Shared Autonomous Taxi Networks:
An Analysis of Transportation
Demand in NJ and a 21st Century
Solution for Congestion
Christopher Kirlin Brownell
April 2013
Advised by Alain Kornhauser
Submitted in partial fulfillment
of the requirements for the degree of
Bachelor of Science and Engineering
Department of Operations Research and Financial Engineering
Princeton University
I hereby declare that I am the sole author of this thesis.
I authorize Princeton University to lend this thesis to other institutions or individuals for
the purpose of scholarly research.
Christopher Kirlin Brownell
April 15, 2013
I further authorize Princeton University to reproduce this thesis by photocopying or by
other means, in total or in part, at the request of other institutions or individuals for the
purpose of scholarly research.
Christopher Kirlin Brownell
April 15, 2013
Acknowledgements:
First and foremost, I would like to thank my advisor Professor Alain Kornhauser. Prof. K,
you opened my eyes to the world of autonomous vehicles in ORF 467 and have led me
through this thesis process with your characteristic wit, boundless knowledge of the
subject matter, and infectious excitement about the future of transportation.
To my brothers in song, the Princeton Nassoons, I cannot put into words how much it has
meant to me to gain twenty-eight superbly faithful companions over my past four years at
Princeton. Throughout all my ups and downs, my six weeks in the ELE department, that
time in 2013 when I grew a lumberjack beard and consumed 43 cans of Monster, and all
my terrible jokes, the Nassoons have been a constant force of song, fun, and camaraderie
for me, a blessing which I know will continue until my melting day.
To Alex, thank you for your love and support throughout this thesis process and our
senior year. Thank you for introducing me to Princeton’s many libraries and the
productivity found therein, and for sharing my love of pirates, television, and highpitched voices.
To my family, I extend the utmost thanks and admiration. While Princeton has filled my
head with equations and opened my mind to the intricacies and wonders of life on this
earth, nothing can compare to the things I have learned from you over the past twentyone years. I thank my lucky stars every day that I am the son of such loving and
supportive parents, and the brother of such an intelligent, adventurous, and fun-loving
sister.
Finally, I would like to thank my mother, Jana Brownell, for the pair of cufflinks she
bought me on my first day at Princeton. As always, she had the brains and foresight to
know that engraving “CKB 2013” on the back would provide the extra kick in the pants I
needed to give this thesis my all and graduate with the illustrious Class of 2013.
Abstract:
As the most mobile people in the world, Americans rely on automobiles for the
majority of their personal travel. Over the past six decades, as private automobiles have
become more affordable and more universal among American families, cars’ previously
uncounted costs have come to the forefront of the modern transportation debate, with
some activists calling for an end to cars. This thesis presents five transit criteria that a
transportation system must satisfy if it hopes to dethrone the individually owned and
operated car as king of the road. They are: 1) a solution to the congestion problem, 2)
safety improvements over conventional manually operated cars, 3) a lesser impact on the
environment, 4) economic feasibility, and 5) comfort and convenience to rival the
automobile. Given recent advancements in the field of vehicle autonomy, a potential
solution to the car’s growing problems has presented itself: an autonomous taxi network
(ATN). Drawing from the classic Personal Rapid Transit model as well as Mark Gorton’s
idea of Smart Para-Transit, two potential designs for an ATN are presented and compared
to one another, and the viability of the ATN concept as a whole is explored using
statewide transportation demand from the state of New Jersey.
Table of Contents
Chapter One: Introduction ............................................................................................................... 1
1.1
American Mobility in Fifty Year Increments .................................................................. 3
1.2
Unsuccessful Challengers to the Automobile .................................................................. 9
1.2.1
The Failure of Personal Rapid Transit ................................................................... 10
1.2.2
Train, Bus, and Subway’s Inability to Become Universal ..................................... 17
Chapter Two: The Problems .......................................................................................................... 21
2.1
Road Congestion by the Numbers ................................................................................. 22
2.2
Safety Concerns ............................................................................................................. 27
2.3
Environmental Issues ..................................................................................................... 28
2.4
Culture of Car Ownership .............................................................................................. 31
Chapter Three: The Solution .......................................................................................................... 33
3.1
Advances in Vehicle Autonomy .................................................................................... 36
3.1.1
Google’s Driverless Car ......................................................................................... 37
3.1.2
Automakers’ Forays into Autonomy...................................................................... 39
3.2
An Autonomous Taxi Network ...................................................................................... 42
3.2.1
ATN’s Improvements to the Congestion Problem ................................................. 43
3.2.2
Safety Improvements in an ATN ........................................................................... 45
3.2.3
Environmental Improvements as a Result of an ATN ........................................... 48
Chapter Four: The Demand ........................................................................................................... 51
4.1
Step One: Generating Individuals .................................................................................. 53
4.2
Step Two: Assigning Primary Role Locations ............................................................... 56
4.3
Step Three: Generating Secondary Trips ....................................................................... 59
4.4
Step Four: Adding the Temporal Domain...................................................................... 62
Chapter Five: The Models ............................................................................................................. 65
5.1
Defining the PRT and SPT Models................................................................................ 66
5.2
Dividing New Jersey into a Pixelated Transit Grid ....................................................... 69
5.3
Calculating Fleet Size and Travel Costs using the Transportation Problem .................. 72
Table of Contents (continued)
Chapter Six: The Results ............................................................................................................... 79
6.1
Average Occupancy for Both Models Given
6.2
Average Occupancy for Both Models Varying
6.3
Further Analysis for SPT and PRT Model ATNs with
.......... 80
and
................................ 82
.... 85
6.3.1
Calculating Fleet Size with Instantaneous Repositioning ...................................... 88
6.3.2
Calculating Fleet Size with One Hour Break between Trips ................................. 90
6.3.3
Varying Chauffeur Speed for the SPT Model ........................................................ 92
6.4
Cost Comparison of the Two Models ............................................................................ 93
6.5
Comparing Comfort and Convenience in the Two Models ........................................... 94
Chapter Seven: Conclusion ............................................................................................................ 95
7.1
Potential Barriers to ATN Implementation .................................................................... 96
7.2
Further Analysis: Using the Transportation Problem to Model Repositioning ............ 98
7.3
Where to Go from Here?............................................................................................... 99
References .................................................................................................................................... 101
Appendix A: Detailed Flowcharts of Methods in Mufti 2012 ..................................................... 106
Appendix B: Code ....................................................................................................................... 109
Chapter One: Introduction
“Baseball, Hot Dogs, Apple Pie, and Chevrolet”
1
In a famous 1975 television commercial, a group of spokespeople for an American
automaker sings a short jingle about four things that are uniquely American: “Baseball,
Hot Dogs, Apple Pie, and Chevrolet.” And though the brief song is only a marketing plug
for the maker of the Corvette and the Malibu, a case can certainly be made for including
Chevrolet, Toyota, Ford, and all major makes of automobiles on the short list of what
makes America the country it is today. The United States is without question the most
mobile nation on earth, and is the most reliant on cars as a result. In 2006 the average
American travelled 18,700 miles, which is more than twice as much as the average
European, and almost three times as much as the average person in Japan.1
Given some of the issues surrounding automobiles today, including safety,
congestion, and the environment, a growing group of activists and urbanists are calling
for an end to cars.2 But even the harshest critics of the automobile admit that personally
owned and operated motor vehicles will not simply disappear from the fabric of
American life in an instant. Cars will only lose their status as the national means of
mobility if a technology arises that can match or surpass the car’s benefits of door-todoor transportation, comfort and convenience while also addressing some of the its major
costs: a growing congestion problem, safety hazards, and emissions associated with
global climate change. When looking into the future of transportation and mobility in the
United States, it is important to consider how we got to where we are. America was not
always as mobile a nation as it is today, nor as reliant upon the car. One way to appreciate
exactly how and why America became the most mobile country in history is to analyze
historical vignettes of American mobility in 1800, 1850, 1900, 1950, and 2000.
1
2
O’Toole 2009
See Dennis and Urry 2009 for an introduction to anti-automobile urbanism
2
1.1
American Mobility in Fifty Year Increments
1800: Humans and Horsepower3
Average Travel Speed
Average Mobility
Per capita GDP
3 miles per hour
1,500 miles per year
$1,2004
In 1800, writes researcher and author Randal O’Toole, “virtually no living person had
ever travelled faster than a horse could run.”5 All modes of transportation, including the
Omnibus (see Figure 1) which was just emerging in Europe, required either manpower or
horsepower to operate. As a result, the average travel speed across all available modes of
transportation was just under the average long-haul horseback riding speed of three or
four miles per hour. Because of these travel speed constraints, Americans, only six
percent of whom lived in urban areas, travelled about 1,500 miles each year on average. 6
This equates to about two three-hour round trips from a rural farm to a market, store, or
some other destination, each week. Shortly after the turn of the nineteenth century,
Meriwether Lewis and William Clark journeyed from Camp Wood, Illinois to the Pacific
Ocean, about three quarters of the way across the North American continent, in just over
sixteen months travelling mainly by canoe and on foot.
1850: The Canal, Steamboat and Rail Revolutions
Average Travel Speed
Average Mobility
Per capita GDP
4 miles per hour
1,600 miles per year
$1,900
3
Vignette titles and figures come from O’Toole 2009
All GDP numbers taken from O’Toole’s text are adjusted to 2007 dollars.
5
O’Toole 2009: p. 7
6
U.S. Department of Commerce 1975: p. 12
4
3
By the year 1850, a number of transportation innovations were underway; the advent
of canals, steamboats, and railroads did not significantly improve the average American’s
mobility right off the bat, but the innovations would set the stage for a boom in travel
down the line. More than 500 steamboats were in use on America’s rivers and the Great
Lakes by the mid-nineteenth century.7 While steamboats did not offer much of a speed
advantage over horses, they did offer a capacity advantage which extended not only to
the mobility of people, but also to the mobility of freight. In addition to travelling on
rivers and lake systems, steamboats also served the many canals that were constructed in
New York, Pennsylvania, and Maryland.
However the biggest transportation revolution that occurred from 1800 to 1850 was
the dawn of the railroad era. Trains had a significant speed advantage over horses and
steamboats, travelling between ten and thirty miles per hour along 9,000 miles of track in
all 30 states.8 Within cities, the horse-powered streetcar (see Figure 2) was the biggest
technological innovation of the era, and contributed primarily to personal mobility,
whereas the previous four revolutions made a much larger impact on freight shipping.
The average miles per year did not increase initially, as a trip on the train, steamboat, or
even urban streetcar was unusual for the majority of the population, who still travelled in
much the same way their parents or grandparents had in 1800: on horseback or by foot.
At the height of the westward migration along the Oregon Trail in 1850, the trip from
Missouri to Oregon, similar in distance to Lewis and Clark’s journey, took less than five
months.9
7
U.S. Department of Commerce 1975: p. 756
O’Toole 2009: p. 8
9
Unruh 1979: p. 403
8
4
State of the Art Transportation 1800-2000
Figure 1: Omnibus on Blackfriar’s Bridge10
London, 1789
Figure 2: Horse-Drawn Rail Car
New York City, 1850
Figure 3: Sprague’s Electric Rail Car
Richmond, 1888
Figure 4: First Production Benz Velo
1894
Figure 6: 2000 Toyota Camry12
2000’s Most Popular Automobile
Figure 5: 1951 Chevrolet Bel Air11
1951’s Most Popular Automobile
10
Images 1-4 from Kornhauser 2011
Image courtesy of www.smcars.net
12
Image courtesy of www.edmunds.com
11
5
1900: Railroads and Electric Streetcars
Average Travel Speed
Average Mobility
Per capita GDP
8 miles per hour
2,000 miles per year
$5,000
By 1900, the railroad revolution had transformed mobility in America in a much more
significant way than it had fifty years prior. A total of 193,000 miles of track traversed
the 45 states and three territories of the United States, and a trip from New York to
Chicago could be completed in twenty hours.13 Still, the train revolution mostly
contributed to increased freight mobility at lower costs. In the realm of human mobility,
two key methods of transportation that were operational by 1900 were the bicycle and the
electric streetcar. Although bicycles were not by any means a dominant mode of
transport, they were the first widespread demand-responsive vehicle since the
domestication of horses millennia beforehand. The idea of a demand-responsive
personally-owned vehicle would lead to an explosion of American personal mobility by
1950. The dominant mode of urban transportation in 1900 was the electric streetcar. First
introduced in the 1880s, these vehicles gained real success in Richmond, Virginia in 1888
under the direction of Frank J. Sprague (see Figure 3).14 By 1902, electric streetcars made
5.8 billion passenger trips in over 500 American cities, at an average speed of eleven
miles per hour.15 By 1900, some 250 companies claimed to be manufacturing
automobiles as well, though only about 8,000 cars were on the road at that time (see
Figure 4). A trip from New York to San Francisco on the Transcontinental Railroad in
1900 took approximately one week.
13
Beebe 1962: p. 14
Kornhauser 2011
15
O’Toole 2009: p. 11
14
6
1950: The Automobile Revolution
Average Travel Speed
Average Mobility
Per capita GDP
24 miles per hour
6,900 miles per year
$12,000
Between 1900 and 1950, the automobile experienced an historic transformation from
a novelty item for the wealthiest Americans to a nearly universal family possession. The
rise of the automobile was assisted by the aging electric streetcar fleets in cities across the
country, and was accompanied by a 67% decrease in intercity rail traffic between 1944
and 1950. In 1944, electric streetcars and intercity rail carried the average American
about 1,600 miles. By 1950, automobiles alone accounted for 4,800 passenger miles per
capita, nearly three times as much mobility as rail travel had provided at its peak.
Although the cost of owning a car was slightly higher than relying on rail transit, most
Americans were willing to pay the premium for the extra freedom that a demandresponsive personally-owned vehicle offered. The sentiment was so widespread that by
1950 there were nearly 50 million automobiles (see Figure 5) on the road in the U.S.,
which amounted to one for every three people.16 One of the main reasons that the
automobile increased American mobility so drastically is its extraordinary improvement
in ground transportation speed, effectively tripling the number from 1900. The explosion
in mobility for upper-class and middle-class families opened up opportunities such as
suburbanizing the areas surrounding large cities, and increased access to jobs, recreation,
and shopping essentially nine times over as compared to the year 1900. This helped
contribute to a per capita GDP growth of over 100 percent during the fifty year period.
16
U.S. Department of Transportation Federal Highway Administration 2011
7
The 1950s also saw the beginnings of commercial air travel, and a cross country flight
would have taken just under 12 hours in a Douglas DC-6B.17
2000: Superhighways and Jetliners
Average Travel Speed
Average Mobility
Per capita GDP
73 miles per hour
18,000 miles per year
$35,000
Although the technological improvements from 1950 to 2000 were minimal
compared to some of the other innovations mentioned in these vignettes, the fifty year
period saw a boom in mobility, specifically on the road and in the air. The only major
breakthrough on the road was the construction of the interstate highway system which
linked all major metropolises in the country by high-speed roads. The main breakthrough
in the air was the transition from propeller-driven aircraft to commercial jets, which is the
predominant force behind the drastic increase in average travel speed. However, the main
story here is the increase in American reliance on the automobile. In 2000, Americans
drove a total of 4.02 trillion miles, a 450 percent increase over 1950’s 730 billion miles.
This drastic increase in miles driven surprisingly enough coincided with a decrease in the
share of personal incomes devoted to driving. In 1950, the average American spent 9.8
percent of his income on driving; fifty years later that percentage had fallen to 9.4
percent.18 The increased affordability and increased use of automobiles culminated in a
2000 census report that 93 percent of American households had access to at least one car
(see Figure 6).
17
18
Smithsonian Air and Space Museum 2007
O’Toole 2009: p. 13
8
These mobility vignettes tell the story of the rise and fall of a number of different
transportation technologies via a process that twentieth century economist Joseph
Schumpeter called “creative destruction.” The 200-year timeline culminates in the rise of
the automobile as a national necessity. However, by the year 2000 the prevalence of
personally owned cars in the United States had contributed to more than just an increase
in American mobility and a coinciding increase in GDP. There are certain negative
externalities associated with the car that arose between 1950 and 2000 and have
worsened since the turn of the millennium; and yet, automobiles have continued to
dominate the personal transportation sector. As the goal of this thesis is to understand
what may come next in American transportation, it is important to analyze why the
automobile’s position on top of the transit food chain has not yet been seriously
challenged, despite cars’ growing congestion, pollution, and safety issues.
1.2
Unsuccessful Challengers to the Automobile
In a 1974 publication entitled Instead of Cars, British academic Terence Bendixson
cites growing congestion problems, air pollution, and an undesirably high accident rate as
the automobile’s “uncounted costs.”19 As a result of the mounting problems, Bendixson
explores various new technologies and potential government regulations in search of a
suitable replacement for the automobile. While he spends considerable energy discussing
improved bus systems and the limits of subways, Bendixson eventually puts his faith in a
new technology that he believes is destined to replace the automobile in a grand display
of Schumpeter’s creative destruction. However, nearly forty years after Bendixson
asserted his faith in the new technology, it has not come to fruition on any sort of large
19
Bendixson 1974: pp. 21-25
9
urban or suburban scale. It bears investigating, then, why exactly Bendixson’s prediction
that Personal Rapid Transit was bound to replace the automobile has not yet come true.
1.2.1
The Failure of Personal Rapid Transit
In a chapter entitled “Gee-Whiz Technology,” Bendixson describes the classic
Personal Rapid Transit model as the transportation of the future. The model includes
small driverless vehicles that travel along an elevated guideway, and stop at offline
stations when there is demand indicated at those stations. PRT has nearly all the benefits
of the automobile, in that it is a demand-responsive, door-to-door transit method, but does
not have the issue of congestion, would generate less pollution, and would be safer too,
having removed the possibility of driver error from the equation. PRT, writes Bendixson
in 1974, “seems almost certain to catch on in the United States.”20
With the benefit of hindsight, we now know that Bendixson was mistaken, and that
PRT has not caught on in the United States or anywhere else in the world by 2013. But it
is easy to see why Bendixson’s argument had some merit in the seventies. The city of
Denver, for example, voted in September 1973 in favor of a citywide PRT system that
included nearly 100 miles of routes.21 However, the program was eventually cancelled
due to a combination of political power shifts and charges of fiscal irresponsibility. This
thesis explores autonomous taxi networks as a new urban and suburban transportation
system, on an even larger scale than Denver’s proposed PRT. While the technology
behind an autonomous taxi system and its myriad benefits will be discussed in the
upcoming chapters, the issues of legal and government support are much harder to
quantify and solve. These legal issues may prove to be the barrier that most threatens the
20
21
Bendixson 1974: p. 207
Ibid: p. 203
10
viability of an autonomous taxi network, so the reasons behind the failure of Denver’s
proposed rapid transit system bear investigation.
