Uncongested Mobility for All: A Proposal for an Area Wide Autonomous Taxi System in New Jersey By Jaison Zachariah ‘13 Jingkang Gao ‘13 Tala Mufti *13 Recent Grads, Operations Research & Financial Engineering Princeton University Alain L. Kornhauser *71 Professor, Operations Research & Financial Engineering Director, Program in Transportation Faculty Chair, PAVE (Princeton Autonomous Vehicle Engineering Princeton University Presented at

Outline What is an autonomousTaxi (aTaxi) Synthesizing an Appropriate Representation of All Person Trips in New Jersey on a Typical Weekday How much Ride-Sharing (AVO) could various aTaxi service offerings stimulate Next Step

Preliminary Statement of Policy Concerning Automated Vehicles Level 0 (No automation) The human is in complete and sole control of safety-critical functions (brake, throttle, steering) at all times. Level 1 (Function-specific automation) The human has complete authority, but cedes limited control of certain functions to the vehicle in certain normal driving or crash imminent situations. Example: electronic stability control Level 2 (Combined function automation) Automation of at least two control functions designed to work in harmony (e.g., adaptive cruise control and lane centering) in certain driving situations. Enables hands-off-wheel and foot-off-pedal operation. Driver still responsible for monitoring and safe operation and expected to be available at all times to resume control of the vehicle. Example: adaptive cruise control in conjunction with lane centering Level 3 (Limited self-driving) Vehicle controls all safety functions under certain traffic and environmental conditions. Human can cede monitoring authority to vehicle, which must alert driver if conditions require transition to driver control. Driver expected to be available for occasional control. Example: Google car Level 4 (Full self-driving automation) Vehicle controls all safety functions and monitors conditions for the entire trip. The human provides destination or navigation input but is not expected to be available for control during the trip. Vehicle may operate while unoccupied. Responsibility for safe operation rests solely on the automated system SmartDrivingCars & Trucks What is a SmartDrivingCar?

Preliminary Statement of Policy Concerning Automated Vehicles Level“Less”Value PropositionMarket ForceSocietal Implications 0 “55 Chevy” Zero 1 “Cruise Control” InfinitesimalSome ComfortInfinitesimal 2 “CC + Emergency Braking” InfinitesimalSome SafetySmall; Needs help From “Flo & the Gecko” (Insurance Industry) “20+%” fewer accidents; less severity; fewer insurance claims 3 “Texting Machine” SomeLiberation (some of the time/places) ; much more Safety Consumers Pull, TravelTainment Industry Push Increased car sales, many fewer insurance claims, Increased VMT 4 “aTaxi “ AlwaysGet to be Chauffeured; Get to Buy Mobility “by the Drink” rather than “by the Bottle” Profitable Business Opportunity for Utilities/Transit Companies Personal Car becomes “Bling” not instrument of personal mobility, VMT ?; Comm. Design ? Energy, Congestion, Environment? What is a SmartDrivingCar?

What if a “Community Design” (New Jersey) only had – Walking, – Bicycling, – NJ Transit Rail – aTaxis for mobility. What are the Societal Implications of that Mobility (Energy, Pollution, Congestion) ? (Hint: It’s all about Ride-Sharing!) What about Level 4 Implications on Energy, Congestion, Environment?

New Jersey’s existing Land-uses generate about 32 million Trips / Day – The Automobile (~ 28 million) – Walking + bicycling (~3 million) – Bus + rail Transit (~1 million) While Concentrated at some Times in some Corridors – Most of those trips are enormously diffuse in time and space New Jersey “Today”

Creating the NJ_PersonTrip file “every” trip that each Traveler is likely to make on a typical day. NJ_PersonTrip file {oLat, oLon, oTime, dLat, dLon, Est_dTime} Start with – NJ_Residentfile (120,000 Census Blocks) – NJ_Employment file (430,000 businesses) – NJ_School file (18,000 schools) Readily assign trips between Home and Work/School – Trip Activity -> Stop Sequence Home, Work, School characteristics synthesized in NJ_Resident file

Overview of Data Production 1.Generate each person that lives or works in NJ 2.Assign work places to each worker 3.Assign schools to each student 4.Assign tours / activity patterns 5.Assign other trips 6.Assign arrival / departure times Project Overview

