., 1993. Autonomous intelligent cruise control. Vehicular Technology, IEEE Transactions on 42 (4), 657–672.
Jonas, A., Shanker, R., Liu, J., Jain, P., Mehta, N., 2015. Shared Autonomy: Put This Chart On Your Wall, It's My Sad Life. Morgan Stanley
Research, New York, NY.
Luna-Reyes, L.F., Andersen, D.L., 2003. Collecting and analyzing qualitative data for system dynamics: Methods and models. System Dynamics
Review 19 (4), 271–296.
Madrigal, A.C., 2012. Driverless Cars Would Reshape Automobiles *and* the Transit System. The Atlantic.
http://www.theatlantic.com/technology/archive/2012/09/driverless-cars-would-reshape-automobiles-and-the-transit-system/262953.
Milakis, D., Snelder, M., van Arem, B., van Wee, B., Correia, G., 2015. Development of automated vehicles in the Netherlands: Scenarios for 2030
and 2050, Delft.
Pavone, M., Smith, S., Frazzoli, E., Rus, D., 2012. Robotic Load Balancing for Mobility-on-Demand Systems. International Journal of Robotics
Research 31 (7), 839–854.
Pfaffenbichler, P., Emberger, G., Shepherd, S., 2010. A system dynamics approach to land use transport interaction modelling: the strategic model
MARS and its application. System Dynamics Review 26 (3), 262–282.
Schwartz, P., 1991. The Art of the Long View. Doubleday/Currency, New York, NY.
Shepherd, S.P., 2014. A review of system dynamics models applied in transportation. Transportmetrica B: Transport Dynamics.
Spieser, K., Treleaven, K., Zhang, R., Frazzoli, E., Morton, D., Pavone, M., 2014. Toward a systematic approach to the design and evaluation of
automated mobility-on- demand systems: a case study in Singapore. Road Vehicle Automation.
Sterman, J., 2000. Business Dynamics. McGraw-Hill, Boston.
Townsend, A., 2014. Re-programming Mobility: the Digital Transformation of Transportation in the United States. Rudin Center for Transportation
Policy and Management.
Vennix, J., 1996. Group Model Building: Facilitating Team Learning Using System Dynamics. Wiley, Chichester.
Wilkinson, A., Kupers, R., 2013. Living in the Futures. Harvard Business Review.