Case Study: The Failure of Denver’s PRT
In the late 1960s and early 1970s, in response to growing congestion and pollution
problems in Los Angeles, many cities began looking into alternatives for highway-based
urban transportation plans. One such city was Denver, Colorado, which began a
comprehensive analysis of public transportation options in 1969, establishing the
Regional Transportation District (RTD), and mandating that it develop and fund a new
plan that “attracts riders, reduces dependence on the automobile, reduces total
transportation costs, and provides benefits to both users and nonusers of the system”
within five years.22 In June 1973, an independent conglomeration of transportation
consultants hired by RTD published their optimal plan. The report suggested a Personal
Rapid Transit system, composed of driverless cars on an elevated guideway, serving
demand as indicated by travelers at offline stations around the grid. The proposed system
was to be very similar to a pilot project in Morgantown, WV, which was under
construction at the time. The Morgantown system strayed from the classic PRT model in
several key ways, however – most notably the twelve-person capacity of its cars, which
made the system an example of Group Rapid Transit rather than true PRT. Although the
Denver plan referred to its suggested system as “Personal Rapid Transit,” it also
incorporated twelve-passenger cars, making it an example of GRT as well.
The year of the Denver transportation report coincided with a push toward innovative
rapid transit systems by the federal government. The Urban Mass Transit Administration
22
Princeton University: Critical History of Transit Planning
11
(UMTA) had in fact received $30 million in federal money to invest in “development of
an automated high-capacity transit system using four-to-six passenger vehicles that would
travel on guideways and provide fast, non-stop transportation from starting point to
destination”23 in early 1972. By October of the same year, the federal government
announced that $11 million had been appropriated to a Denver PRT project, but limited
the system to one of the four technologies that had been displayed at the 1972
transportation exposition, dubbed “Transpo.” The exposition included group rapid transit
models by Ford Motor Company and Bendix-Dashaveyor whose cars had total capacities
of 24 and 32 respectively. It also included true personal rapid transit models by RohrMonocab and Transportation Technology, Inc., each with a capacity of six.
Figure 7: Proposed plan for PRT in Denver, September 197324
23
24
Witkin 1972
Princeton University: Critical History of Transit Planning
12
In conjunction with the federal funding for the system, the people of Denver approved
the plan for PRT in their city, and an increased tax to help fund the project, in a
September 1973 vote. The proposed system of demand-responsive driverless pods was
meant to operate on the elevated guideways indicated by solid black lines on the map in
Figure 7, and to be supported by a system of feeder buses, the routes of which are
indicated by dotted black lines. It was marketed to voters as a true PRT system, though
the map and detailed plans implied a GRT with dual-directional guideways very similar
to the Morgantown system.25 The voters had no way of knowing that there was an
inherent incompatibility between true PRT and the proposed system map, so they
approved the tax, and the system that supposedly came along with it.
Even before the election in the fall of 1973 however, the plans for PRT in Denver
were in jeopardy. Earlier in the year, the UMTA awarded the city a technical studies
grant of $179,000 to “very carefully consider the alternatives of express buses and
exclusive bus lanes.”26 Because Denver had already spent years on an exhaustive
alternatives analysis that culminated in the June 1973 plan, it became clear that the
UMTA’s support of PRT or GRT in Denver was waning, and that they were funding an
analysis that would “emphasize costs not service and virtually preclude the introduction
of new systems.”27 By 1975, UMTA administrator Frank Herringer released the
following statement to the Senate Committee on Appropriations:
There is absolutely no commitment to Denver that we would participate
in such a project. Proceeding with the PRT test project should not be read
as a commitment to that, either. It is quite independent of the Denver
situation, and as I mentioned, Denver is analyzing alternatives right now
under a contract to TRW [the systems analysis firm contracted to lead
25
Burke 1979: p. 181
Ibid: p. 202
27
Ibid
26
13
Denver’s alternatives analysis]. They are looking at all ways in which
they might solve their transportation need and problems.
The results of the TRW analysis are what eventually killed the prospect of a largescale urban PRT system in Denver. Twice in the process, a plan for a true PRT system
was presented and subsequently thrown out. The main reason for PRT’s rejection was
UMTA’s cost-centric feasibility metric, coupled with the assumption that a PRT system
in Denver would need to employ dual-directional guidelines like the system in
Morgantown, and that each 6 passenger pod car would cost the same as a traditional fifty
passenger bus.28 Another reason cited in PRT’s rejection was the public’s concern
regarding guideways in their neighborhoods. In several studies conducted by TRW,
citizens objected to twenty-foot dual-directional guideways with two-foot service
walkways weaving through their neighborhoods. While these specifications describe the
guideways used in the Morgantown GRT system, they are much more intrusive than the
three-foot guideways needed to support a mono-directional PRT system with lighter
weight, six passenger pod cars. Unfortunately, the acceptability of such guideways was
never measured.
The final report of TRW’s alternatives analysis examined six alternatives for the city
of Denver: (1) bus service along existing lines; (2) express bus lanes and other
improvements; (3) light rail streetcars; (4) conventional rail transit similar to San
Francisco’s BART; (5) Advanced Rapid Transit (ART) – a simple automated system with
forty-seat vehicles; and (6) Demand Responsive Transit (DRT) – essentially a GRT
network with offline stations and twenty-seat vehicles.29 As is immediately evident, PRT
did not make it into the alternatives analysis at all. The report suggested, in the end, the
28
29
Burke 1979: p. 207
Ibid: p. 211
14
ART system. Upon receiving the report, the UMTA rejected the alternatives analysis and
refused to fund the ART system. Instead they made a $100 million commitment to
improve the city’s bus services.30 Officials involved in the TRW and UMTA analyses
have described the attitude towards PRT as one of ridicule. Catherine Burke quotes one
analyst from the TRW team, who explained that the organization “felt [that] even to
critique [the PRT proposal] was beneath them, and that the fact of a review might lend
credence to a system that was too ridiculous to discuss.”31
Conclusions Drawn from Denver
The statewide New Jersey autonomous taxi network that this thesis explores at length
would likely encounter many of the same issues as the Denver PRT plan of the 1970s. It
is important, therefore, to understand why an innovative system that was approved by the
democratic process and partially funded by a federal administration was abandoned. One
key reason why the classic PRT model was seen by the analysts as “ridiculous” is that, to
be efficient and effective, a true PRT system needs to serve a good portion of the trips
that a private automobile could just as easily serve. This idea was reasonable to
Bendixson, who introduces the system in his book as a replacement for the car. But to the
government officials and consultants involved with the Denver project, such an idea
seemed ridiculous.
Returning to Joseph Schumpeter’s theory of creative destruction, it is easy to see that
in the mid-1970s, the automobile was still gaining momentum as the nation’s primary
mode of transportation, and had not yet begun to level off, thereby creating an
opportunity for a more efficient method of personal transit to arise. The two graphs below
30
31
Burke 1979: p. 207
Ibid: p. 209 – directly quoting a participant in the Denver Study
15
chart the growth and decline of electric railways, which dominated urban transportation
in the 1900 mobility vignette (Figure 8) and a graph of annual vehicle miles traveled in
the United States from 1970 through 2011 (Figure 9). While the Federal Highway
Administration data for vehicle miles traveled extends only back to 1970, the rise of the
car began in much the same way as the electric streetcar, and as automobiles really took
__
Figure 8: The rise and fall of electric railways32
Figure 9: The rise and leveling-off of vehicle miles traveled33
32
33
Kornhauser 2011
U.S. Department of Transportation Federal Highway Administration 2012
16
off in popularity and affordability in the 1920s, the decline of electric railway trackage is
very evident in Figure 8. The idea, then, of a technology such as PRT replacing the car in
the mid-1970s does seem a bit farfetched, as the automobile’s rise to prominence was still
very much underway. Thankfully for the prospects of an autonomous taxi network as
outlined in this thesis, the vehicle miles traveled figure seems to have leveled off in the
past 6 years or so, indicating that the time may be right for a new technology to take over
the world of personal transportation in the United States.
1.2.2
Train, Bus, and Subway’s Inability to Become Universal
Bendixson was not alone in 1974 as a vocal opponent of the automobile. Experts in
transportation have been concerned about cars’ uncounted costs in the areas of
congestion, safety, and environmental implications for many decades. In some special
cases, alternatives to the automobile have arisen and enjoyed considerable success. New
York City’s subway system carries 55% of the commuting population each day. This rate
of public transit usage is unheard of in America’s less dense cities. Augusta, GA for
example is nearly identical in square mileage to New York City, though its working
population is only 80,335 to New York’s 3.72 million. Accordingly, only one percent of
workers in Augusta commute via public transportation, while 87% commute by
automobile.34 Cities less dense than New York, NY, which is to say most every city in
the United States, simply have been unable to experience New York’s public
transportation success, and have therefore come to be dominated by the automobile.
Many city governments have taken a hard stance on the car due to its uncounted costs,
subsidizing public transportation and refusing new road construction in an attempt to
34
FindTheData 2013
17
force people to turn away from their cars. This strategy has not worked in any major city,
and has been a spectacular failure in some, such as Sacramento, California. In Chapter
Seven, I discuss the implementation of an autonomous taxi network in the state of New
Jersey. If a system is to ever come to fruition, it will be important to learn from the
mistakes of a city like Sacramento when figuring out exactly how twenty-first century
drivers might be successfully lured out of their cars.
Case Study: Sacramento’s Failed Transportation Policies
In Sacramento’s 2006 Metropolitan Transportation Plan, city officials admit that “the
region needs a new transportation vision and plan,” and that “many expectations during
the past 25 years have not worked out.”35 These expectations included the notion that
“lack of road building and the resulting congestion” would encourage people to use
public transit rather than personal automobiles. Congestion in 2006, after twenty five
years of the policy in Sacramento, was worse than ever before, yet still more than half of
the citizens who commuted via transit did so because they did not have access to a car
rather than by choice. In fact the percentage of transit riders without access to a car rose
throughout the 1990s, indicating that public transit was increasingly serving those who
could not otherwise choose to drive. Oddly enough, despite the recognition of these failed
policies, the city’s 2006 plan suggests an expansion of light rail mileage by more than
one hundred percent, and an increase in the bus fleet, the very same systems that
residents have been eschewing in favor of the car for the past twenty five years.36
Furthermore, the financial burden of these expansions falls primarily on the city’s
taxpayers – to the tune of $2.8 billion – since rider fares pay only thirty percent of light
35
36
SACOG 2006: p. 3
Ibid: p. 4
18
rail operating costs.37 For all this investment, the city projects that the mode share of
public transit will increase from 0.9 percent in 2005 to just 1.1 percent by the year
2027.38 The cost figures and expected return on investment don’t look like they should be
found in the same report, but they are. Conventional public transit has only been
successful in a handful of American cities, yet scores and scores of municipal
governments, from Sacramento to the Twin Cities to St. Louis are investing billions of
dollars in transit systems that have proven over multiple decades to not offer any real
challenge to the automobile.
From the policies of Sacramento, it is clear that over the past several decades, cities
have tried to use congestion as a means to an end. In Sacramento, city officials made the
conscious effort to increase congestion, thereby forcing drivers to use alternative systems.
The decision resulted in a congestion problem, as expected, but not the coinciding shift to
public transit – the policy’s main contributions to city life in Sacramento were longer
commutes and increased wear on neighborhood roads used to bypass bumper-to-bumper
highways. If a transportation revolution is to occur in the United States, governments
must learn from the mistakes of Sacramento and realize that conventional modes of
public transit are actually niche-products which have found considerable success in the
New York metropolitan area, Chicago, Boston, and Washington, but nowhere else.39
The role of government when it comes to public transportation should not be to force
citizens into buses and trains that have proven time and time again unviable in most
urban and suburban settings, but rather to invest in a system that shows that it can satisfy
37
Ibid: p. 5
SACOG 2006: p. 29
39
These four areas are the only places in the United States where public transportation serves more than ten
percent of commuters.
38
19
five key transit criteria: 1) The system must reduce congestion and decrease commuting
times; 2) it must be safer than automobiles; 3) it must have fewer negative environmental
impacts than automobiles; 4) it must be economically viable and financially feasible; and
5) it must offer its passengers comfort and convenience to rival the automobile. In
Sacramento’s 2006 Metropolitan Transportation Plan, the city earmarked funds for light
rail, which satisfies only constraints two and three above, a bus system which satisfies
constraints three and four, and highway expansion which satisfies constraints one and
four. Denver’s proposed PRT system in the 1970s satisfied constraints one, two, and
three, may or may not have satisfied constraint five, and was never undertaken because it
did not satisfy constraint four. Over the past fifty years, the car’s challengers have failed
to replace it because of their inability to satisfy each of the five transit criteria listed
above. However, due to recent advances in vehicle autonomy, a truly viable
transportation system that satisfies all five transit criteria may be just around the corner.
Using the state of New Jersey as a subject, this thesis proposes a new system of personal
rapid transit: a network of public, demand-responsive, fully autonomous taxis.
20
Chapter Two: The Problems
Negative Side-Effects of America’s Automobile Addiction
21
The first three transit criteria that a successful challenger to the car must satisfy are
very much in line with Terence Bendixson’s research on the car’s uncounted costs in the
1970s. Breaking these three criteria down one by one, a viable alternative to automobiles
must possess 1) congestion improvements, 2) safety improvements, and 3) environmental
improvements over the car. As such, before presenting a network of autonomous taxis, I
will define the current state of road congestion, safety concerns, and environmental
concerns as they pertain to the automobile, with a special emphasis on the state of New
Jersey where appropriate.
2.1
Road Congestion by the Numbers
Samuel Staley and Adrian Moore’s 2009 publication Mobility First explores
congestion data at the beginning of the twenty-first century, and the trajectory for years to
come. Using a congestion measurement known as the “travel time index” (TTI), the
paper studies time added to commutes due to congestion in the United States’ 86 largest
metropolitan areas.40 A TTI of 1.0 represents free flow traffic in most urban areas, while
an index of more than 1.0 indicates congestion-induced slower speeds. A TTI of 1.5 for
example indicates that during peak hours in a given city, trips take 50% more time to
complete than during free flow. Staley and Moore make sure to point out TTI’s one main
flaw, which is that the United States’ most congested cities such as Los Angeles really
have no “free-flow” speed, as there is congestion at all times. In cases of severely
congested metropolitan areas, the TTI measures relative congestion for peak periods as
compared to non-peak periods.
40
Staley and Moore 2009: p. 14
22
Using a time horizon of the year 2030, Staley and Moore estimate the fluctuation in
TTI from a starting point in the year 2005, assuming no major road construction and a
continuation of basic population trends in each metropolitan area. They define “severe”
road congestion as a TTI greater than 1.3, or annual congestion averaging 40 hours or
more – an entire work week – per commuter. In the year 1982, only one city met either of
these criteria. Unsurprisingly this city was Los Angeles, which met both. By the year
2005, thirty two U.S. cities met at least one of the two criteria, and the average annual
time spent in traffic had grown to 72 hours in Los Angeles, which had a TTI of 1.75. By
2030, an estimated 52 cities are expected to suffer from “severe” congestion, provided
the course of transportation maintains its trajectory as of late. The city of Los Angeles is
expected to see nearly a 100% increase in travel time during peak hours, with a TTI of
1.94. In the two metropolitan areas that most affect transportation in New Jersey – New
York/Newark and Philadelphia – the projected TTIs in 2030 are 1.74 and 1.61
respectively.41
These staggering congestion estimates do make one key assumption that is very
unlikely – the lack of significant roadway improvements and expansion between now and
2030. A 2006 paper by David T. Hartgen and M. Gregory Fields uses the 2030 TTI
projections to quantify the costs of building roads to reduce congestion to tolerable levels.
This approach can be seen as adapting the infrastructure to meet the increased demand of
automobile travelers, and is seen by many as the only viable option to decrease roadway
congestion in the near future. Since no transportation system has arisen that can meet all
five transit criteria for an alternative to cars, the idea behind capacity increases seems a
sound one. The map in Figure 10 shows the costs associated with road building to
41
Hartgen and Fields 2006: p. 7
23
decrease congestion to the point where it no longer meets either definition of “severe,”
broken up into spending on “Urban Interstates and Freeways,” “Other Principal
Arterials,” and “Minor Arterials, Collectors, and Local Streets.” In the map, the state of
New Jersey is entirely obscured by the costs associated with relieving congestion mainly
in metro-New York and metro-Philadelphia, but also to a lesser extent by costs in
Atlantic City and Bucks County, Pennsylvania. In total, Hartgen and Fields estimate that
2,446 lane miles are required to relieve congestion in the New York metropolitan area,
and 1,929 lane miles are required in metro-Philadelphia. The combined cost estimates for
this road construction total $58.2 billion.42
Figure 10: Costs Associated with Construction-Based Congestion Relief43
42
43
Hartgen and Fields 2006: p. 13
Ibid: p. 14
24
Congestion’s problems are bigger than the nuisance of waiting bumper-to-bumper
during rush hour, however. It is no coincidence that as the United States has grown from
an agricultural nation to a post-industrial super power, mobility has continued to increase.
The mobility vignettes in Chapter 1 indicate a strong correlation between increasing
mobility and increasing GDP per capita, as shown in Figure 11. The idea of the “travel
shed” supports this data, and suggests a very serious downside to TTIs in the range of 1.6
to 2.0 as Hartgen and Fields predict. Staley and Moore define the travel shed as “the
geographic area within which commuters will live” around their given destination.44
Public policy expert Ted Balaker calls this area the “opportunity circle.” 45 The bigger the
circle, the more employment, entertainment, and residential opportunities someone has.
The travel shed is determined not by distance, but by time, which explains why mobility
and GDP have increased so drastically since the 1800s.
Mobility and GDP 1800-2000
40000
35000
30000
25000
20000
Annual Mobility (miles)
15000
GDP Per Capita (USD)
10000
5000
0
1800
1850
1900
1950
2000
Year
Figure 11: Mobility and GDP over Time 46
44
Staley and Moore 2009: p. 41
Balaker 2006: p. 1
46
Mobility Data and GDP Data taken from O’Toole 2009
45
25
As travel speed increases, more opportunities exist within a travel shed of one hour.
The benefits of larger and larger travel sheds extend to many different sectors of the
American lifestyle. For one, the increased mobility in the 1950s and 60s allowed urban
workers to move to suburban single-family homes, and still live within a commutable
distance to work. Increased mobility has also led to better healthcare opportunities, a
greater access to food sources, and greater access to employment. In 1800, Princeton
New Jersey’s one hour travel shed extended maybe as far as Lawrenceville, New Jersey.
As of 2013, Princeton’s one hour travel shed includes Philadelphia, PA and New York,
NY, both major centers of business, entertainment, and healthcare.