Motivation – Publicly available TRAVEL Data do NOT contain: – Spatial precision Where are people leaving from? Where are people going? – Temporal precision At what time are they travelling? Trip Synthesizer (Activity-Based) Project Overview

Synthesize from available data: “every” NJ Traveler on a typical day NJ_Resident file – Containing appropriate demographic and spatial characteristics that reflect trip making “every” trip that each Traveler is likely to make on a typical day. NJ_PersonTrip file – Containing appropriate spatial and temporal characteristics for each trip

Creating the NJ_Resident file for “every” NJ Traveler on a typical day NJ_Resident file Start with Publically available data:

2010 Population Level – 8,791,894 individuals distributed 118,654 Blocks. CountyPopulationCensus BlocksMedian Pop/ BlockAverage Pop/Block ATL 274,549 5, BER 905,116 11, BUR 448,734 7, CAM 513,657 7, CAP 97,265 3, CUM 156,898 2, ESS 783,969 6, GLO 288,288 4, HUD 634,266 3, HUN 128,349 2, MER 366,513 4, MID 809,858 9, MON 630,380 10, MOR 492,276 6, OCE 576,567 10, PAS 501,226 4, SAL 66,083 1, SOM 323,444 3, SUS 149,265 2, UNI 536,499 6, WAR 108,692 2, Total 8,791, ,

Bergen Block Level CountyPopulationCensus Blocks Median Pop/ Block Average Pop/Block BER 907,128 11,

Publically available data: Distributions of Demographic Characteristics – Age – Gender – Household size – Name (Last, First) Ages (varying linearly over interval):input:output: [0,49]67.5% [50,64]18.0%17.9% [65,79]12.0%12.1% [80,100]2.5% Gender:Input:Output: female51.3% Household:Size:Probability:cdf:Expectation: couple couple couple couple couple couple + grandparent: single woman single mom single mom single mom single mom single man single dad single dad single dad

Beginnings of NJ_Resident file County 2010 Census # People, Lat, Lon, For each person Vital Stats RandomDraw: Age, M/F, WorkerType, Task 1 County Person Index Household IndexLast Name First Name Middle InitialAgeGender Worker IndexWorker Type Home Latitude Home Longitude 0 11PREVILLERICHARDG.24FALSE5worker PREVILLEJACKJ.7FALSE0grade School PREVILLECHARLESX.1FALSE7under DEVEREUXSUEB.24TRUE6at-home-worker DEVEREUXANTONP.2FALSE7under DEVEREUXKATIES.6TRUE0grade School WHEDBEELINDAC.26TRUE6at-home-worker CARVERROBERTZ.24FALSE5worker CARVERJENNIFERP.25TRUE6at-home-worker TINSLEYELLENU.23TRUE4college on campus: WorkerType IndexWorkerType String:Distribution: 0grade school100% ages [6,10] 1middle school100% ages [11,14] 2high school100% ages [15,18] 3college: commuteSate-wide distribution 4college: on campusSate-wide distribution 5workerDrawn to match J2W Stats by County 6at-home worker and retiredRemainder + 100% ages [65,79] 7nursing home and under 5100% ages [0,5] and 100% ages [80,100]

Using Census Journey-to- Work (J2W) Tabulations to assign Employer County County Person Index Household IndexLast Name First Name Middle InitialAgeGender Worker IndexWorker Type Home Latitude Home Longitude Employer County 011PREVILLERICHARDG.24FALSE5worker PREVILLEJACKJ.7FALSE0grade School PREVILLECHARLESX.1FALSE7under DEVEREUXSUEB.24TRUE6at-home-worker DEVEREUXANTONP.2FALSE7under DEVEREUXKATIES.6TRUE0grade School WHEDBEELINDAC.26TRUE6at-home-worker CARVERROBERTZ.24FALSE5worker CARVERJENNIFERP.25TRUE6at-home-worker TINSLEYELLENU.23TRUE4college on c ampus: WorkCounty Destination RandomDraw: Journey2Work Home County C2C Journey2Work Work County Task Home State Home CountyCounty Name Work State Work CountyCounty NameWorkers 34 1Atlantic Co. NJ659Orange Co. CA Atlantic Co. NJ685Santa Clara Co. CA9 34 1Atlantic Co. NJ103New Castle Co. DE Atlantic Co. NJ105Sussex Co. DE9 6 37L. A. Co. CA341Atlantic Co. NJ Riverside Co. CA341Atlantic Co. NJ7 9 3Hartford Co. CT341Atlantic Co. NJ5 9 5Litchfield Co. CT341Atlantic Co. NJ4

Using Employer Data to assign a Workplace Characteristics Name County NAICS Code NAICS Description Employ ment Latitude Longitude 1 VIP SKINDEEP Atlantic Other Personal Care Acres Motel Atlantic Hotels & Motels Ex Casino Grand Street Investors Atlantic Misc Financial Inves S Main St LLC Atlantic Lessors Of Res Buildg th Floor Creative Group Atlantic Motion Picture Prod Cab Co Atlantic Taxi Svc Junk Car Removal Atlantic Used Merch Stores Bar Atlantic Drinking Places Got-Junk? Atlantic Other Non-Haz Waste Disp Employment-Weighted Random Draw Employment-Weighted Random Draw

Using School Data to Assign School Characteristics

Assigning a Daily Activity (Trip) Tour to Each Person

Final NJ_Resident file Home County Person Index Household Index Full Name Age Gender Worker Type Index Worker Type String Home lat, lon Work or School lat,lon Work County Work or School Index NAICS code Work or School start/end time ATL 274,549 BER 905,116 BUR 448,734 CAM 513,657 CAP 97,265 CUM 156,898 ESS 783,969 GLO 288,288 HUD 634,266 HUN 128,349 MER 366,513 MID 809,858 MON 630,380 MOR 492,276 OCE 576,567 PAS 501,226 SAL 66,083 SOM 323,444 SUS 149,265 UNI 536,499 WAR 108,692 NYC 86,418 PHL 18,586 BUC 99,865 SOU 13,772 NOR 5,046 WES 6,531 ROC 32,737 Total: 9,054,849

Assigning “Other” Locations Attractiveness (i)= (Patrons (I)/AllPatrons)/{D(i,j) 2 + D(j,k) 2 }; Where i is destination county; j is current county; k is home county Attractiveness (i)= (Patrons (I)/AllPatrons)/{D(i,j) 2 + D(j,k) 2 }; Where i is destination county; j is current county; k is home county 1. Select Other County Using: Attractiveness-Weighted Random Draw 1. Select Other County Using: Attractiveness-Weighted Random Draw 2. Select “Other” Business using: Patronage-Weighted Random Draw within selected county 2. Select “Other” Business using: Patronage-Weighted Random Draw within selected county

Assigning Trip Departure Times For: H->W; H->School; W->Other Work backwards from Desired Arrival Time using Distance and normally distributed Speed distribution, and Non-symmetric early late probabilities Else, Use Stop Duration with non-symmetric early late probabilities based on SIC Cod For: H->W; H->School; W->Other Work backwards from Desired Arrival Time using Distance and normally distributed Speed distribution, and Non-symmetric early late probabilities Else, Use Stop Duration with non-symmetric early late probabilities based on SIC Cod Distribution of Arrival/Departure Times Trip Type; SIC Time Generator: RandomDraw: Time Distribution Trip Departure time (SeconsFromMidnight) Task 8

NJ_PersonTrip file 9,054,849 records – One for each person in NJ_Resident file Specifying 32,862,668 Daily Person Trips – Each characterized by a precise {oLat, oLon, oTime, dLat, dLon, Est_dTime} All Trips Home County TripsTripMilesAverageTM #Miles ATL 936,585 27,723, BER 3,075,434 40,006, BUC 250,006 9,725, BUR 1,525,713 37,274, CAM 1,746,906 27,523, CAP 333,690 11,026, CUM 532,897 18,766, ESS 2,663,517 29,307, GLO 980,302 23,790, HUD 2,153,677 18,580, HUN 437,598 13,044, MER 1,248,183 22,410, MID 2,753,142 47,579, MON 2,144,477 50,862, MOR 1,677,161 33,746, NOR 12, , NYC 215,915 4,131, OCE 1,964,014 63,174, PAS 1,704,184 22,641, PHL 46,468 1,367, ROC 81,740 2,163, SAL 225,725 8,239, SOM 1,099,927 21,799, SOU 34,493 2,468, SUS 508,674 16,572, UNI 1,824,093 21,860, WAR 371,169 13,012, WES 16, , Total 32,862, ,178,