If the TTI estimates for 2030 come to fruition, the travel sheds of New York and
Philadelphia will shrink to a level where a number of Princeton residents would no longer
choose to work in the two cities, due to increased time in traffic during the morning and
evening commutes. A shrinking travel shed is an inevitable side-effect of severe
congestion, and would result in decreased access to healthcare, jobs, entertainment and
residential opportunities. Most importantly, it would reverse what has until now been the
global trend of the “flattening” of the world – the increased connectedness of all people.
While some cities and states do have the space for added lanes on highways and major
roads as suggested by Hartgen and Fields, other areas such as the northeast corridor lack
the land area required to build their way out of congestion. Furthermore, in the areas
where increased road capacity is an option, it is easily argued that fifty years down the
road the problem of congestion will return, prompting the very same debate, until space
runs out and physical constraints eventually do prohibit the construction of additional
lanes. Therefore the only real, long-term solution to congestion is to build cars that use
26
road space more efficiently, or develop an alternative to the personal automobile, such as
a shared autonomous taxi network.
2.2
Safety Concerns
In the year 2010, motor vehicle accidents resulted in 35,332 deaths in the United
States, according to the Center for Disease Control.47 This figure puts car accidents just
outside of the nation’s top ten causes of death for that year, at number eleven. The
automobile is, year after year, more deadly than Parkinson’s disease, liver disease, and a
multitude of other conditions. Worldwide, car accidents are more common and more
often fatal than in the United States alone. In the 2004 World Report on Road Traffic
Injury Prevention, the World Health Organization projected that by the year 2020 road
traffic injuries will be the third largest cause of death in the world, after only heart
disease and unipolar major depression.48 The root of this increase in traffic-related deaths
is the growth of wealth and access to cars in developing countries, where there are fewer
laws and regulations regarding road safety. The general consensus in these countries, and
in the United States as well, is that traffic deaths are the price society must pay in
exchange for the unparalleled mobility provided by the automobile.
In addition to fatalities, traffic accidents also result in millions of injuries each year in
the United States, and billions of dollars spent on recovery. In 2009 there were more than
2.3 million non-fatal injuries to motor vehicle occupants in the U.S., and the total
estimated cost of these injuries was $70 billion, including the cost of medical care,
treatment, rehabilitation, and lost productivity.49 Despite these staggering numbers, the
47
U.S. Center for Disease Control 2010
World Health Organization 2004: p. 2
49
U.S. Center for Disease Control 2010
48
27
United States has actually seen a decline in annual motor vehicle accident fatalities since
the all-time high of 54,589 in 1972. In the fifteen years leading up to 2006, the fatality
count stabilized around 42,000 deaths annually, and it has fallen to the 35,000 range in
the last five years or so. While this trend is encouraging, the fact of the matter is that,
even with all of the laws and regulations on the road, car accidents are still a leading
cause of death.
That the United States and the world tolerate such danger on the road speaks volumes
for the importance placed on mobility. Civilians and governments alike see automobiles
as so far above the field in terms of mobility that they willingly submit to the danger
associated with them. Although conventional public transit has a much better safety
record than personal automobiles, it does not fulfill enough of the five transit criteria to
be a preferred mode of transport – the same can be said for the traditional PRT model. If
a transit system is to replace the car on a major scale, increased safety must be a
component, but it is far from the only concern. The increase in safety provided by a
revolutionary new transportation system must coincide with a decrease in congestion, as
discussed in Section 2.1, and benefits to the environment, if it is to challenge the carheavy status quo.
2.3
Environmental Issues
The Environmental Protection Agency’s Clean Air Act of 1963 was designed to address
the problem of air pollution in the United States. As such, it identified six criteria
pollutants and set acceptable levels for each pollutant, known as the National Ambient
Air Quality Standards.50 The six pollutants are carbon monoxide, lead, oxides of nitrogen,
50
Giammar 2013
28
sulfur dioxide, particulate matter, and ozone. Of these six pollutants, automobiles
contribute five directly. The sixth, ozone, is formed in the atmosphere by volatile organic
compounds, which are also emitted in car exhaust. Since the instatement of the Clean Air
Act, technological advances in the field of exhaust emissions and fuel burning technology
have resulted in a reduction in each of the six criteria pollutants. Figure 12 shows the
breakdown of criteria pollutant emissions by source in 2012 according to the EPA. Motor
vehicles contributed the majority of the United States’ carbon monoxide and oxides of
nitrogen due to the high carbon intensity of petroleum gasoline, as well as the
characteristics of combustion engines in automobiles, which create nitrous oxides at high
temperatures, and in a fuel lean, air rich environment.51
Figure 12: EPA 2012 Data: Six Criteria Pollutants by Source52
While the clean air act has been successful in reducing the criteria pollutants
associated with automobile exhaust (see Figure 13), a very important compound was not
included in the EPA’s original list. That compound is carbon dioxide, a greenhouse gas
that is a major contributor to global climate change. While technological advances have
51
52
Giammar lecture
Data from U.S. Environmental Protection Agency 2012
29
allowed us to decrease the amounts of criteria pollutants emitted since 1975, the car’s
march to dominance has continued, as shown in Figure 9 which plots the annual vehicle
miles traveled over time, a number that has only just leveled off in the past several years.
This increase in miles traveled, combined with increased time on the road due to
congestion, has resulted in more gasoline being burned, and more carbon dioxide being
emitted into the atmosphere.
Figure 13: Emissions of Carbon Monoxide and Nitrous Oxides since 1975 53
Now that the dangers of carbon dioxide have been identified and made public, it is
evident that the compound must be treated in a similar manner to pollutants such as NOx
and CO. Unlike these pollutants which we have been able to target and control, however,
carbon dioxide is a significant and necessary byproduct of combustion, which cannot be
targeted in the same way as the original six criteria pollutants. The main method of
reducing carbon emissions is by increasing fuel economy standards, so that a car can
53
Data from U.S. Environmental Protection Agency 2012
30
travel further on each tank of gas. Lighter vehicles require less power to move and
therefore consume less gasoline, so they are being heralded as the future of improved fuel
economy. These lighter cars do not achieve record breaking fuel economy without a cost,
however. The price that drivers pay to drive lighter vehicles is the price of safety. A 40
mile per gallon subcompact will not fare well against a 15 mile per gallon sport utility
vehicle in a collision, despite how well it might outperform the gas-guzzler in an
emissions test.
2.4
Culture of Car Ownership
Congestion, safety, and greenhouse gas emissions are some of the car’s biggest
problems. However, these issues alone do not make personally-owned automobiles seem
unattractive or dangerous to the majority of American consumers. Conventional mass
transit often outperforms the car in all three areas. The vast majority of the country’s
trains, busses, and subways run (nearly) on schedule, and even those systems with the
worst records do not take 50 to 75% more time than expected during rush hour, as cars do
in the nation’s largest cities. Conventional mass transit’s safety records are also much
better than automobiles’. In the European Union in 2001-02, there were an average of 7.0
automobile deaths per billion person kilometers, compared to 0.7 deaths per billion
person kilometers on buses and 0.35 deaths per billion person kilometers on trains.54 In
the United Kingdom, the Camden Council estimates that a diesel bus at just 25%
occupancy emits 75 grams of carbon dioxide per person kilometer, and a regional diesel
train emits 105 grams of carbon dioxide per person kilometer at 25% occupancy. A single
54
Folsom 2011: p. 3
31
passenger “small family petrol” car, however, emits 214 grams of carbon dioxide per
person kilometer.55
Despite conventional mass transit’s benefits over the car in these three key areas,
people undeniably prefer to own and use a car. If the shared autonomous taxi network
presented in this thesis is to seriously challenge the present reality of personally owned
automobiles, it will no doubt need to satisfy the first three criteria. However besting the
car in congestion-clearing capability, passenger safety, and environmental issues is only
half the battle, or perhaps less than half. The car rules the road despite its major issues
because it is a convenient, demand-responsive, and affordable means of personal door-todoor transit. If a new transportation system can best the car in these categories, which is
to say achieve criteria four and five as well as criteria one through three, then it may be
the technology that finally dethrones the car as king of the road.
55
The Guardian Data Blog 2009
32
Chapter Three: The Solution
A Shared Autonomous Taxi Network
33
The theme of the 1939-40 New York World’s Fair was “The World of Tomorrow,”
and its most popular exhibit, General Motors’ Futurama, offered a glimpse into the
landscape of the United States in then far-off 1960. The exhibit was comprised of a
35,000 square foot diorama containing cities, suburbs, farms, mountains, and half a
million individual buildings, all linked by what designer Norman Bel Geddes had named
“magic motorways.”56 In a book published shortly after the World Fair’s closing, Geddes
describes the motorways as “transcontinental roads built for a maximum [speed] of one
hundred miles per hour and a minimum of fifty miles per hour,” navigated by “cars that
are automatically controlled, which can be driven safely even with the driver’s hands off
the wheel.”57
Figure 14: An interchange between two “magic motorways” in 1939’s Futurama58
56
Geddes 1940: p. 4
Ibid: p. 8
58
Ibid: p. 98
57
34
The roadways shown in the exhibit (see Figure 14) bear a striking resemblance to the
interstate highways that would begin spanning the nation in the year 1956, save one key
difference: the vehicle autonomy in 1939’s Futurama was not incorporated into the
interstate system. Geddes describes vehicle autonomy on the “magic motorways” as an
“electrical conductor imbedded within the road surface, carrying an electric current
producing an electro-magnetic field,”59 that would control the cars at speeds of 50, 75, or
100 miles per hour, depending upon which lane each vehicle occupied. Furthermore, in
Geddes’s system cars would communicate with one another via radio signals, which
would ensure adequate stopping distance between vehicles. Knowing that the technology
required to build his “magic motorways” already existed in 1939, Geddes expected them
to be a reality 1960, though he noted that it would require modifications to roads and cars
alike before the system would become operational.
At the turn of the millennium sixty-one years after the Futurama exhibit debuted,
Geddes’s “magic motorways” seemed no closer to reality than they had been in 1939,
except for a brief exhibit in 1997 during which a platoon of eight driverless Buick
LeSabres traveled 7.6 miles on a San Diego freeway spaced at 6.5 meters (21 feet) and
guided by magnets embedded in the road.60 Although the demonstration realized
Geddes’s design from decades before, the high cost of embedding magnets in all of the
nation’s highways seemed a daunting and prohibitively expensive task. In the first
thirteen years of the twenty-first century, however, more advances to vehicle autonomy
have been made than ever before, and technologies have arisen that no longer require
roadway modification to achieve total vehicle autonomy. Driverless vehicles that can
59
60
Geddes 1940: p. 75
PR Newswire 1996
35
navigate today’s roads are now legal and operational in Nevada, Florida, Texas, and
California. Autonomous vehicles that do not require dedicated (and costly) guideways
serve as the backbone to a system of autonomous taxis that can meet all five of the
transportation criteria and become a true challenger to the personally-owned automobile
in the first half of the twenty-first century.
3.1
Advances in Vehicle Autonomy
In the year 2002, in response to the United States Congress’s mandate that one-third
of all U.S. military ground vehicles be robotic by 2015, the Defense Advanced Research
Projects Agency (DARPA) initiated the DARPA Grand Challenge, a competition
between fully autonomous vehicles for a $1 million grand prize. When the race was run
in 2004, its track was defined as a 142-mile course through the Mojave Desert, to be
completed in 10 hours or less.61 Carnegie Mellon University Red Team’s vehicle, dubbed
Sandstorm, was the favorite going into the race, and did outlast its 14 competitors in the
main event, traveling 7.3 miles before getting stuck on an embankment.62 Because no
team completed more than five percent of the course in 2004, the million dollar prize
went unclaimed, and a second Grand Challenge was scheduled for 2005, this time with a
prize of $2 million.
The 2005 Grand Challenge saw major improvements over the contest just a year
before, with all but one of the twenty-three entrants travelling farther than Sandstorm had
in 2004, and five cars completing the course in its entirety. The winner was “Stanley,” a
modified Volkswagen Touareg by Stanford University that finished the 132-mile course
61
62
Hooper 2004: p. 1
Ibid
36
in less than seven hours to claim the $2 million prize.63 The follow-up to the two Grand
Challenges was 2007’s Urban Challenge, which brought autonomous vehicles out of the
desert and into an urban environment in which the cars were required to travel 60 miles in
less than six hours while obeying all traffic regulations and sharing the road with other
vehicles, human-controlled and autonomous. Stanford’s vehicle, a modified Ford Escape
named “Junior” crossed the finish line first, but Tartan Racing’s Chevrolet Tahoe called
“Boss” was awarded first prize for better obeying California traffic laws and finishing
just one minute behind Junior. The lead engineer for Stanford University’s “Stanley” and
“Junior” was a man by the name of Sebastian Thrun, who now serves as the director of
Google’s driverless car project.
3.1.1
Google’s Driverless Car
The first autonomous car legally allowed to operate on public roads was a modified
Toyota Prius designed by Thrun’s team at Google, which passed the Nevada state
driver’s test in May of 2012.64 Since 2011, Google’s fleet of autonomous cars – which
now totals upwards of twelve vehicles – has driven more than 500,000 miles
autonomously on roads in all four states where autonomous cars are legal, but mostly in
California, near Google’s Mountain View headquarters.65 Google’s engineers refer to the
Velodyne 64-beam laser mounted on top of the car as the “heart” of their system. 66 The
laser continuously rotates as the car drives, generating detailed 3D maps of the
environment in real time (see Figure 15), and combines these measurements with highresolution maps of the world, carrying passengers to their destination while obeying all
63
DARPA 2005
Slosson 2012
65
Blodget 2013
66
Guizzo 2011
64
37
traffic laws and speed limits. In addition to the Velodyne laser, the car is equipped with
four bumper radars, a camera that detects traffic lights, and an inertial measurement unit
that keeps track of its movements.67
Figure 15: A 3D Scan generated by a Google vehicle in the process of making a left turn 68
Google’s autonomous vehicle team sees the future of vehicle autonomy as a world
where, according to Thrun, “generations will look back at us and say how ridiculous it
was that humans were driving cars.”69 Despite Google’s push toward full autonomy, laws
in all four states where these vehicles are legal require a licensed driver to be in the
passenger’s seat at all times, ready to take the wheel if something goes awry. Despite this
mandated human oversight, evidence exists that the cars are safer when they are driving
themselves. Google’s cars have only been involved in two reported accidents in the
500,000 miles they have driven – in the first collision, a human driver piloting the vehicle
67
Guizzo 2011
Cook 2012
69
Thrun 2011
68
38
rear-ended another car, and in the second incident Google’s car was rear-ended by a
human-controlled vehicle while waiting at a red light.70
In the United States much of the discussion surrounding autonomous vehicles over
the past several years has been focused on Google. In early 2013, Forbes Magazine
published a six-article series by contributor Chunka Mui, entitled “Fasten Your Seatbelts:
Google’s Driverless Car is Worth Trillions,” which estimated that the autonomous
vehicles could be worth up to $2 trillion in annual revenue, while reducing the number of
cars on the road, the wasted energy and time spent in traffic, and the number of casualties
from car accidents, each by a factor of ninety percent. 71 While economists debate the
economic potential of Google’s system and legislators argue laws and liabilities, the
company continues to test its vehicles and amass road data to be used by autonomous
cars in the future.
3.1.2
Automakers’ Forays into Autonomy
Although Google is receiving much of the media attention surrounding driverless
cars, there are a number of other players in the field of vehicle autonomy in the U.S., and
even more abroad. Many of the companies are in fact automakers that, while tech giant
Google jumped headlong into full autonomy, have been focusing on incremental
improvements and driver-assist systems for years. These systems, such as the “adaptive
cruise control” feature offered by BMW, Mercedes Benz, Audi, and others, incorporate
much of the same technology as Google’s vehicles, just on a smaller scale that is already
commercially available.
70
71
Yarow 2011
Mui 2013: p. 1
39
In January of 2013, Japanese auto giant Toyota announced that it had equipped a
Lexus LS with forward- and side-facing radar sensors, a 360-degree laser scanner and an
onboard computer that operates the car’s controls. Dubbed the Advanced Safety Research
Vehicle, the car’s development and testing is motivated by a desire to “eliminate future
traffic-related fatalities and injuries,” according to Lexus general manager Mark Templin.
In addition to this safety-driven research, Templin added that Toyota does not see
“autonomous” as being synonymous with “driverless,” and that the features that already
exist in Lexus vehicles, such as radar-based collision avoidance, are only the first step in
a “layered introduction of automated technologies,” that will result not in a driverless
Toyota automobile, but a “skilled, intelligent, and attentive copilot.”72 German automaker
Audi echoed this sentiment, using the phrase “piloted driving” when talking about future
cars that use autonomous technology. Said Audi electronics developer Richard Hudi, “we
speak of piloted driving because, like with a plane, the ultimate responsibility rests with
the pilot, the driver.”73 This notion of responsibility, or liability, is a very important one,
which will be discussed at length in Chapter 7 as one of today’s primary barriers to full
autonomy.
American automaker Ford has differentiated itself from foreign competitors Audi and
Toyota in two key ways when it comes to autonomous vehicles and predictions for the
future of cars. The first difference is that while Toyota currently offers driver assist
technologies in its luxury brand Lexus, and Audi in its higher-end models, Ford is the
first company to incorporate such technology in its mass-market vehicles.74 The 2012
Ford Fusion, an affordable midsized sedan, and Explorer, a full-size crossover utility
72
Simonite 2013: Toyota Unveils an Autonomous Car
Simonite 2013: Audi Shrinks the Autonomous Car
74
Fitchard, August 2012
73
40
vehicle, offer Ford’s “Lane Keeping System” which tracks lane boundaries and, in the
event of an unintentional lane departure, alerts the driver by vibrating the steering wheel
and automatically repositions the car to the center of the lane.75 This mass-market
introduction of small-scale automation is an effort on Ford’s part to make drivers more
comfortable with a car that can “do more.” The decision came after consumer clinics and
driver questionnaires by Ford found that people are still uncomfortable with the idea of
fully autonomous cars, but are “very open to the idea of their cars becoming more
intelligent and aware.”76 Ford’s vehicle engineering supervisor for driver assistance
technologies Mike Kane explains that while incorporating driver assist into all Ford
models is a good first step, “it’s going to take a decade before the masses fully accept the
autonomous car, but they’ll get there.”77
People “getting there” is a requirement given executive chairman Bill Ford’s
predictions about the future of mobility. At the Mobile World Congress in February of
2013, Ford outlined a future world in which there will be four billion cars on the road, as
compared to today’s one billion. If any current automakers hope to remain afloat and
successful in such a world, he argued, they would have to be open to the idea of a fully
autonomous network of vehicles constantly communicating with one another, as well as
with traffic signals, pedestrian crosswalks, and bicycles, as part of a “giant grid of multimodal transit intelligence.”78 The idea of a large, fully integrated transportation network
is very much in line with the research and development Google has been doing and
75
Ford Motor Company 2011
Fitchard, August 2012
77
Ibid
78
Fitchard, February 2012
76
41
would serve as the backbone for the fully autonomous taxi network explored in this
thesis.