NJ_PersonTrip file

Intra-pixel Trips Warren County Population: 108,692

What if the only way to get around was by – Walking, – Bicycling, – NJ Transit Rail – aTaxis What are the Societal Implications of this System (Mobility, Energy, Pollution, Congestion) ? (Hint: It’s all about Ride-Sharing!) aTaxi Implications on Mobility, Energy, Congestion, Environment

No Change in Today’s Walking, Bicycling and Rail trips – Today’s Automobile trips become aTaxi or aTaxi+Rail trips with hopefully LOTS of Ride-sharing opportunities aTaxi Implications on Mobility, Energy, Congestion, Environment

“AVO < 1” RideSharing – “Daddy, take me to school.” (Lots today) “Organized” RideSharing – Corporate commuter carpools (Very few today) “Tag-along” RideSharing – One person decides: “I’m going to the store. Wanna come along”. Other: “Sure”. (Lots today) There exists a personal correlation between ride-sharers “Casual” RideSharing – Chance meeting of a strange that wants to go in my direction at the time I want to go “Slug”, “Hitch hiker” Kinds of RideSharing

“AVO < 1” RideSharing – Eliminate the “Empty Back-haul”; AVO Plus “Organized” RideSharing – Diverted to aTaxis “Tag-along” RideSharing – Only Primary trip maker modeled, “Tag-alongs” are assumed same after as before. “Casual” RideSharing – This is the opportunity of aTaxis – How much spatial and temporal aggregation is required to create significant casual ride-sharing opportunities. aTaxis and RideSharing

By walking to a station/aTaxiStand – At what point does a walk distance makes the aTaxi trip unattractive relative to one’s personal car? – ¼ mile ( 5 minute) max Spatial Aggregation

Pixelation of New Jersey NJ State Grid Zoomed-In Grid of Mercer

Pixelating the State with half-mile Pixels Pixelating the State with half-mile Pixels xPixel = floor{ * (longitude )} yPixel = floor{138.2 * (latitude – 38.9)) xPixel = floor{ * (longitude )} yPixel = floor{138.2 * (latitude – 38.9))

a PersonTrip {oLat, oLon, oTime (Hr:Min:Sec),dLat, dLon, Exected: dTime} a PersonTrip {oLat, oLon, oTime (Hr:Min:Sec),dLat, dLon, Exected: dTime} O O D P1P1 An aTaxiTrip {oYpixel, oXpixel, oTime (Hr:Min:Sec), } An aTaxiTrip {oYpixel, oXpixel, oTime (Hr:Min:Sec), } An aTaxiTrip {oYpixel, oXpixel, oTime (Hr:Min:Sec),dYpixel, dXpixel, Exected: dTime} An aTaxiTrip {oYpixel, oXpixel, oTime (Hr:Min:Sec),dYpixel, dXpixel, Exected: dTime}

P1P1 O Common Destination (CD) CD=1p: Pixel -> Pixel (p->p) Ride-sharing Common Destination (CD) CD=1p: Pixel -> Pixel (p->p) Ride-sharing TripMiles = L TripMiles = 2L TripMiles = 3L

P1P1 O PersonMiles = 3L aTaxiMiles = L AVO = PersonMiles/aTaxiMiles = 3 PersonMiles = 3L aTaxiMiles = L AVO = PersonMiles/aTaxiMiles = 3

Elevator Analogy of an aTaxi Stand Temporal Aggregation Departure Delay: DD = 300 Seconds Elevator Analogy of an aTaxi Stand Temporal Aggregation Departure Delay: DD = 300 Seconds Kornhauser Obrien Johnson 40 sec Henderson Lin 1:34 Popkin 3:47

Samuels 4:50 Henderson Lin Young 0:34 Popkin 2:17 Elevator Analogy of an aTaxi Stand 60 seconds later Elevator Analogy of an aTaxi Stand 60 seconds later Christie Maddow 4:12

By walking to a station/aTaxiStand – A what point does a walk distance makes the aTaxi trip unattractive relative to one’s personal car? – ¼ mile ( 5 minute) max By using the rail system for some trips – Trips with at least one trip-end within a short walk to a train station. – Trips to/from NYC or PHL Spatial Aggregation

D a PersonTrip from NYC (or PHL or any Pixel containing a Train station) a PersonTrip from NYC (or PHL or any Pixel containing a Train station) NYC O Princeton Train Station NJ Transit Rail Line to NYC, next Departure aTaxiTrip An aTaxiTrip {oYpixel, oXpixel, TrainArrivalTime, dYpixel, dXpixel, Exected: dTime} An aTaxiTrip {oYpixel, oXpixel, TrainArrivalTime, dYpixel, dXpixel, Exected: dTime}

By walking to a station/aTaxiStand – A what point does a walk distance makes the aTaxi trip unattractive relative to one’s personal car? – ¼ mile ( 5 minute) max By using the rail system for some trips – Trips with at least one trip end within a short walk to a train station. – Trips to/from NYC or PHL By sharing rides with others that are basically going in my direction – No trip has more than 20% circuity added to its trip time. Spatial Aggregation

P1P1 P2P2 O CD= 3p: Pixel ->3Pixels Ride-sharing

P1P1 P5P5 O P3P3

P1P1 P5P5 O SP 4 P3P3 SP 1 SP 5 P6P6 SP 6 CD= 3sp: Pixel ->3SuperPixels Ride-sharing

c

Results

What about the whole country? Extending the Activity-Based Person-Trip Synthesizer to all 330 million Americans Judy Sun ‘14 & Luke Cheng ’14 ORF467 F13

Public Schools in the US

Quick stats on Public Schools (2011) School Type# of CHARTER# of PUBLICTotal Primary 2,584 51,79354,377 Middle ,33216,947 High 1,316 19,76221,078 Other 1,145 5,8476,992 No Answer 564 3,5254,089 Total6,22497,259103,483

Nation-Wide Businesses RankState Sales VolumeNo. Businesses 1California$1,8891,579,342 2Texas$2,115999,331 3Florida$1,702895,586 4New York$1,822837,773 5Pennsylvania$2,134550,678 9New Jersey$1,919428,596 45Washington DC$1,31749,488 47Rhode Island$1,81446,503 48North Dakota$1,97844,518 49Delaware$2,10841,296 50Vermont$1,55439,230 51Wyoming$1,67935, Million Businesses {Name, address, Sales, #employees}

US_PersonTrip file will have.. ~330 Million records – One for each person in US_Resident file Specifying ~1.2 Billion Daily Person Trips – Each characterized by a precise {oLat, oLon, oTime, dLat, dLon, Est_dTime} Will Perform Nationwide aTaxi AVO analysis Results ????

Thank You Discussion!

Scope of “Automated Vehicles” Rivium > Aichi, Japan, 2005 Expo Automated Guided Vehicles Tampa Airport 1 st APM 1971 Tampa Airport 1 st APM 1971 Rio Tinto Automated Truck Tesla Car Transporter CityMobil2 Copenhagen Metro Heathrow PodCar Milton Keynes, UK Elevator Mercedes Intelligent Drive Rio Tinto Automated Train

VisLab.it (U. of Parma, Italy) – Had assisted Oshkosh in DARPA Challenges – Stereo vision + radars SmartDrivingCars: Post-DARPA Challenges (2010-today) (Video has no sound)

Crash Mitigation (air bags, seat belts, crash worthiness, …) Crash Mitigation (air bags, seat belts, crash worthiness, …) Good News: Effective in reducing Deaths and Accident Severity Bad News: Ineffective in reducing Expected Accident Liability & Ineffective in reducing Insurance Rates Bad News: Ineffective in reducing Expected Accident Liability & Ineffective in reducing Insurance Rates Up to today: The Primary Purview of

Intelligent Drive (active steering ) Volvo Truck Emergency braking Click images to view videos S-Class WW Launch May ‘13 Frankfurt Auto Show Sept ‘13 MB Demo Sept ‘13