3.2
An Autonomous Taxi Network
Although Ford’s Mike Kane estimates that the people of the United States will not
accept driverless cars for at least a decade, and Bill Ford’s time horizon for a multi-modal
transit grid of four billion vehicles is even further into the future, the benefits of a
network of self-driving cars will be very real if the technology comes to fruition on a
large scale. A network of driverless vehicles sharing information and optimizing traffic
flow lends itself to an Autonomous Taxi Network (ATN). An ATN can be defined by two
key characteristics. First, it consists of a number of fully autonomous, constantly
communicating vehicles – the taxis – which drive passengers to their requested
destinations. Second, the taxis are demand-responsive, which is to say they do not operate
on a regular schedule like a bus or train, but rather only run when a passenger has
indicated demand. As long as a system obeys these two guidelines, it can be classified as
an ATN. Chapters Five and Six focus on a number of different variables that can be
altered within the structure of an ATN and how those differing systems would operate in
New Jersey. The adjustable variables include decisions such as whether or not the taxis
will be able to automatically reposition themselves while empty. These variables will
affect how well the system satisfies transit criteria four and five, the economic viability of
the system and comfort-convenience factor, but criteria one, two, and three will be
satisfied by any system classified as an ATN.
42
3.2.1
ATN’s Improvements to the Congestion Problem
Congestion was a growing issue in American cities in 1939, and one of the major
motivations for Norman Bel Geddes’s Futurama project. In a world of four billion
vehicles, the planet’s only hope to avoid massive congestion would be the kind of
intelligent transit grid that Bill Ford proposed at the Mobile World Congress. An ATN by
definition is built on a foundation of interconnected autonomous vehicles; research by
Columbia University’s Patcharinee Tientrakool in 2011 suggests that a highway
populated exclusively by driverless, communicating vehicles could be up to 273% more
efficient than today’s roads. The left hand graph in Figure 16, taken from her research,
shows safe vehicle headway distances at varying speeds in three cases: 1) when all cars
are manually driven, 2) when all vehicles have sensors, and 3) when all cars are
connected, autonomous, and communicating. Scenario 3) would be the case if an ATN
were the only mode of transportation on a given road, or if ATN vehicles communicated
with any and all vehicles on the road that did not belong to the ATN.
Figure 16: Tientrakool’s Vehicle Headway and Highway Capacity Graphs 79
79
Tientrakool 2011
43
At typical highway speeds of 65 miles per hour, or approximately 105 kilometers per
hour, vehicle headway in an ATN would be, on average, 5 meters. In contrast, the current
system of manually operated vehicles traveling at that speed requires a vehicle headway
distance of approximately 30 meters. These distances correspond to 0.17 seconds of
headway in the ATN and 1.1 seconds of headway in the conventional model, and result in
the total highway capacities shown in the right hand graph. While highway lanes today
have an approximate capacity of 3,000 cars per hour at a speed of 65 miles per hour, an
ATN with the same speed would experience a capacity of around 11,000 vehicles per
hour per lane, a 267% increase. While less dramatic than this near tripling of highway
capacity in an ATN, local roadway capacity would also increase 180% at speeds of 30
miles per hour.
The increased efficiency offered by an ATN would be sufficient to eliminate the
growing Travel Time Indices (TTIs) in major cities like New York and Philadelphia, as
discussed in Section 2.1. Even in Los Angeles, where the TTI in 2030 is estimated to
approach a value of 2.0, a network in which 100% of cars are communicating with one
another would result in congestion-free roads. In addition to decreased vehicle headway,
the concept of ridesharing also has the potential to reduce congestion in a major way.
Ridesharing in an ATN is defined as an instance in which two or more people
simultaneously occupy the same vehicle due to similar origin and destination locations
for trips taken at approximately the same time. The key benefit of ridesharing is that it
results in a decrease in the total number of vehicles required to meet all demand in an
ATN. The concept of ridesharing and its place in a statewide New Jersey ATN are
discussed at length in Chapters Five and Six.
44
3.2.2
Safety Improvements in an ATN
That Google’s twelve driverless vehicles have driven over 500,000 miles
autonomously without causing a single accident speaks to the safety benefits of driverless
cars, and an ATN’s ability to meet the second transit criterion. The Velodyne laser
mounted on the roof of every Google car completes 10 revolutions per second, gathering
data 360 degrees around the car. Naturally this system is able to “see” much more than
the average, or even the best, human driver. Furthermore, the car’s “reaction time” is all
but immediate, and it lacks the basic human ability to fall asleep behind the wheel, get
distracted, or drive under the influence. In fact, safety benefits were one of the major
reasons that Google’s Sebastian Thrun decided to devote his life’s work to the
advancement of autonomous vehicles.
Driverless cars on the road today must approach collision avoidance probabilistically,
as they are interacting almost exclusively with unpredictable human drivers. A paper by
Matthias Althoff et al. in 2007 used Markov chains to model human-controlled cars from
the perspective of an autonomous vehicle. Given the scenario shown in Figure 17, the
___
Figure 17: Althoff’s Experiment80
80
Althoff 2007: p. 2
45
autonomous car must swerve to avoid the static obstacle as well as the approaching
human-controlled car. Preparing for the worst-case scenario, the research assumes that
the red car moves only parallel to the road’s longitudinal y-axis, which is to say it will not
swerve from the center of its lane regardless of the autonomous vehicle’s path.
Figure 18: Probabilistic Rendering of Althoff’s Experiment at Four Discrete Time Steps81
The Markov chain model for the human-controlled car includes four discrete states:
1) standstill, 2) driving the speed limit, 3) breaking, and 4) accelerating. The autonomous
vehicle is able to track which state the human-controlled vehicle is in at any point in time
given its relative position, and is thereby able to calculate the probability that the humancontrolled car will stop completely, coast, apply the break, or accelerate at each discrete
time step based on the model’s transition probabilities. The four discrete time steps
shown in Figure 18 represent a case in which the autonomous vehicle would calculate
that it had enough time and space to clear the static obstacle without colliding with the
81
Althoff 2007: p. 6
46
oncoming car. The road area is discretized into cells, which are shaded blue based on the
relative probabilities of the car occupying that cell at the time step shown. In panel (d),
the closest the two cars come to a collision, the vehicles occupy the same cell with a
probability of zero. If the probability of a collision were greater than zero, the
autonomous vehicle would not proceed around the obstacle until the oncoming car had
passed.
Autonomous vehicles can employ this kind of sophisticated decision-making and
probabilistic modeling in a fraction of a second, while even the most alert human driver
would be forced to make a decision given incomplete information, and potentially cause
an accident. According to John Maddox of the National Highway Traffic Safety
Administration, human error is “the critical reason for 93 percent of [automobile]
crashes.”82 By eliminating the possibility for human error, driverless vehicles have the
potential to decrease car accident fatalities by orders of magnitude.
On top of the inherent benefits of a driverless vehicle interacting with unpredictable
human drivers as in Althoff’s Markov-chain-based model, the safety benefits of an ATN
are even more promising. In a system where all vehicles are constantly communicating,
there is no need for probabilistic models – the network becomes almost completely
deterministic. In an ATN, the vehicles all share one mind, so to speak. In the event of an
unexpected event, such as a tree falling into a roadway during a storm, the “decisions” of
every car on the road would be made simultaneously, with all vehicles knowing with
100% certainty the other vehicles’ next moves. No transportation system will ever
operate without incident, but the ATN’s removal of human error through the use of
82
Flammang 2012
47
autonomous vehicles makes it a significant improvement over the current system in the
field of passenger safety, thereby satisfying transit criterion two.
3.2.3
Environmental Improvements as a Result of an ATN
In addition to providing a less congested and safer method of transportation than the
current system of personally-owned manually-operated cars, an ATN also satisfies the
third transit criterion: it promises less of a negative impact on the environment. In the
past two decades, researchers and automakers alike have made great strides in the fuel
economy of combustion engines, as well as the development of hybrid gas-electric
powered vehicles, some of which report gas mileage as high as 47 miles per gallon.83
However, the reality of combustion engines is that they do need to burn fuel to provide
the power that propels the car forward, taking passengers from point A to point B. This
power requirement is equal to the sum of energy changes needed to counter rolling
resistance and aerodynamic drag. The equations below come from a 2011 paper by Tyler
C. Folsom on the subject of autonomous urban land vehicles such as the taxis in an ATN.
Equation (1) defines the change in energy needed to overcome rolling resistance
Equation (2) is the expression for the amount needed to counter aerodynamic drag
83
84
Sabatini 2012
Folsom 2011: p. 4
48
while
.84
The advances made in fuel economy thus far have been mosty due to increases in
from Equation (1), the mechanical efficiency of the transmission. The main improvement
an ATN can provide to reduce
is a reduction in
, the total mass of the
vehicle. Because of autonomous vehicles’ major safety benefits discussed in Section
3.2.2, the need for heavy cars that better survive accidents would disappear in an ATN.
Especially in a network in which every vehicle is communicating, collisions with other
vehicles would be exceedingly rare, and as a result a large SUV would have almost no
safety benefit over a much lighter, smaller car. Furthermore, an ATN would also offer a
reduction to
, as shown in Equation (2). The value of
varies from car to car,
as does the value of , based on the design; the product of the two values,
, is known
as “drag area.” A car with a low drag area value requires less power to overcome
aerodynamic drag than one with a large drag area. Of cars currently on the road, one of
the top performers is the Toyota Prius with a drag area of 6.24 square feet. A Hummer
H2 on the other hand will have a
of more than four times that amount, around 26.5
square feet.85
An ATN not only allows for lighter vehicles with more aerodynamic designs, it also
opens up the possibility for much closer following distances as explained in the
Tientrakool paper. This has an even greater affect on
than replacing all cars on
the road with Priuses – it allows vehicles to drive in the slipstream of other vehicles, at
85
Butts 2012
49
distances of less than two tenths of a meter. According to 2012 research by Kloiber et al.,
one car following another at a speed of 80 km/h (approx. 50 mph) and a distance of one
meter experiences a 93% reduction of its drag coefficient
.86 Platooning is a term for
running two or more vehicles at very close distances to reduce drag, and is shown in
Figure 19. Because all vehicles communicate with one another in an ATN, platoons such
as the six-vehicle example shown would be employed extensively in an effort to use road
space more efficiently and decrease drag. This decreased drag, combined with the lighter
Figure 19: Visualization of five sedans platooning behind a large truck 87
weight of the vehicles, would result in a significant reduction of power required to propel
those vehicles; the result would be a transportation system that burns significantly less
fossil fuel per passenger mile than today’s cars, thereby reducing the harmful emissions
associated with petroleum gasoline. The key factor in an ATN that allows its vehicles to
use road space more efficiently, avoid accidents, and use less fuel is the interconnected
nature of the network. As long as the system can operate as a single, intelligent entity
rather than tens of millions of independent parts, any ATN will satisfy transit criteria one,
two, and three.
86
87
Kloiber et al. 2012
O’Kane 2009
50
Chapter Four: The Demand
Mapping New Jersey’s Trips on an Average Work Day
51
Chapter Three shows that an ATN, simply by consisting of interconnected fullyautonomous demand-responsive vehicles, satisfies transit criteria one, two, and three. The
system’s ability to fulfill criteria four and five, the economic viability and comfortconvenience factor, depends more on the specifics of the network. These specifics
include the network’s scope and the logistical methods used to organize such a complex
and interconnected transportation system. In Chapters Five and Six, I focus on two
distinct models for the ATN, the first being the PRT model, similar in many ways to the
classic Personal Rapid Transit design and the plans for Denver discussed in the case
study in Section 1.2.1. The second model derives some of its basic layout from the work
of Mark Gorton in 2008 regarding a system he named “Smart Para-Transit”; as a result I
have named my second model the SPT model. The specifics of these models are
described at length in Section 5.1, and the two are analyzed and compared using a
number of parameters including estimated daily cost and the consumer behavior required
to make the system work on a large scale.
My analysis of each model assumes that the given ATN is operating at maturity,
rather than at an incipient stage where market share is low. In addition, the network I
have designed for each model is a state-wide system. As a result of these two
assumptions, the analysis performed in Chapters Five and Six requires input data that
successfully mirrors the comings and goings of New Jersey’s 8.8 million residents on an
average business day. Luckily the generation of such data has been underway in
Princeton University’s Operations Research and Financial Engineering department over
the past two years, in the form of two major undertakings, the first being Synthesizing
Individual Travel Demand in New Jersey, a report compiled by Professor Alain
52
Kornhauser and the members of his Transportation Systems Analysis class in the fall of
2011, and the second entitled Synthesis of Spatially & Temporally Disaggregate Person
Trip Demand: Application for a Typical New Jersey Weekday, a Master’s Thesis by
Princeton graduate student Talal Mufti.
The idea behind both projects was to “integrate large amounts of demographic,
employment, industry, school, and human behavioral data to create a high-resolution
snapshot of travel demand, via each individual trip made by each individual [New Jersey]
resident and each individual out-of-state commuter that works in New Jersey.”88 For the
purpose of generating the data used in Chapters Five and Six of this thesis, I have
employed Mufti’s six-module approach, with very few updates and alterations. The most
significant deviation from Mufti’s model is the use of updated patronage and employee
data for New Jersey’s many employers and trip destinations. The patronage-employee
data I have used was manually updated by Dr. Kornhauser, who repaired several outliers
and glaring errors in the data that caused unfeasibly large amounts of trip attractions in
some areas of the state. Detailed flowcharts of Mufti’s six modules can be found in
Appendix A, but in this chapter I have separated the process into the four main steps I
have used to generate the 8,791,894 New Jersey residents and 262,955 out-of-state
commuters used in my model.
4.1
Step One: Generating Individuals
The first step in generating each and every trip that takes place in New Jersey during
an average business day is to generate the people making the trips. In the case of Mufti’s
model, that means generating twelve data points for each individual person, county-by-
88
Mufti 2012 p. 8
53
county. A sample output from Atlantic County is shown in Figure 20. In this file, each
row corresponds to an individual who resides in Atlantic County, and contains twelve
identifying cells. Explanations for these twelve cells can be found in Table 1.
Figure 20: Output of Person Generation in Atlantic County
Res County
HH ID
HH Type
Res Lat
Res Long
Person ID
Age
Sex
Traveler Type
Income Bracket
Income
Work County
County of residence, listed in alphabetical order
Household identification number within a county
Household type, ranging from 0-8; See Table 2
Latitude of the centroid of the household’s census block
Longitude of the centroid of the household’s census block
Unique identifier for each individual
Chosen based on state demographics
Chosen based on state demographics; M = 0, F = 1
Traveler type, ranging from 0-7; See Table 2
Income bracket ranging from 0-9; See Table 2
A personal income within the range specified above
County of employment, listed in alphabetical order
Table 1: The Twelve Values Generated in Step One
In the sample output shown in Figure 20, for example, we can focus on the person
with ID tag ATL00000082, a 34-year-old man. Based on the integer codes given in Table
2, this man is a part of a family composed of two people, presumably him and his 36-year
old wife or girlfriend who live in a census block in Atlantic County centered at latitude
and longitude (39.3549, -74.4583). This particular man is a working adult with traveler
54
type 5 in income bracket 5, who earns a total annual income of $46,284 at his job in
Atlantic County. His female partner on the other hand does not work in Atlantic County,
but rather in Bucks County, Pennsylvania, which corresponds to the integer value 44 seen
in the last cell of the “Work County” column.
Traveler Types
0 Do-Not-Travel
1 School-No-Work
2 School-Work in County
3 College-No-Commute
4 College-Work in County
5 Working Adult
6 At-Home-Worker
7 Out-of-State-Worker
Household Types
0 Family
1 Non-Family
2 Correctional Facility
3 Juvenile Detention
4 Retirement Home
5 Institutionalized Homes
6 Collegiate Dormitory
7 Military Quarters
8 Other
Income Brackets ($)
0
No Income
1
Up to 10,000
2
10,000 - 14,999
3
15,000 - 24,999
4
25,000 - 34,999
5
35,000 - 49,999
6
50,000 - 74,999
7
75,000 - 99,999
8 100,000 - 149,999
9
150,000 or more
Table 2: Integer Codes for Traveler Types, Household Types, and Income Brackets
For each of New Jersey’s 21 counties and seven out-of-state locations, I have
generated a person file like the one shown in Figure 20 to use in my analysis. The input
data for these files, generated using Mufti’s Module 1 code, comes from the 2010 United
States Census. The data is broken up into census blocks, and includes each block’s
population by sex by age, population in households by age, household type by
relationship, and household size. As a result, the data generated in Step One almost
perfectly replicates the number of people of every age and gender as reported for each
census block. The data for traveler type is not available on a census-block-by-censusblock basis, so the statewide distribution is sampled for each individual based on age and
household type. Similarly, income bracket data is only available by census tract, a
collection of census blocks that generally coincides with city or town boundaries. A
resident of a given census block is assigned an income bracket based on the income
55
distribution in his or her census tract.89 The end result of this first step is a collection of
28 person files – one for each county or out-of-state location – which are used as the
basis for the final product: the 28 trip files used in my analysis.
4.2
Step Two: Assigning Primary Role Locations
Given the lists of people by county generated in Step One, the task of Step Two is to
assign to each resident the geographic location of their primary role in society, to and
from which they would travel on a typical week day. In the case of this research, the three
main types of individuals are workers, students, and neither. Thus, the primary role
locations for these types of people are their workplace, school, or home respectively. In
the case of high school or college students who also work, both the individual’s school
and workplace are considered primary role locations.
Figure 21: Step Two Output for Atlantic County for the Individuals shown in Figure 20
A sample output for Step Two is shown in Figure 21, using the same individuals
shown in Figure 20. Examining the couple discussed in Section 4.1, the 34-year-old male
ATL00000082 works in real estate rental and leasing, denoted by industry code 10. This
integer value is generated for each individual using data from the 2010 American
Community Survey, which publishes exact numbers of workers in each county for each
89
For a full explanation of the person generation process, see Mufti 2012
56
industry sector, along with median incomes and gender proportions for that sector. Each
individual’s exact income, as shown in Figure 20, is used to determine their work
industry via a discrete distribution built using Equation (3).90
(3)
In Equation (3),
is the number of workers in industry i in a given county,
and each individual’s attraction to that industry is weighted by the squared inverse of the
difference between the individual’s income and the industry median. After a work
industry is chosen for an individual via the distribution generated from Equation (3), the
individual’s exact workplace is chosen from within their work county generated in Step
One. In the case of our 34-year-old man, his Atlantic County real estate employer was
chosen to be the “Plaza Condominium Association,” located at latitude and longitude
(39.3448, -74.4630), a distance of 0.74 miles from his home. The close proximity of his
workplace to his residence is a function of the attraction equation for each individual to
their place of employment, as shown in Equation (4).
(4)
In much the same way that workers are assigned industries before they are mapped to
individual workplaces, students are assigned a school type based upon their age and
statewide enrollment data. Once they have been divided into public and private
elementary, middle, and high schools, as well as commuter and non-commuter
universities, individual schools are chosen based on one of two methods. The first
90
Equations from Mufti 2012
57
method is applied for all public elementary, middle, and high schools, and is a simple
great circle distance minimization, as shown in Equation (5).91
(5)
This simple assignment rule is in fact very much in line with the reality of the
situation, as the vast majority of school districts are zoned such that public school
children attend the school closest to their homes. Private schools and universities function
more similarly to the employment choice method, which assumes that residents will often
but not always prefer closer jobs to those further afield, and models that choice using a
weighted distribution that varies with the inverse of the squared distance. In fact, that is
exactly the method used to assign private school children and college students to their
primary role locations, as shown in Equation (6).
(6)
In the example output shown in Figure 21, the two students are girls aged 14 and 17,
located in rows 73 and 74. Being both of school type 2, they are assigned to their nearest
public high school, Oceanside Charter High School. The assignment of primary role
locations is fundamental to the generation of New Jersey’s daily trips, as each
individual’s workplace or school becomes a very important location during their daily
activity. However, very few people only commute to and from work or school – there are
91
The calculation of great circle distance, denoted in my equations as the function
, is meant to approximate the distance in miles between
two points on the earth’s surface, as indicated in the following equation:
⁄
√
58
many other trips made in New Jersey at a variety of times for a variety of reasons. These
trips are generated in Step Three, and are referred to as “secondary trips.”
4.3
Step Three: Generating Secondary Trips
While assigning primary role locations is relatively straightforward due to the daily
nature of traveling to one’s workplace or school, the assignment of secondary trip
locations is considerably more complicated. In his analysis, Mufti uses 18 different “tour
types” as possible travel patterns for New Jersey’s nine million residents. The tour types
range in number of trip ends – or locations visited on the day in question – from zero to
seven, and are shown in Figure 22.
Figure 22: Mufti's eighteen tour types
In the Visual Representations columns, H corresponds to an individual’s home, W to
their workplace, S to their school, and O to any other – or secondary – trip locations.
Tour types zero through four correspond to people who, on this particular average
59
weekday, simply visit their primary role location(s) before returning home. The
subsequent thirteen tour types include one or more O destinations, which could
correspond to lunch restaurants or mid-day errands, as in tour type eleven, or after-work
recreation and shopping, as in tour type five, among many other possibilities. For a
detailed probability distribution of tour type by traveler type, please see Appendix A. In
the case of stay-at-home individuals, tour types with W destinations were selected, and
all W values replaced with O, thereby generating only secondary trips for those
individuals.
Once a tour type has been assigned to each individual, the destination of every O trip
is specified by drawing a selection from a distribution constructed via the weighting
method shown in Equation (7). Given an individual’s location at the time that the O trip
is taken, the model chooses from any of the establishments in the county of current
location and any bordering counties based on the patronage data that was used by Mufti
and subsequently updated by Dr. Kornhauser. Furthermore, lunch time trips in tour type
eleven are limited to a distance of five miles or less, based on typical behavior and time
constraints.
(7)
An example of Step Three output can be found in Figure 23, for the 34-year-old man
ATL00000082 and his 36-year-old female partner ATL00000083 discussed in Sections
4.1 and 4.2. The man has been assigned tour type 16, which has seven trip ends including
three secondary O trips. The woman has been assigned tour type 5, which includes just
one O trip, that is made on her way home from work in Buck’s County, Pennsylvania. As
60
a result, the Step Three output for ATL00000082 includes Nodes 0-7, for each of which
there are cells identifying whether the node can be identified as H, W, S, or O, which
county the node is in, the node’s pointer in the data file, the latitude and longitude, the
industry type or school type where applicable, and the distance from the previous node.
ATL00000083 on the other hand has values for only Nodes 0-3, including an O trip in
Ocean County, which borders her home county of Atlantic and is identified with integer
code 29.
Figure 23: Step Three Output for ATL00000082 and ATL00000083
Given the output of Step Three for every county and out-of-state region, the trip files
are nearly complete, as they include the origin and destination for every trip made by
every individual who lives or works in the state of New Jersey. The only missing element
61
is the temporal one – each trip needs a time stamp associated with it. This is achieved in
Step Four.
4.4
Step Four: Adding the Temporal Domain
In my analysis in Chapters Five and Six, I use the trip files generated by Step Four to
measure the viability of the PRT model versus the SPT model for an autonomous taxi
network as it pertains to system cost and the comfort-convenience factor. One of the key
factors that make this data exceedingly useful is the inclusion of the exact time that each
trip begins and ends. Mufti’s code assumes an average speed of 15 miles per hour for
trips to schools and 30 miles per hour for all other trips, and uses these values along with
calculated trip distances to generate arrival times for O and H trips given their departure
times. For all S and W trips, the arrival time is chosen via a triangular distribution with a
given minimum, maximum, and mode. An example of morning arrival times for three
different industries of work is shown in Figure 24. Jobs in the agriculture sector require
an arrival time at work that varies between 6:00am and 6:45am, while utilities jobs start
Morning Commute Input CDF
1
0.8
0.6
Agriculture
0.4
Utilities
0.2
Administrative
0
Arrival Time at Work (AM)
Figure 24: An example of morning commute times for 3 industries
62
between 7:00am and 8:15am. Administrative jobs, on the other hand, are not as specific
in their starting times as the first two examples, and range from 6:00am until 9:00am,
resulting in a smoother cumulative density function in the figure. These input
distributions are almost precisely replicated by an empirical CDF generated from Atlantic
County’s workers in the given industries. The empirical CDFs of arrival times generated
for agricultural, utilities, and administrative workers are shown in Figure 25.
Figure 25: Output Realization for Atlantic County Morning Commute
School starting times, as well as ending times for both work and school, are chosen
from triangular distributions like the three in Figures 24 and 25, while O trip times are
generated randomly based on where they fall in the day. The result of these various
temporal generation strategies can be seen in Figure 26, which corresponds to the final
trip file output for ATL00000082 and ATL00000083, the couple from the previous
examples. The columns “Node N: Arr Time” and “Node N: Dep Time” have been added
63
to the output from Step Three, in units of seconds after midnight. It is this final trip data
that serves as my input in Chapters Five and Six, expanded to include not just the two
individual subjects that have been discussed thus far, but rather all nine million residents
and workers in the state of New Jersey.
Figure 26: Step Four Output for ATL00000082 and ATL00000083
64
Chapter Five: The Models
An Introduction of Two Potential ATN Designs
65
In the preceding chapters I have defined an Autonomous Taxi Network as any
transportation system which obeys the two criteria defined in Section 3.2. They are 1) the
system consists of fully autonomous, constantly communicating vehicles and 2) the taxis
are demand-responsive, and only operate when a passenger has indicated demand. Given
these two specifications, an ATN can take on any number of arrangements in practice.
Furthermore, while any system that satisfies the two definitional qualities of an ATN is
an improvement over the current system for transit criteria one through three – as shown
in Sections 3.2.1 through 3.2.3 – the additional specifications that can be made affect how
well the ATN will satisfy transit criteria four and five. In this chapter I will define two
distinct arrangements that an ATN might take on: the PRT model and the SPT model.
Given the specifications of these two models, I will then discuss how each can be set up
as a classic transportation problem to determine the optimal fleet size and total cost
associated with operation, which in turn will determine some of the financial
requirements of each system.
5.1
Defining the PRT and SPT Models
One key parameter that must be defined when designing an Autonomous Taxi
Network is the method by which travelers are picked up and dropped off by the vehicles.
In their 2013 paper Uncongested Mobility for All, While Improving Safety, Energy, and
Environmental Consequences: New Jersey’s Area-wide aTaxi System, Kornhauser et al.
design a system that borrows its layout from the classic implementation of a Personal
Rapid Transit, or PRT network. This model establishes stations – or taxi stands – across
the state of New Jersey, spaced 0.5 miles apart from one another. The PRT model
assumes that passengers will walk to their closest station, which is at most 0.35 miles
66
away. At a typical human walking speed of 3 mph, this corresponds to a seven minute
walk at the absolute most, though the majority of passengers would require five minutes
or less to travel to their nearest station. Similarly, at the end of their trip, passengers
would disembark at a station and walk to their destination, again a maximum distance of
0.35 miles away. In the PRT model for an ATN, two riders will take the same taxi if their
origin and destination locations exist within the same 0.5-mile-by-0.5-mile pixels as one
another, and they arrive at the taxi stand within
seconds of one another.
The second model which this thesis explores derives its set-up from a 2008 report by
Mark Gorton which introduces a transportation mode called Smart Para-Transit. Figure
27 shows an example of how a Smart Para-Transit system could condense twelve
individual trips from northern New Jersey to Manhattan into just two SPT trips. Gorton’s
system assumes that the vehicles used will be operated by human drivers, but they could
just as easily be substituted by autonomous taxis in an ATN. The basic idea behind
Gorton’s SPT system is that individual people request a trip to a given destination, at
which point they are picked up by the SPT vehicle at a “central transit point.” Along the
way, the vehicle may stop at one or two other “central transit points” to pick up more
passengers. The drop-off works similarly to the pick-up, with the vehicle stopping at one
to several “central transit points” to drop off its passengers close to their final destination.
The SPT model for an ATN allows the pixels of the statewide transit grid to become
larger, as the vehicle can move around within the origin pixel to pick up multiple
passengers before heading to the destination pixel. In an SPT model ATN, “central transit
points” can even be discarded, because the autonomous taxis can drive themselves to the
passengers’ doorsteps, and let these passengers off at the doorsteps of their destinations.
67
In this way, the vehicle takes the place of the individual for intra-pixel travel. While the
PRT model requires its users to walk up to 0.35 miles to the nearest station, the SPT
model has the vehicle moving around, gathering up passengers before a trip departs. Not
only is this a benefit to the passengers, who exert less energy to get to their taxis, it is also
allows for a major increase in pixel size. While two people who live 0.6 miles from one
another with a taxi stand directly in between them would have to walk for six minutes
each in the PRT model prior to their trip, an autonomous taxi in the SPT model would be
able to pick up two passengers up to 2.5 miles apart in the same six minutes, assuming an
average speed of 25 mph.92 In the SPT model, I have increased the distance between taxi
stands to 1.5 miles, meaning that the maximum distance to the center of the pixel
increases to 1.06 miles.
Figure 27: Representation of Mark Gorton's Smart Para-Transit System93
92
Walking: (
93
Gorton 2008: p. 3
⁄
)
Driving: (
68
⁄
)
5.2
Dividing New Jersey into a Pixelated Transit Grid
For both the PRT model and the SPT model, the state of New Jersey and its
surrounding areas must be broken up into a grid in which each pixel has a side length of
. In the case of the PRT model,
model
is equal to 0.5 miles, and in the case of the SPT
is equal to 1.5 miles. Rearranging the formula for great circle distance found in
footnote 90 on page 58, the X and Y coordinates for any point P located at
are calculated using Equations (9) and (10). In this analysis, the origin point O is located
at (38.0, -76.0) or (38°N, 76°W), the bottom left corner of the orange square in Figure 28.
(
)
Figure 28: Map of New Jersey area showing boundary coordinates for the PRT Model
69
The resulting grid that is formed via these equations can be seen in Figure 29,
overlaid atop the Princeton area. The pixels in the figure have side length
= 0.5, a
representation of the PRT model ATN, but the layout of the SPT model grid simply
combines each three-pixel-by-three-pixel square from the PRT model into a single pixel.
Once the state has been divided up into a grid of pixels, the trip files generated and
discussed in Section 4.4 can be broken down into individual trips from an origin pixel
(O_X, O_Y) to a destination pixel (D_X, D_Y) at a given departure time
. In the data
set I have generated for use in this thesis, there are a grand total of 32,770.528 trips taken
by 9,054,849 individuals, for an average of 3.62 trips per person. Once each trip has been
assigned an (O_X, O_Y), (D_X, D_Y), and
, the next step is to order the file by
,
then by (D_X, D_Y), and finally by (O_X, O_Y), resulting in an ordered trip file for the
entire state, which can be seen in Figure 30.
Figure 29: PRT Model Gridding Overlaid on the Princeton Area94
94
Figures 28 and 29 are taken from Kornhauser et al. 2013, in which the PRT model ATN was analyzed
70
Dep.Time O_X
66099.33
61999.55
62410.14
66002.38
66039.86
44955.29
62659.35
44154.39
65876.19
65883.56
45556.94
66515.48
44112.65
45880.44
52039.23
65353.07
66260.12
65979.74
66369.47
O_Y
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
16
D_X
73
73
73
73
73
73
73
73
73
73
73
73
73
73
73
73
73
73
73
D_Y
16
16
16
16
16
17
17
17
17
17
18
18
18
18
18
18
18
19
19
75
80
80
80
80
75
75
76
76
76
73
73
74
74
74
74
74
69
69
…
…
…
Dep.Time O_X
29822.99
58036.25
58281.50
58884.98
25024.62
25695.85
25793.31
27298.21
30241.38
30335.80
30341.03
30578.84
31399.16
31420.73
31635.02
31751.79
32042.39
32124.87
32383.93
O_Y
81
81
81
81
81
81
81
81
81
81
81
81
81
81
81
81
81
81
81
D_X
139
139
139
139
139
139
139
139
139
139
139
139
139
139
139
139
139
139
139
D_Y
75
75
75
75
75
75
75
75
75
75
75
75
75
75
75
75
75
75
75
136
137
137
137
138
138
138
138
138
138
138
138
138
138
138
138
138
138
138
Figure 30: The First 19 and Last 18 Entries in the SPT Model Ordered Trip File
The trips listed on the left of Figure 30 are those that originate at the westernmost
point, (16, 73) which corresponds to a pixel that contains only one point of interest, Fort
Mott State Park, and lies due south of Wilmington, DE at the western edge of New
Jersey. The park employs approximately 10 people, is visited by 80 patrons each day,
according to the Employee and Patronage data updated by Professor Kornhauser, and
while only the first 19 trips originating at the park are shown in the figure, a grand total
of 65 trips in the ordered trip file originate from (16, 73), and by extension, from the
park. While this number is lower than the expected 90 visits, it is a very reasonably
realization for an average work day. The trips on the right of Figure 30 originate at the
easternmost point, (81, 139), which corresponds to Westchester County, NY, one of the
seven out-of-state locations in which New Jersey workers live and New Jersey residents
work. The final fifteen trips originating in Westchester County in the ordered trip file
share a common destination of (75, 138) and range in time from 6:57am until 9:00am,
71
indicating a potential for ridesharing during the morning commute to (75, 138), a pixel
that includes the towns of Rockleigh, NJ and Northvale, NJ, both in Bergen County. The
ridesharing potential across the entire state will be discussed in Section 5.3, and again in
Chapter Six.
5.3
Calculating Fleet Size and Travel Costs using the Transportation Problem
In order to compare the PRT model ATN with the SPT model ATN as they pertain to
transit criterion four – the economic feasibility – the required fleet size and travel costs
for each system need to be determined. The first step in that process is to combine any
individual person-trips that share the same origin pixel and destination pixel into one taxi
trip, provided they depart within a given time window
. This is done by stepping
through the ordered trip file person-trip-by-person-trip (row-by-row), performing
Equation (10) to determine
equation,
, the number of passengers present in taxi trip x. In the
corresponds to the row entry of the first person-trip in taxi trip x. Given a
maximum vehicle occupancy of
, taxi trips are filled by the first
travelling from point A to point B within
indicated in
, which is denoted
acceptable time slot,
passengers
seconds of the original departure time
. If fewer than
passengers arrive within the
is equal to the total number of person trips that originate within
that time slot for the given origin-destination pair.
∑
(10)
The result of running Equation (10) through the ordered trip file can be seen in Figure
31. The output rows have a very similar format to the rows in Figure 30, and come from,
again, the very beginning and very end of the ordered trip file. The difference is that the
72
O_X, O_Y, D_X, D_Y, and
values no longer apply to person-trips, but to taxi trips,
and for every taxi trip x, an occupancy
has been added in the final column. The
value for this output is set at six passengers.
Dep Time O_X
O_Y
D_X
D_Y
Q_x
Dep Time O_X
O_Y
D_X
D_Y
Q_x
58277
81
139
75
135
1
66099
16
73
16
75
1
28391
81
139
75
136
1
61999
16
73
16
80
1
…
29123
81
139
75
136
2
62410
16
73
16
80
1
29676
81
139
75
136
4
66002
16
73
16
80
2
58036
81
139
75
137
2
44955
16
73
17
75
1
58884
81
139
75
137
1
62659
16
73
17
75
1
25024
81
139
75
138
1
44154
16
73
17
76
1 …
25695
81
139
75
138
2
65876
16
73
17
76
2
27298
81
139
75
138
1
45556
16
73
18
73
1
30241
81
139
75
138
3
66515
16
73
18
73
1
…
30578
81
139
75
138
1
44112
16
73
18
74
1
31399
81
139
75
138
3
45880
16
73
18
74
1
31751
81
139
75
138
2
52039
16
73
18
74
1
32124
81
139
75
138
2
65353
16
73
18
74
1
Figure 31: Ordered Taxi Trip File with Capacities
The data in Figure 31 come from the SPT model for an ATN, and comparing it to the
data in Figure 30, it is clear that the trips originating in the Fort Mott State Park pixel, at
the left side of the figure, do not offer as much opportunity for ridesharing as those
originating in Westchester County, NY. The fifteen person-trips from (81, 139) in
Westchester County to (75, 138) in Bergen County have been condensed into eight taxi
trips, with vehicle occupancies ranging from one to three, whereas in the Fort Mott pixel
at (16, 73), only two of the trips leaving the park offer the possibility for ridesharing.
In its entirety, the SPT output file shown in the figure and its counterpart for the PRT
model list all the taxi trips needed to meet New Jersey’s transportation demand under the
characteristics of that model. A grand total of 32,770,528 individual person-trips are
reduced to a smaller number of taxi trips; the exact number depends on which model is
73
selected, as well as the values of
and
. The quantity of taxi trips required to
meet the given demand in a number of different situations is discussed at length in
Chapter Six. Whatever these parameters may be, the taxi trip output file shows the
demand for vehicles over time at each node throughout the day, which becomes the basic
input information for setting up a classic, cost-minimizing, transportation problem.
Figure 32: Visual Representation of the Classic Transportation Problem95
The transportation problem is a network flow problem that traditionally represents the
shipment of goods along a transportation network, from supply nodes – or sources – to
demand nodes – destinations. In the classic model, there is a cost matrix
contains the unit costs associated with sending goods from supply node
demand node
that
to
. In my analysis the goods will be the individual passengers, the
source nodes will be their (O_X, O_Y) pairs and the destination nodes will be their
(D_X, D_Y) pairs. An additional implementation of the transportation problem in
determining the network flow of uninhabited autonomous vehicles between trips will be
95
Gim et al. 2012
74
discussed in Chapter Seven. In my model, the source node and destination node for each
trip is deterministic, as it has to correspond to the given origin-destination pair for the taxi
trip in question. The cost minimization problem will be a direct comparison of the two
models, to determine whether the PRT model ATN or the SPT model ATN serves New
Jersey’s transportation demand in a more cost efficient way.
The cost function I employ in Chapter Six is linearly dependent upon the distance the
vehicle travels between picking up passengers at supply node m and dropping them off at
demand node n. In addition, the analysis requires knowledge of both the departure time of
each taxi trip and the arrival time at its destination, at which point the vehicle can be
repurposed to serve another trip in the area. The arrival time, too, depends on the distance
of the trip. Because the taxi trip output file only includes data regarding the origin pixel,
destination pixel and departure time, it is necessary to calculate the distance. However it
is important to note that the pick-up and drop-off behavior of the two systems differs
quite significantly. While a taxi trip in the PRT model ATN ends as soon as the vehicle
reaches the centroid of its destination pixel, the SPT model autonomous taxi must drive
around within both the origin and destination pixel to pick up and drop off its passengers.
This behavior is modeled via the addition of the function
to the distance
calculation (11) in the case of the SPT model. I have named this term the Chauffeur
Function, and it varies with
, the number of passengers in taxi trip x.
( √
)
75
(11)
○
○
○
●
○
○
●
○
○
○
●
○
○
○
○
○
○
●
○
○
○
○
○
●
●
○
○
1 Person Trip:
2 Person Trip:
3 Person Trip:
●
○
●
●
○
●
●
○
●
○
○
○
○
●
○
○
●
●
●
○
●
●
○
●
●
○
●
4 Person Trip:
5 Person Trip:
miles
6 Person Trip:
Figure 33: Realizations of the Chauffeur Factor for Trip Occupancies of 1-6
The Chauffeur Function, shown for
from one to six above, is a worst-case-scenario
pick-up and drop-off approximation for the SPT model autonomous taxi network. If taxi
trip x has only one passenger, the taxi can pick up the passenger at his point of origin,
drive him directly to his destination and then be free to relocate and serve another trip.
Given the fact that this pick-up and drop-off can occur anywhere within the pixel, the
value for
is equal to 0 miles, and the distance between the centroids of the two
pixels is multiplied by the circuity multiplier
to determine the distance. However,
when more than one passenger is present in an SPT model taxi trip, the vehicle cannot
76
necessarily make only one pick-up and drop-off stop. In the worst case scenario for a
three passenger trip, for example, the vehicle would have to travel
2.24 miles
between picking up passenger one and passenger three, and another
2.24 miles
while dropping them off. The remaining worst case scenarios are plotted in Figure 33,
using nine PRT model pixels to approximate one pixel in the SPT model grid. Of course,
there is a possibility that in a three passenger trip, two or more of the passengers actually
originate at the same place, but the Chauffeur Function is meant as a representation of the
worst-case scenario, with the knowledge that, in practice, the distances travelled in the
SPT model may be slightly shorter than the value calculated in Equation (11). In the case
of the PRT model,
is equal to zero regardless of the value of
.
Once the distance traveled per trip has been calculated using Equation (11), the trip
time is calculated by multiplying the distance by the inverse of the average vehicle speed.
In much of the analysis in Chapter Six, the average speed is assumed to be 30 miles per
hour, though in Figure 43 in Section 6.3.3 the inter-pixel driving is assigned a speed of 30
miles per hour while the “chauffeuring speed” – the distance added to the SPT model via
the Chauffeur Function – is given a slower average value of 15 miles per hour to account
for stops at passengers’ destinations. In both cases however, the trips in the SPT model
ATN are expected to have longer travel times than their counterparts in the PRT model
ATN, due to the extra driving inherent in the SPT model’s pick-up and drop-off scheme.
Given the calculation of taxi trip distances and arrival times, the total cost of the PRT
system and SPT system can be compared. In each case the total cost is a function of permile travel cost and fleet size, shown in Equation (12). The total operational cost is
77
approximated by the sum of the trip distances multiplied by a per-mile cost constant c
and the total vehicle cost is the product of the fleet size and a per-vehicle cost constant
∑
.
(12)
The required fleet size to meet all demand is denoted by
model is employed, as well as the vehicle occupancy
, and depends on which
and the time delay
. Fleet
size is calculated by discretizing time into 48 thirty-minute segments, and determining
how many vehicles are actively en route during each time segment. Whichever half-hour
time period has the greatest number of active vehicles will be the one that determines the
fleet size. In Chapter Six I explore two different methods for determining fleet size. In the
first I assume that any vehicle that has finished a trip can be instantaneously repurposed
to serve another trip anywhere in the state. In the second method, I require that each
vehicle wait one hour between trips, giving it ample time to refuel or recharge and
relocate to a nearby pixel where demand for a taxi trip has been indicated. The reality of
the repositioning and refueling time is likely somewhere in between these two methods,
and a more sophisticated means of calculating repositioning time is discussed in Section
7.2. The total cost of a PRT model ATN and an SPT model ATN under a number of
different scenarios is discussed in Section 6.4.
78
Chapter Six: The Results
Comparing the PRT and SPT Models for an ATN
79
Both the PRT and SPT model describe a system that significantly outperforms the
personally owned and operated car in transit criteria one through three. In this chapter I
present the results of the equations and models discussed in Chapter Five, and using these
values for average taxi trip occupancy, fleet size, average distance per taxi trip, and total
cost, I compare the two models to one another as they pertain to transit criteria four and
five, and additionally compare each model to its true competitor, the current system of
personally owned and operated cars.
6.1
Average Occupancy for Both Models Given
The first implementation of Equation (10) that I performed assumed a
equal to
five minutes, and did not impose a maximum vehicle occupancy. The resulting average
statewide taxi trip occupancy values are shown in the table below. At first glance it is
clear that the PRT model sees much less ridesharing than the SPT model. This result is to
be expected, as each pixel in the SPT model has nine times the amount of land area of a
single pixel in the PRT model. The trade-off between the two models is that while
ridesharing is more prevalent in the SPT model, trips take a longer amount of time due to
increased distance traveled within the origin and destination node, picking up and
dropping off passengers. This effect is discussed and implemented in Section 6.3.
Model
PRT
SPT
Statewide Occupancy for
Total Person-Trips
32,770,528
32,770,528
Total Taxi Trips
25,824,326
15,174,736
Average Occupancy
1.269
2.160
Running a simulation without imposing a maximum vehicle occupancy allows for
sanity checks regarding origin-destination pairs that experience very high volumes of
travel within specific five-minute intervals throughout the day. In the SPT model, one
80
such route takes place beginning at 28,177 seconds after midnight, or 7:50am, and lasts
until 7:55am. The trip serves 1,896 passengers within this five minute span, and
originates at pixel (71, 126), which corresponds to coordinates (40.73517, -74.04422) and
is indicated by the red marker in Figure 34. The destination that these nearly two
thousand passengers share is pixel (73, 126) which corresponds to coordinates (40.73517,
-73.98913) and is indicated by the green marker in Figure 34. The purpose of this sanity
check is to decide whether it is reasonable to expect 1,896 people to travel from the
Jersey City-Hoboken area into Manhattan between 7:50am and 7:55am on an average
business day. Indeed this origin-destination pair would be expected to serve a large
number of people just before 8:00am on a business day, precisely when the commuters
who live across the river from Manhattan (many to avoid its famously high rent) travel to
work in the city.
Figure 34: Origin and Destination of SPT Model’s Busiest Trip when
96
Image retrieved from Google Maps, April 2013
81
96
6.2
Average Occupancy for Both Models Varying
and
Having calculated the absolute best case average occupancy numbers for both
models, it is evident that in practice neither model will achieve the occupancies shown in
Section 6.1. To achieve these occupancies, the autonomous taxi network would need to
include at least one vehicle with an occupancy of 1,896 to serve the trip from Hoboken to
Manhattan at 7:50am. In reality, a limit needs to be set and the vehicle size must be
selected. Theoretically, an ATN could be served using vehicles as large as buses, with
occupancies of 48 or more, and on the other end of the spectrum the taxis could be tiny
pods with only two seats. Average statewide occupancies for a PRT model ATN with
varying from two to forty-eight seats, and with
equal to five or seven minutes
are shown in the table below and plotted in Figure 35.
48
24
12
8
6
5
4
3
2
48
24
12
8
6
5
4
3
2
Statewide Occupancy for PRT Model Varying
Total Taxi Trips
5 min
25,828,582
5 min
25,842,589
5 min
25,898,095
5 min
25,987,269
5 min
26,105,757
5 min
26,219,983
5 min
26,421,530
5 min
26,833,103
5 min
27,917,911
7 min
24,934,732
7 min
24,953,503
7 min
25,025,032
7 min
25,134,696
7 min
25,277,242
7 min
25,412,173
7 min
25,648,059
7 min
26,124,301
7 min
27,362,145
82
and
Average Occupancy
1.269
1.268
1.265
1.261
1.255
1.250
1.240
1.221
1.174
1.314
1.313
1.310
1.304
1.296
1.290
1.278
1.254
1.198
Statewide Occupancy for PRT Model
Average Statewide Occupancy
1.34
1.32
1.30
1.28
1.26
1.24
Tmax = 5 min
1.22
Tmax = 7 min
1.20
1.18
1.16
0
4
8 12 16 20 24 28 32 36 40 44 48
Vehicle Size (number of seats)
Figure 35: Statewide Occupancy for PRT Model Given Variable
Focusing on the solid line in which
is equal to five minutes as in section 6.1, the
PRT model average occupancy is actually equal to the best case,
occupancy of
1.269 people per taxi trip when the vehicle occupancy is set to forty-eight passengers. For
between six and forty-eight, the average statewide taxi occupancy ranges from
1.255 to 1.269, but with a maximum vehicle occupancy less than six there is a rapid dropoff in average trip occupancy to a value of 1.221 for a
when
of three, and a mere 1.174
is equal to two.
The same calculations were performed for the SPT model and, as expected, the
occupancy numbers were significantly higher than in the PRT model. A table of
statewide average occupancy in an SPT model ATN with
forty-eight and
varying from two to
values of five minutes and seven minutes is shown on the top of the
next page. It is followed by Figure 36 which is a plot of the SPT model taxi trip
occupancies for the range of
values. The contour of the plot is very similar to the
83
PRT model in Figure 35, and a major fall-off occurs between maximum occupancy of
eight and six.
and
Average Occupancy
2.155
2.145
2.111
2.066
2.012
1.965
1.890
1.761
1.514
2.374
2.358
2.310
2.246
2.174
2.111
2.015
1.854
1.562
Statewide Occupancy for SPT Model
Average Statewide Occupancy
48
24
12
8
6
5
4
3
2
48
24
12
8
6
5
4
3
2
Statewide Occupancy for SPT Model Varying
Total Taxi Trips
5 min
15,204,783
5 min
15,279,703
5 min
15,522,727
5 min
15,864,220
5 min
16,287,928
5 min
16,677,475
5 min
17,339,134
5 min
18,613,650
5 min
21,648,455
7 min
13,803,641
7 min
13,896,478
7 min
14,188,442
7 min
14,588,688
7 min
15,076,195
7 min
15,520,223
7 min
16,264,001
7 min
17,676,762
7 min
20,979,557
2.40
2.30
2.20
2.10
2.00
1.90
1.80
1.70
1.60
1.50
1.40
TD = 5 min
TD = 7 min
0
4
8 12 16 20 24 28 32 36 40 44 48
Vehicle Size (number of seats)
Figure 36: Statewide Occupancy for SPT Model Given Variable
84
Despite the clear benefits of using larger vehicles in both the PRT and SPT model
ATNs, there are also significant costs associated with vehicles built for twelve people, for
example, versus vehicles built for six people. For the remainder of the chapter I will
focus on two specific layouts of an ATN, one a PRT model and the other an SPT model,
each with a
equal to five minutes, which is more convenient to passengers than a
of seven minutes, and a
equal to six passengers. The value of six has been
selected because it is the beginning of the occupancy drop-off and because six-passenger
autonomous vehicles could easily be mass-produced on current sedan platforms, and
would be approximately the size of today’s vehicles.
6.3
Further Analysis for SPT and PRT Model ATNs with
The histogram for PRT model ATN occupancies when
is shown in Figure
37. Of the 26 million taxi trips taken in this scenario, 22.8 million are single occupancy.
Two-person trips are the next most common, with 1.9 million occurrences, and the
remaining four possibilities make up only 1.3 million trips, just five percent of the total.
Figure 37: Histogram of Taxi Trip Occupancies for
85
in PRT Model
In Figure 38 the same histogram has been plotted for the SPT model. As expected,
multiple-passenger trips are much more common in the SPT model than in the PRT
model. Of the 16 million taxi trips, 2.5 million are two-person trips and 1.8 million are
six-person trips. This explosion of six-person trips as compared to the PRT model comes
from origin-destination pairs like the one discussed in Section 6.1, where 1,896
passengers would be split up into exactly 316 taxi trips, each of which would carry six
passengers into Manhattan. Of course, this is not really the most efficient way to move
almost 2,000 people across the Hudson River, but for the purpose of this analysis I have
assumed that the ATN in question meets any and all travel demand in New Jersey, with
the exception of intra-pixel trips in the PRT model, which must be met by individuals
walking or biking, as the autonomous taxis in that model run only from pixel centroid to
pixel centroid.
Figure 38: Histogram of Taxi Trip Occupancies for
86
in SPT Model
Continuing the analysis of the two different ATN models for a
a
equal to six and
equal to five minutes, it is important to discuss the trade-off between ridesharing
and total trip distance mentioned in Section 6.1. As shown in the histograms and the
vehicle occupancy plots in Section 6.2, the SPT model offers much more opportunity for
ridesharing than the PRT model. This more efficient use of vehicle space does not come
for free however. Because vehicles in the SPT model drive around their origin and
destination pixels, the trip distances and trip times are longer than in the PRT model. In
the table below, the average trip distance for each model has been calculated using
Equation (11), and sure enough, the average trip distance in the SPT model is 4.88 miles
longer than the average in the PRT model. Despite this disparity however, the
approximate total distance traveled by all vehicles over the entire day in the SPT model is
still more than 80 million miles less than the total distance traveled in the PRT model.
Average Distance per Trip and Approximate Total Distance for Both Models
Model
Total Taxi Trips Average Trip Distance Approximate Total Distance
PRT
26,105,757
16.29 miles
425,263,000 miles
SPT
16,287,928
21.17 miles
344,815,000 miles
Because the total mileage of the PRT model is greater than that of the SPT model,
and the number of trips needed to meet demand is drastically larger in the PRT model as
well, the expectation is that the PRT model’s cost will be much greater than the cost of an
SPT model ATN. In the following three sections I will calculate the required fleet size for
both models in three ways: 1) assuming instantaneous repositioning capability, 2)
assuming one hour of down-time between taxi trips, and 3) implementing variable speed
in the SPT model when chauffeuring passengers around their origin and destination
87
pixels. Given these fleet sizes, it will become clear which model is truly more
economically viable, and a better candidate to satisfy transit criterion number four.
6.3.1
Calculating Fleet Size with Instantaneous Repositioning
The assumption that vehicles can instantaneously reposition themselves to meet any
demand that arises within the network allows for a relatively easy calculation of the best
case scenario for fleet size. As mentioned in Section 5.3, time is discretized into fortyeight segments, each a half hour long. For each time segment, the total number of cars en
route is calculated assuming an average speed of 30 miles per hour, and plotted in a graph
such as the one in Figure 39, which corresponds to the PRT model. The main peak in the
morning rush hour occurs between 7:30am and 8:00am, when there are 1,775,225
vehicles on the road in the PRT model. Evening rush hour is much busier than morning
rush hour however, as the majority of secondary trips take place in the afternoon and
evening, resulting in a daily maximum fleet size of 2,409,736 vehicles on the road
between 5:00pm and 5:30pm.
PRT Model ATN Fleet Requirement
0:00-0:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
12:00-12:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
2,600,000
2,400,000
2,200,000
2,000,000
1,800,000
1,600,000
1,400,000
1,200,000
1,000,000
800,000
600,000
400,000
200,000
-
Figure 39: Vehicles Required at 48 Time Steps in PRT Model Assuming Instantaneous Repositioning
88
In 2011, there were 7,609,467 vehicles registered in the state of New Jersey, so a fleet
size of 2,410,000 would be a very considerable reduction in the number of vehicles
needed to meet New Jersey’s transportation demand. The reduction in fleet size would
further reduce congestion and the danger of accidents, and would furthermore be better
for the roads. As expected, the fleet size required for an SPT model ATN is even smaller
than that of a PRT model ATN. The busiest time of day in the SPT model is the time
between 6:00pm and 6:30pm, when there are 1,609,073 vehicles actively moving
passengers from their origins to their destinations.
SPT Model ATN Fleet Requirement
1,800,000
1,600,000
1,400,000
1,200,000
1,000,000
800,000
600,000
400,000
200,000
0:00-0:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
12:00-12:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
-
Figure 40: Vehicles Required at 48 Time Steps in SPT Model Assuming Instantaneous Repositioning
While the instantaneous repositioning method is the most straightforward way to
calculate approximate fleet size, its results are almost certainly better than the reality of
the system in practice. In an effort to determine a more realistic, if slightly over-cautious
89
fleet size for both models, I have implemented a system in Section 6.3.2 in which every
taxi is required to take an hour break between trips to refuel and reposition itself.
6.3.2
Calculating Fleet Size with One Hour Break between Trips
As expected, in the implementation that includes an extra hour of wait time between
trips, the calculated fleet size is much higher than in the case of instantaneous
repositioning. The contours of the PRT model fleet requirement are very similar in Figure
41 to what they looked like in the instantaneous repositioning graph in Figure 39, and the
busiest time period has actually changed from 5:00pm-5:30pm to 6:00pm-6:30pm. This
makes sense because as the afternoon rush begins, every trip that is undertaken carries
with it an hour of waiting. It is no coincidence that the new peak time is one hour after
the peak time in Figure 39. As for the actual capacity calculation for the PRT model in
Figure 41, there are 4,450,701 cars on the road between 6:00pm and 6:30pm.
PRT Model With 1 Hour of Repositioning Time
5,000,000
4,500,000
Required Fleet Size
4,000,000
3,500,000
3,000,000
2,500,000
2,000,000
1,500,000
1,000,000
500,000
0:00-0:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
12:00-12:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
-
Time
Figure 41: Vehicles Required at 48 Time Steps in PRT Model Assuming 1 Hour Repositioning Time
90
The results of the SPT model with one hour of repositioning and refueling time after
every trip are very similar to those shown in Figure 41. Like the PRT model, the SPT
model experiences the busiest streets from 6:00pm until 6:30pm, with 2,789,391 vehicles
on the road during that time.
SPT Model with 1 Hour of Repositioning Time
3,000,000
Required Fleet Size
2,500,000
2,000,000
1,500,000
1,000,000
500,000
0:00-0:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
12:00-12:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
-
Time
Figure 42: Vehicles Required at 48 Time Steps in SPT Model Assuming 1 Hour Repositioning Time
Assuming that reality lies somewhere between instantaneous repositioning and hourlong required waiting time, the necessary fleet size for a PRT model ATN would be
somewhere between 2,409,736 vehicles and 4,450,701 vehicles. For an SPT model ATN,
the range would be between 1,609,073 and 2,789,391 vehicles. While the SPT model
appears to be a better set-up than the PRT model, it is worth noting that both models
outperform the automobile in terms of limiting the vehicles on the road without
sacrificing mobility. In the end, that is the goal of any alternative to the car: to afford the
people of the future all the freedom and mobility we experience today, without any of the
91
negative externalities including hassle, safety issues, environmental concerns, time
wasted behind the wheel, etc.
6.3.3
Varying Chauffeur Speed for the SPT Model
One final area of increased analysis regarding fleet size is the specification of intrapixel travel speed in the SPT model. In Sections 6.3.1 and 6.3.2 I have assumed that
every trip has an average speed of 30 miles per hour, but that assumption actually
benefits the SPT model and hurts the PRT model, because the pick-up and drop-off
speeds in the SPT model would almost certainly be slower than the “open road” speeds.
Figure 43 plots the fleet size requirement over time when the chauffeur speed – or intrapixel speed – is equal to 30 miles per hour as it was in the previous two sections, and
when it is half that speed, at 15 miles per hour. While there is definitely an effect on the
necessary fleet size, which increases 5.4 percent from 1,609,073 to 1,695,460, it is not
enough of an effect to change the SPT model’s preferred status over the PRT model.
SPT Fleet Size with Chauffeur Speed 15mph vs. 30mph
1800000
1600000
1400000
1200000
1000000
800000
15 mph
600000
30 mph
400000
200000
0:00-0:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
12:00-12:30
1:00-1:30
2:00-2:30
3:00-3:30
4:00-4:30
5:00-5:30
6:00-6:30
7:00-7:30
8:00-8:30
9:00-9:30
10:00-10:30
11:00-11:30
0
Figure 43: Decreasing the Chauffeur Speed by 50% Increases the Fleet Size by 5%
92
6.4
Cost Comparison of the Two Models
Given the minimal effect of changing the chauffeur speed in the SPT model, I will
calculate the total cost figures for both models assuming an average trip speed of 30
miles per hour, and compare the PRT and SPT models using both the upper and lower
bound for fleet size, rather than attempting to pick a reasonable value in between the two
bounds. Reiterating Section 5.3, the equation for total daily operation cost for each model
is found in Equation (12) below.
∑
(12)
The total cost estimates based on the results presented in this chapter are contained in
the table at the bottom of this page. For the value of c I have used $0.17 per mile, which
is the average cost of operating a personal vehicle according to research by Gary Barnes
and Peter Langworthy at the University of Minnesota. For the value of
, the estimated
cost of each vehicle, there is a lot of leeway as estimates for the cost of fully autonomous
cars range anywhere from $100,000 to $300,000. Assuming that by the time an ATN is
implemented, the equipment will be more reasonably priced, I have selected a cost per
vehicle of $100,000 and divided it equally over five years, coming to $54.76 per car per
day. The resulting daily costs for the PRT model range from $204 million to $316
million, or a per capita price of $22.69 to $35.11 for the entire system. The SPT model
ranges from $147 million to $211 million total, or $16.30 to $23.49 per capita.
Model
PRT Inst.
PRT 1 Hr.
SPT Inst.
SPT 1 Hr.
Estimating the Total Cost of the Two Models
c
Sum(Dist)
$0.17 425,263,000 miles
$54.76
2,409,736
$0.17 425,263,000 miles
$54.76
4,450,701
$0.17 344,815,000 miles
$54.76
1,609,073
$0.17 344,815,000 miles
$54.76
2,789,391
93
Cost
$ 204.2 M
$ 316.0 M
$ 146.7 M
$ 211.4 M
While there is a slight overlap between the two cost windows, it is evident that the
SPT model outperforms the PRT model in terms of transit criterion four. The comparison
of the two models as they pertain to transit criterion five however is much more
qualitative.
6.5
Comparing Comfort and Convenience in the Two Models
An autonomous taxi network, regardless of the model on which it is built, has one
inherent comfort and convenience benefit over the current system of personally owned
and operated cars. That benefit is that travelers can have productive commutes. In today’s
system, driving one’s own car 45 minutes to work requires that one put aside email,
work, and phone calls for those 45 minutes. In an ATN, every traveler can have a
productive commute, during which they are free to work on projects, prepare for
meetings, or do whatever they’d like to do. While both the PRT model and SPT model
share this benefit, I do believe that the SPT model has other preferable qualities to the
PRT model that make it better suited to satisfy transit criterion five.
Most of all, the SPT model caters to the individual much more than the PRT model
does. In the PRT model, every passenger must walk to the taxi stand at the centroid of his
transit pixel to meet up with his vehicle. The SPT model offers what seems like a much
more personalized service, picking up travelers very close to their trip origin coordinates.
While this extra distance traveled could be seen as an inconvenience compared to the
PRT model, it does allow for such a significant increase ridesharing that the total distance
traveled by all vehicles is lower in the SPT model despite a longer average trip distance.
Presumably this lower cost would be passed on to the consumers as lower rates to use an
SPT model ATN as compared to a PRT model ATN, again making it more convenient.
94
Chapter Seven: Conclusion
Transition, Implementation and Further Research
95
Chart-topping personal mobility has been a major contributor to American success
and wealth throughout the twentieth century. If we as a nation seek to maintain the
personal freedoms and quality of life enjoyed in the late twentieth century for the
remainder of the twenty-first century, we cannot and should not pursue transportation
systems that offer decreased personal mobility. The result of such policies can already be
seen in the thousands of miles of lightly-occupied light rail and expansions of bus
systems that are rarely filled to twenty-five percent of vehicle capacity in cities like
Sacramento. Any successful transportation system will need to obey the five transit
criteria presented in this thesis if it hopes to enjoy true commercial success in America.
While a network of autonomous taxis may or may not be the transportation system of the
future, it is possible that – using Mark Gorton’s principle of a Smart Para-Transit system
– an ATN could serve one hundred percent of travel demand in New Jersey, while
costing its passengers between $16 and $25 per day. However, for all their many benefits,
there is a possibility that truly autonomous vehicles could be kept out of commercial
markets, and thereby kept off the road (with the exception of current test vehicles) for a
decade or longer.
7.1
Potential Barriers to ATN Implementation
The successful operation of Google’s twelve autonomous vehicles for over 500,000
accident-free miles, as well as the successful completion of the DARPA Urban Challenge
in 2007 and the integration of autonomous technology in current vehicles is indicative of
the fact that the technology required to create an ATN is largely available. The main
barrier at the moment to autonomous vehicles becoming widespread on current roads is a
combination of consumer wariness, evidenced by Ford’s focus groups and questionnaires
96
that indicate drivers’ unwillingness to completely relinquish control, and legal
uncertainty in the event of an accident. If a driverless car causes an accident, the
argument goes, who will be liable for the damages? In conventional cars the liability
belongs to the driver, but in autonomous vehicles, the question is, would the
manufacturer be at fault?
The interesting fact behind this argument is that autonomous vehicles have the
potential to be remarkably safer than conventional cars, and much more reliable in their
decision-making than human drivers. If an insurance company actually incentivized its
clients to use autonomous vehicles, or cars with autonomous features, the implied
liability of these automobiles would actually decrease. If in fact the implied liability of an
autonomous vehicle is much less than the implied liability of a manually-operated
vehicle, which seems to follow from the discussion in Chapter Three, insurance
companies would stand to make a large profit insuring autonomous or semi-autonomous
cars. If an insurance company decides to pursue this money-making opportunity, the
incentives could start right away, encouraging drivers to buy cars with lane-assist or
adaptive cruise control, and could eventually be used to defray the high cost of fully
autonomous vehicles down the line.
An ATN also requires all vehicles on the road to communicate with one another, and
assumes that vehicles belong to the network rather than being personally owned. Such a
system could itself be seen by some Americans as an affront on personal liberties like the
ability to own and drive one’s own car. Of course the counter-argument to this suggestion
comes from the world of horses and canoes. In the early 1800s, horses we the primary
means of mobility in the United States, and the canoe was the fastest way for Lewis and
97
Clark to cross the continent. In the year 2013, however, canoes and horses have been
replaced by faster, more efficient systems. Nonetheless, many Americans still own horses
or canoes, not for the mobility they offer, but rather for the pure enjoyment of riding the
horse or paddling the canoe as a form of recreation. “Baseball, hot dogs, apple pie, and
Chevrolet” may be a part of the American identity for hundreds of years to come, but
there is no reason that individuals’ love of driving should prevent the implementation of a
more efficient transportation system in the near future. In a world dominated by ATNs,
individuals could still own and operate personal vehicles for the pure enjoyment of doing
so; they would just need to limit their recreational driving to race tracks, off-road courses,
and private land.
7.2
Further Analysis: Using the Transportation Problem to Model Repositioning
While I employed the classic transportation problem to perform a cost minimization
between the PRT and SPT models, there is a possibility to further sophisticate the
methods used in Chapters Five and Six to determine more specific fleet sizes than the
ranges I present on page 93. This analysis would use the transportation model to
determine the network flows of the empty taxis in between trips. The supply and demand
nodes would be reversed from the problem discussed in Chapter Five. Because the taxis
empty out and become “available” again at the destination pixel of their passengers, the
taxi trip destinations would be the supply nodes. The demand nodes would be pixels in
which passengers indicated demand – which is to say their trip origin locations.
For a number of discrete time steps like the ones I employed in Chapters Five and
Six, the transportation problem would assign empty taxis from supply nodes to node arcs
that connected them to demand nodes where a taxi was needed, all while minimizing the
98
cost of travel, which is equivalent to minimizing the distance to the demand node. If there
are not enough taxis at all the supply nodes combined to meet the demand in a given time
step, an additional taxi would be added to the fleet size to meet this trip. Therefore the
system would calculate fleet size in a very similar manner to my analysis in Chapter Six,
but would include the cost of repositioning as a function of distance, a parameter that I
have not included in my results.
In addition to a more precise fleet size, the benefits of implementing this
transportation problem minimization would be the ability to visually simulate all the
vehicle travel in a New Jersey ATN, including the repositioning of empty vehicles. The
model discussed in this section would provide information as to where the ATN vehicles
were located when not serving passengers; because the trip files generated using Mufti’s
methods provide information as to where the vehicles are when they are serving
passengers, the combination of the two outputs would be all the data necessary to put
together a comprehensive visual model of how an ATN would truly operate. If the
popularity of Norman Bel Geddes’s Futurama exhibit at the 1939 World’s Fair is any
indication, a visual rendering of an ATN would likely get people excited about the
possibility of implementing such a system.
7.3
Where to Go from Here?
The importance of public support for an ATN cannot be downplayed if the system
hopes to have any commercial success. Theoretically, an autonomous taxi network would
outperform the car in congestion mitigation, safety, environmental friendliness, total
economic cost, and comfort and convenience. However, the same things have been said
about many failed transportation systems including Personal Rapid Transit. The only way
99
to know for sure whether an ATN modeled after Mark Gorton’s Smart Para-Transit
system will truly be able to move passengers in as efficient and effective a manner as is
presented in this thesis is for such a network to be implemented on at least a small scale.
The first step toward the implementation of a potentially revolutionary network of
autonomous taxis is establishing a method for today’s automobiles to communicate with
one another, like the ATN vehicles would do. The second step is a movement toward the
cultural acceptance of driverless cars. Only when roads in the United States are filled
with constantly communicating autonomous and semi-autonomous vehicles will a
statewide ATN such as the one I have presented become truly feasible. Luckily, the
technology required for both those steps, as well as for the implementation of an ATN,
already exists. It is the task of automakers, governments, private-sector profit-seekers,
and the American people as a whole to embrace and encourage this new technology, and
eventually to reap its benefits for centuries to come.
100
References
Agatz, Niels et al. Sustainable Passenger Transportation: Dynamic Ride-Sharing. ERIM
Report Series Research in Management. February 2010.
Althoff, Matthias, Olaf Stursberg, and Martin Buss. Safety Assessment of Autonomous
Cars using Verification Techniques. American Control Conference. 2007.
Retrieved from: http://www.lsr.ei.tum.de/fileadmin/backup/ACC_final.pdf
Balaker, Ted. Why Mobility Matters. Reason Foundation. August 2006. Retrieved from:
http://reason.org/files/52551973a35d9e8d342b78197e6d31d9.pdf
Barnes, Gary and Peter Langworthy. The Per-Mile Costs of Operating Automobiles and
Trucks. Minneapolis, MN: Humphrey Institute of Public Affairs. 2003. Retrieved
from: http://www.hhh.umn.edu/centers/slp/pdf/reports_papers/per_mile_costs.pdf
Beebe, Lucius. 20th Century: The Greatest Train in the World. Berkeley CA: HowellNorth Books. 1962.
Bendixson, Terence. Instead of Cars. London: Temple Smith. 1974.
Blodget, Henry. Here Are Some Of The Problems Google Is Having With Its Self-Driving
Cars. Business Insider. March 2013. Retrieved from:
http://www.businessinsider.com/google-self-driving-car-problems-2013-3
Burke, Catherine G. Innovation and Public Policy: The Case of Personal Rapid Transit.
USA: D.C. Heath and Company. 1979.
Butts, Mike. Automotive Aerodynamics: Drag Area – Size Matters. Curbside Classic.
February 2012. Retrieved from: http://www.curbsideclassic.com/ automotivehistories/automotive-aerodynamics-drag-area-size-matters/
Cook, Sam. Google’s Self-Driving Cars, and How Society Copes with New Technology.
Geek. May 2012. Retrieved from: http://www.geek.com/news/ googles-selfdriving-cars-and-how-society-copes-with-new-technology-1492453/
DARPA. A Huge Leap Forward for Robotics R&D. 2005. Retrieved from:
http://archive.darpa.mil/grandchallenge05/gcorg/index.html
Dennis, Kingsley and John Urry. After the Car. Cambridge: Polity Press. 2009.
FindTheData. New York, New York Transportation Statistics. 2013. Retrieved from:
http://transportation-modes-city.findthedata.org/l/1447/New-York-New-York
101
Fitchard, Kevin. Ford is Ready for the Autonomous Car. Are Drivers? Bloomberg
Businessweek. August 2012. Retrieved from: http://www.businessweek.com/
articles/2012-04-09/ford-is-ready-for-the-autonomous-car-dot-are-drivers#p1
Fitchard, Kevin. If Cars Could Talk to One Another, What Could (and Should) They Say?
Gigaom. February 2012. Retrieved from: http://gigaom.com/2012/02/ 28/if-carscould-talk-to-another-what-could-and-should-they-say/
Flammang, James M. Era of Driverless Cars May Be Near, Ideally Reducing Crashes
and Fatalities. AutoMedia.com. July 2012. Retrieved from:
http://www.automedia.com/Blog/post/Era-of-Driverless-Cars-May-Be-NearIdeally-Reducing-Crashes-and-Fatalities.aspx
Folsom, Tyler C. Social Ramifications of Autonomous Urban Land Vehicles. May 2011.
Retrieved from: http://www.enviroteach.com/social_ramifications.pdf
Ford Motor Company. Ford Lane Keeping System Helps Fusion Drivers Stay Alert and
Between the Lines. December 2011. Retrieved from:
http://media.ford.com/article_display.cfm?article_id=35776
Geddes, Norman Bel. Magic Motorways. 1940. Retrieved from:
http://ia600208.us.archive.org/10/items/magicmotorways00geddrich/magicmotor
ways00geddrich.pdf
Giammar, Daniel. Lecture: Motor Vehicles and Air Quality. Presented to CEE 304 at
Princeton University. February 2013.
Gim, Bonjin, Kyung Jin Boo, and Sang Min Cho. A Transportation Model Approach for
Constructing the Cost Effective Central Hydrogen Supply System in Korea.
Dankook University. International Journal of Hydrogen Energy. January 2012.
Retrieved from: http://www.sciencedirect.com/science/article/pii/
S0360319911022671
Google Maps. (40.73517,-74.04422) to (40.73517,-73.98913). April 2013. Retrieved
from: https://maps.google.com/maps?hl=en&gs_rn=8&gs_ri=psyab&gs_mss=(40.73517,-74.04422)+to++(40.73517,-73.98913)&tok=KTpgdyM9eO4I82jDME3Zg&pq=58226/3600&cp=23&gs_id=l&xhr=t&q=(40.73
517,-74.04422)+to+(40.73517,-73.98913)&bav=on.2,or.r_cp.r_qf.&biw=
1920&bih=971&bs=1&um=1&ie=UTF-8&sa=N&tab=wl&authuser=0
Gorton, Mark. Using Information Technology to Achieve a Breakthrough in
Transportation in New York City. The Open Planning Project. August 2008.
Retrieved from: http://www.streetsblog.org/wp-content/pdf/SmartParaTransit.pdf
The Guardian Data Blog. Emissions by Transport Type. 2009. Retrieved from:
http://www.guardian.co.uk/environment/datablog/2009/sep/02/carbon-emissionsper-transport-type
102
Guizzo, Erico. How Google’s Self-Driving Car Works. Discovery News. October 2011.
Retrieved from: http://news.discovery.com/autos/future-of-transportation/ howgoogle-self-driving-car-works-111018.htm
Hartgen, David T. and Gregory Fields Building Roads to Reduce Traffic Congestion in
America’s Cities: How Much and at What Cost? Reason Foundation. August
2006. Retrieved from: http://reason.org/files/ps346.pdf
Hooper, Joseph. From DARPA Grand Challenge 2004: DARPA’s Debacle in the Desert.
Popular Science. June 2004. Retrieved from:
http://www.popsci.com/scitech/article/2004-06/darpa-grand-challenge2004darpas-debacle-desert
Kloiber, Bernhard, Thomas Strang, and Fabian de Ponte Muller. Slipstream Cooperative
Adaptive Cruise Control – A Conceptual ITS Application for Electric Vehicles.
German Aerospace Center. 2012. Retrieved from:
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6183170&tag=1
Kornhauser, Alain. Lecture: Enhancing Mobility Through Technology in a Congested
Urban Environment. Presented to ORF 467, Fall 2011.
Kornhauser, Alain et al. Synthesizing Individual Travel Demand in New Jersey. Princeton
University, January 2012.
Kornhauser, Alain et al. Uncongested Mobility for All, While Improving Safety, Energy,
and Environmental Consequences: New Jersey’s Area-wide aTaxi System.
Princeton University. January 2013.
Mufti, Talal R. Synthesis of Spatially & Temporally Disaggregate Person Trip Demand:
Application for a Typical New Jersey Weekday. Princeton University, September
2012.
Mui, Chunka. Fasten Your Seatbelts: Google’s Driverless Car is Worth Trillions. Forbes
Magazine. January 2013. Retrieved from: http://www.forbes.com/sites/
chunkamui/2013/01/22/fasten-your-seatbelts-googles-driverless-car-is-worthtrillions/
O’Kane, Tony. SARTRE Project Aims to Develop Autonomous Car Technology. The
Motor Report. October 2009. Retrieved from: http://www.themotorreport.com.au/
45030/sartre-project-aims-to-develop-autonomous-car-technology
O’Toole, Randal. Gridlock: Why We’re Stuck in Traffic and What to Do About It.
Washington D.C.: The Cato Institute. 2009.
103
PR Newswire. Buick LeSabres to Demonstrate Automated Highway Technologies. June
1996. Retrieved from: http://www.thefreelibrary.com/BUICK+LESABRES+
TO+DEMONSTRATE+AUTOMATED+HIGHWAY+TECHNOLOGIESa018428882
Princeton University. Critical History of Transit Planning and Decisionmaking.
Retrieved from: http://www.princeton.edu/~ota/disk3/1976/7606/760605.PDF
Sabatini, Jeff. 2013 Ford C-Max Hybrid. Car and Driver. November 2012. Retrieved
from: http://www.caranddriver.com/reviews/2013-ford-c-max-hybrid-test-review
SACOG (Sacramento Area Council of Governments). 2006 Metropolitan Transportation
Plan. March 2006. Retrieved from: http://www.sacog.org/mtp/ pdf/
MTP2006/2006%20MTP%203-16-06.pdf
Simonite, Tom. Audi Shrinks the Autonomous Car. MIT Technology Review. January
2013. Retrieved from: http://www.technologyreview.com/news/509676/ audishrinks-the-autonomous-car/
Simonite, Tom. Toyota Unveils an Autonomous Car, but Says It’ll Keep Drivers in
Control. MIT Technology Review. January 2013. Retrieved from:
http://www.technologyreview.com/news/509616/toyota-unveils-an-autonomouscar-but-says-itll-keep-drivers-in-control/
Slosson, Mary. Google Gets First Self-Driven Car License in Nevada. May 2012.
Reuters. Retrieved from: http://www.reuters.com/article/2012/05/08/uk-usanevada-google-idUSLNE84701320120508
Smithsonian Air and Space Museum: America by Air. 2007. Retrieved from:
http://airandspace.si.edu/americabyair/flyacross/index.cfm
Staley, Samuel and Adrian Moore. Mobility First: A New Vision for Transportation in a
Globally Competitive Twenty-First Century. Plymouth U.K.: Rowman &
Littlefield. 2009.
Thrun, Sebastian. TED Talk: Google’s Driverless Car. March 2011. Retrieved from:
http://www.ted.com/talks/sebastian_thrun_google_s_driverless_car.html
Tientrakool, Patcharinee. Reliable Neighborcast Protocol for Vehicular Ad Hoc
Networks. Columbia University. 2011. Retrieved from:
http://academiccommons.columbia.edu/catalog/ac:134198
Unruh, John. The Plains Across: The Overland Emigrants and the Trans-Mississippi
West, 1840-60. 1979. Retrieved from: http://books.google.com/books?id=
hM2IXMY3jQC&printsec=frontcover&source=gbs_ge_summary_r&cad=0#v=o
nepage&q&f=false
104
U.S. Center for Disease Control. Detailed Tables for the National Vital Statistics Report
(NVSR). 2010. Retrieved from: http://www.cdc.gov/nchs/data/dvs/
deaths_2010_release.pdf
U.S. Department of Commerce. Historical Statistics of the United States: Colonial Times
to 1970. 1975. Retrieved from: http://books.google.com/
books?id=60AB17kg6ZIC&printsec=frontcover&dq=historical+statistics+of+the
+united+states&hl=en&sa=X&ei=wU0PUd-wHq00AGXtIC4Ag&ved=
0CC0Q6AEwAA
U.S. Department of Transportation Federal Highway Administration. Highway Statistics
Summary to 1995: Table MV-200. April 2011. Retrieved from:
http://www.fhwa.dot.gov/ohim/summary95/section2.html
U.S. Department of Transportation Federal Highway Administration. Historical Monthly
VMT Report. March 2012. Retrieved from: http://www.fhwa.dot.gov/
policyinformation/travel/tvt/history/
U.S. Environmental Protection Agency. National Emissions Inventory Air Pollutant
Emissions Trends Data. 2012. Retrieved from: http://www.epa.gov/
ttnchie1/trends/
Witkin, Richard. Wide Technological Drive Planned by Administration. New York
Times. January 21, 1972. Retrieved from: http://jfk.hood.edu/Collection/
White%20Materials/Nixon%20Administration/Nixon%201440.pdf
World Health Organization. World Report on Road Traffic Injury Prevention. 2004.
Retrieved from: http://www.who.int/violence_injury_prevention/
publications/road_traffic/world_report/summary_en_rev.pdf
Yarow, Jay. Human Driver Crashes Google’s Self Driving Car. Business Insider. August
2011. Retrieved from: http://www.businessinsider.com/googles-self-driving-carsget-in-their-first-accident-2011-8
105
Appendix A: Detailed Flowcharts of Methods in Mufti 2012
Module One: Generating the New Jersey Residents
106
Module 2a and 2c: Generating Workers from Out of State and Assigning their Primary
Role Locations
Module 2b: Assigning Primary Role Locations to NJ-Resident Workers
Module 3: Assigning Primary Role Locations to NJ-Resident Students
107
Module 4: Assigning Tour Type 0-17
Module 5: Generating Secondary Trips
Module 6: Adding the Temporal Dimension
97
All Images in the Appendix are taken from Mufti 2012
108
Appendix B: Code
/*******************************************************/
/* File: tripfile.c
*/
/* Author: Chris Brownell
*/
/*
*/
/* Description: -Implementation of Equations 8 and 9
*/
/*
-Generates Output in Figure 30
*/
/*
*/
/*******************************************************/
#include
#include
#include
#include
<stdio.h>
<math.h>
<stdlib.h>
<string.h>
int main(int argc, char *argv[]) {
FILE * pInput;
FILE * pOutput;
int i_Person = 0;
int origin_x;
int origin_y;
int past_x;
int past_y;
float home_lat;
float home_lon;
float distance;
float origin_lat;
float origin_lon;
char c = 8;
int i;
int i_Type;
float f_DepTime;
float f_ArrTime;
float f_DepTimeNext;
int counter = 0;
float size = 0.5;
char thefilename[100];
strcpy(thefilename, argv[1]);
strcat(thefilename, "Module6NN2ndRun.csv");
pInput = fopen(thefilename, "r");
pOutput = fopen(argv[2], "a");
/* kill first line */
while (c != '\n') {
fscanf(pInput, "%c", &c);
}
start:
/* kill 26 cells */
for (i=0; i<=25; i++) {
c=0;
while (c != ',') {
c = fgetc(pInput);
if (c == EOF) return 0;
}
}
i_Person++;
/* read in Trip Type */
fscanf(pInput, "%d", &i_Type);
c = fgetc(pInput);
/* kill 3 cells */
for (i=0; i<=2; i++) {
109
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
/* Store Node0Lat and Node0Lon as origin x and y */
fscanf(pInput, "%f", &home_lat);
past_y = (int) ((69.1/size)*(home_lat - 38));
c = fgetc(pInput);
fscanf(pInput, "%f", &home_lon);
past_x = (int) ((54.454/size)*(home_lon + 76));
c = fgetc(pInput);
/* kill 2 cells */
for (i=0; i<=1; i++) {
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
/* if Trip Type = 0, kill rest of line and start over */
if (i_Type == 0) {
c=0;
while (c != '\n'){
c = fgetc(pInput);
}
counter++;
goto start;
}
/* read in Node 0 Departure time */
else {
fscanf(pInput, "%f", &f_DepTime);
c = fgetc(pInput);
/* Case: At-Home Worker */
if (f_DepTime < 0) {
/* kill 8 cells */
for (i=0; i<=7; i++) {
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
/* read in Node 1 departure time */
fscanf(pInput, "%f", &f_DepTimeNext);
c = fgetc(pInput);
/* Kill 1 Cell */
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
/* Case: Anything else */
else {
/* Kill 4 Cells */
for (i=0; i<=3; i++) {
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
110
/* Read in Node 1 Lat and Lon */
fscanf(pInput, "%f", &origin_lat);
origin_y = (int) ((69.1/size)*(origin_lat - 38));
c = fgetc(pInput);
fscanf(pInput, "%f", &origin_lon);
origin_x = (int) ((54.454/size)*(origin_lon + 76));
c = fgetc(pInput);
/* Kill 1 Cell */
c=0;
while (c != ',') {
c = fgetc(pInput);
}
/* Read in ArrTime, DepTimeNext, distance */
fscanf(pInput, "%f", &f_ArrTime);
c = fgetc(pInput);
fscanf(pInput, "%f", &f_DepTimeNext);
c = fgetc(pInput);
fscanf(pInput, "%f", &distance);
c = fgetc(pInput);
/* print ID, Times, Coordinates, Distances */
/* fprintf(pOutput, "%d,", i_Person);*/
fprintf(pOutput, "%f,", f_DepTime);
/* fprintf(pOutput, "%f,", f_ArrTime);*/
fprintf(pOutput, "%d,", past_x);
fprintf(pOutput, "%d,", past_y);
fprintf(pOutput, "%d,", origin_x);
fprintf(pOutput, "%d\n", origin_y);
/* fprintf(pOutput, "%f,", distance); */
/* print new trip distance based on grid*/
/* fprintf(pOutput, "%f\n",
size*sqrt(((origin_x-past_x)*(origin_x-past_x)) +
((origin_y-past_y)*(origin_y-past_y))));*/
}
/* Kill 1 Cell */
c=0;
while (c != ',') {
c = fgetc(pInput);
}
/***** NORMAL *****/
while (c != '\n') {
f_DepTime = f_DepTimeNext;
if (f_DepTime < 0) {
counter++;
goto start;
}
past_x = origin_x;
past_y = origin_y;
/* Kill 2 Cells */
for (i=0; i<=1; i++) {
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
/* Read in Node n Lat and Lon */
fscanf(pInput, "%f", &origin_lat);
c = fgetc(pInput);
if (origin_lat == -1) {
/* Kill 3 Cells */
111
for (i=0; i<=2; i++) {
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
/* Read in DepTimeNext */
fscanf(pInput, "%f", &f_DepTimeNext);
c = fgetc(pInput);
/* Kill 1 Cell */
c=0;
while (c != ',') {
c = fgetc(pInput);
}
}
else {
origin_y = (int) ((69.1/size)*(origin_lat - 38));
fscanf(pInput, "%f", &origin_lon);
origin_x = (int) ((54.454/size)*(origin_lon + 76));
c = fgetc(pInput);
/* Kill 1 Cell */
c=0;
while (c != ',') {
c = fgetc(pInput);
}
/* Read in ArrTime, DepTimeNext, distance */
fscanf(pInput, "%f", &f_ArrTime);
c = fgetc(pInput);
fscanf(pInput, "%f", &f_DepTimeNext);
c = fgetc(pInput);
fscanf(pInput, "%f", &distance);
c = fgetc(pInput);
/* print ID, Times, Coordinates, Distances */
fprintf(pOutput,
fprintf(pOutput,
fprintf(pOutput,
fprintf(pOutput,
fprintf(pOutput,
"%f,", f_DepTime);
"%d,", past_x);
"%d,", past_y);
"%d,", origin_x);
"%d\n", origin_y);
/* print new trip distance based on grid*/
/* fprintf(pOutput, "%f\n",
size*sqrt(((origin_x-past_x)*(origin_x-past_x)) +
((origin_y-past_y)*(origin_y-past_y)))); */
}
if (c == ',') {
while (c == ',') {
c = fgetc(pInput);
}
}
if(c != '\n') c = fgetc(pInput);
}
counter++;
goto start;
}
printf("Error Line %d\n", counter);
return 0;
}
112
/*******************************************************/
/* File: buckets.c
*/
/* Author: Chris Brownell
*/
/*
*/
/* Description: -Implementation of Equation 10
*/
/*
-Generates Output in Figure 31
*/
/*
*/
/*******************************************************/
#include <stdio.h>
#include <stdlib.h>
int main(int argc, char * argv[]) {
int i = 0;
char c = 8;
FILE * pInput;
FILE * pOutput;
FILE * pAppend;
int O_X = 0;
int O_Y = 0;
int D_X = 0;
int D_Y = 0;
float time = 0;
int O_X2 = 0;
int O_Y2 = 0;
int D_X2 = 0;
int D_Y2 = 0;
float time2 = 0;
float btime = 0;
int triptot = 0;
int passtot = 0;
int size = 6;
int limit = 5;
pAppend = fopen(argv[1], "a");
fprintf(pAppend,"%f,", -100.00);
fclose(pAppend);
pInput = fopen(argv[1], "r");
pOutput = fopen(argv[2], "w");
sscanf (argv[3],"%d",&size);
sscanf (argv[4],"%d",&limit);
/* killl one line */
while (c != '\n') {
fscanf(pInput, "%c", &c);
}
/* Read in five values */
fscanf(pInput,"%f", &time);
c = fgetc(pInput);
fscanf(pInput, "%d", &O_X);
c = fgetc(pInput);
fscanf(pInput, "%d", &O_Y);
c = fgetc(pInput);
fscanf(pInput, "%d", &D_X);
c = fgetc(pInput);
fscanf(pInput, "%d", &D_Y);
c = fgetc(pInput);
i = 1;
btime = time;
/*
/* Print out five values */
fprintf(pOutput, "%d,%d,%d,%d,%d,",(int) time,O_X,O_Y,D_X,D_Y);*/
triptot = 1;
start:
113
/* Read in five values */
fscanf(pInput, "%f", &time2);
/* printf("%f\n", time2); */
if (time2 < 0) {
/* fprintf(pOutput,"%d\n",i); */
passtot +=i;
goto end;
}
c = fgetc(pInput);
fscanf(pInput, "%d", &O_X2);
c = fgetc(pInput);
fscanf(pInput, "%d", &O_Y2);
c = fgetc(pInput);
fscanf(pInput, "%d", &D_X2);
c = fgetc(pInput);
fscanf(pInput, "%d", &D_Y2);
c = fgetc(pInput);
if ((D_X == D_X2) && (D_Y == D_Y2)
&& (O_X == O_X2) && (O_Y == O_Y2)) {
if (((time2 - btime) <= ((float) limit*60))
&& (i<size)) {
i++;
O_X = O_X2;
O_Y = O_Y2;
D_X = D_X2;
D_Y = D_Y2;
goto start;
}
else {
/* fprintf(pOutput,"%d\n",i);*/
passtot += i;
/* fprintf(pOutput, "%d,%d,%d,%d,%d,",
(int) time2,O_X2,O_Y2,D_X2,D_Y2);*/
triptot++;
O_X = O_X2;
O_Y = O_Y2;
D_X = D_X2;
D_Y = D_Y2;
btime = time2;
i = 1;
goto start;
}
}
else {
/* fprintf(pOutput,"%d\n",i); */
passtot += i;
/* fprintf(pOutput, "%d,%d,%d,%d,%d,",
(int) time2,O_X2,O_Y2,D_X2,D_Y2);*/
triptot++;
O_X = O_X2;
O_Y = O_Y2;
D_X = D_X2;
D_Y = D_Y2;
btime = time2;
i = 1;
goto start;
}
end:
printf("Total Trips = %d\n", triptot);
printf("Total Passengers = %d\n", passtot);
fclose(pInput);
fclose(pOutput);
return 0;
}
114
/*******************************************************/
/* File: counter.c
*/
/* Author: Chris Brownell
*/
/*
*/
/* Description: -Implementation of Equation 11
*/
/*
-Outputs Average Taxi Trip Distance
*/
/*
*/
/*******************************************************/
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
int main(int argc, char * argv[]) {
int index = 0;
FILE * pOutput;
FILE * pInput;
int i = 0;
char c = 8;
FILE * pAppend;
int dtime = 0;
int O_X = 0;
int O_Y = 0;
int D_X = 0;
int D_Y = 0;
int counters[48] = {0};
int atime = 0;
int qx = 0;
float fch = 0.0;
float side = 0.5;
int adddist = 0;
int dist = 0;
pAppend = fopen(argv[1], "a");
fprintf(pAppend,"%d,",-100);
fclose(pAppend);
pInput = fopen(argv[1], "r");
pOutput = fopen(argv[2], "w");
start:
fscanf(pInput,"%d",&dtime);
if (dtime < 0) goto end;
c = fgetc(pInput);
fscanf(pInput,"%d",&O_X);
c = fgetc(pInput);
fscanf(pInput,"%d",&O_Y);
c = fgetc(pInput);
fscanf(pInput,"%d",&D_X);
c = fgetc(pInput);
fscanf(pInput,"%d",&D_Y);
c = fgetc(pInput);
fscanf(pInput,"%d",&qx);
c = fgetc(pInput);
if (side == 1.5) {
if (qx == 1) fch=0;
else if (qx == 2) fch=1.41;
else if (qx == 3) fch=2.24;
else if (qx == 4) fch=3;
else fch = 3.41;
}
dist = (int) (((1.2*side*sqrt(((D_Y-O_Y)*(D_Y-O_Y))+
((D_X-O_X)*(D_X-O_X)))) + 2*1.2*fch));
atime = dtime +
(int) (((1.2*side*sqrt(((D_Y-O_Y)*(D_Y-O_Y))+
((D_X-O_X)*(D_X-O_X))))*120)+(2*1.2*fch*240));
for(i=0;i<48;i++) {
115
if ((dtime <= (30*60 + 30*60*i)) &&
(atime >= 30*60*i)) {
counters[i]++;
}
}
adddist = adddist+dist;
index++;
goto start;
end:
for(i=0;i<48;i++) {
fprintf(pOutput,"%d\n",counters[i]);
}
printf("Avg Dist = %2.2f\n",((float) adddist)/((float) index));
printf("index = %d\n",index);
fclose(pInput);
fclose(pOutput);
return 0;
}
116