Tutorial on AI-based Analytics in Traffic Management
Upcoming SlideShare
Loading in...5
×
 

Tutorial on AI-based Analytics in Traffic Management

on

  • 331 views

This is the tutorial on AI analytical techniques for traffic management presented at the IJCAI 2013 conference, Beijing, China presented by Biplav Srivastava and Akshat Kumar.

This is the tutorial on AI analytical techniques for traffic management presented at the IJCAI 2013 conference, Beijing, China presented by Biplav Srivastava and Akshat Kumar.

Statistics

Views

Total Views
331
Views on SlideShare
331
Embed Views
0

Actions

Likes
0
Downloads
12
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Tutorial on AI-based Analytics in Traffic Management Tutorial on AI-based Analytics in Traffic Management Presentation Transcript

  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Biplav Srivastava, Akshat Kumar IBM Research August 2013 IJCAI 2013 Tutorial, Beijing, China 1
  • 2 What to Expect: Tutorial Objectives •  The aim of the tutorial is to present •  AI-enabled analytics related to transportation in a consolidated manner from varied domains like transportation, social networks, planning and graph theory to early and experienced researchers. •  With this, we hope that results from different areas can be better reused leading to increased attention of the AI-community enabling smarter traffic management. •  Relation to other tutorials •  Tutorial describing the traffic space and relevance of AI techniques was held at 26th Conference on Artificial Intelligence (AAAI-12), at Toronto, Canada. Formally called “Tutorial Traffic Management and AI”, by Biplav Srivastava and Anand Ranganathan, its details are at: http://www.aaai.org/Conferences/AAAI/2012/aaai12tutorials.php •  Tutorial highlighting planning and scheduling techniques for traffic management was held at ICAPS 2010. Formally called “Planning and Scheduling for Traffic Control” by Scott Sanner. Its details are available at: http://users.cecs.anu.edu.au/~ssanner/Papers/traffic_tutorial.pdf •  Disclaimer: we are only providing a sample of traffic analytics space intended to match audience profile in the available time. IJCAI 2013 Tutorial, Beijing, China
  • 3 Outline 1.  Traffic Management Ecosystem Overview 1.  2.  3.  2.  Analytics 1.  2.  3.  4.  3.  Problem Perspective Basic Concepts Key Performance Indicators Journey Planning Traffic Light Coordination Carpooling Sensor Optimization Supporting topics 1.  2.  3.  Competitions, Datasets Practical Considerations Case Studies IJCAI 2013 Tutorial, Beijing, China View slide
  • 4 Outline – AAAI 2012 Tutorial (Reference) 1.  2.  3.  4.  5.  Traffic Management Problem Instrumentation 1.  Sensing traffic 2.  Traffic state estimation 3.  Optimizing and combining sensor data Interconnection 1.  Middleware 2.  Traffic standards Intelligence 1.  Path planning 1.  Simple Illustration 2.  Path Planning for Individual Vehicles 2.  End-user analytics 1.  Bus arrival prediction and journey planning, with state-of-art instrumentation 2.  Multi-modal journey planning, without sensors Supporting topics 1.  Traffic Simulators 2.  Practical considerations for real-world pilots Source: AAAI 2012 Tutorial, Traffic Management and AI, Biplav Srivastava and Anand Ranganathan IJCAI 2013 Tutorial, Beijing, China View slide
  • 5 Outline – ICAPS 2010 Tutorial (Reference) Outline • • • • • Motivation History Fundamentals Simulation Control – Single Intersection – Multiple Intersection • Future Source: ICAPS 2010 Tutorial, Planning and Scheduling for Traffic Control, Scott Sanner IJCAI 2013 Tutorial, Beijing, China
  • 6 Acknowledgements All our collaborators, and especially those in: •  City agencies around the world •  Boston, USA; New York/ New Jersey area, USA; Bay Area, USA; Dubuque, IA •  Dublin, Ireland, Stockholm, Sweden •  Ho Chi Minh City, Vietnam; New Delhi, India; Tokyo, Japan; Bengaluru, India; Singapore •  Nairobi, Kenya •  Academia •  IBM: Many including – Anand Ranganathan, Raj Gupta, Ullas Nambiar, Srikanth Tamilselvam, L V Subramaniam, Chai Wah Wu, Anand Paul, Milind Naphade, Jurij Paraszczak, Wei Sun, Laura Wynter, Olivier Verscheure, Eric Bouillet, Francesco Calabrese, Tsuyoshi Ide, Xuan Liu, Arun Hampapur, Nithya Rajamani, Vivek Tyagi, Raguram Krishnapuram, Shivkumar Kalyanraman, Manish Gupta, Nitendra Rajput, Krishna Kummamuru, Raymond Rudy, Brent Miller, Jane Xu, Steven Wysmuller, Alberto Giacomel, Vinod A Bijlani, Pankaj D Lunia, Tran Viet Huan, Wei Xiong Shang, Chen WC Wang, Bob Schloss, Rosario UscedaSosa, Anton Riabov, Magda Mourad, Alexey Ershov, Eitan Israeli, Evgenia Gyana R Parija, Ian Simpson, Jen-Yao Chung, Kohichi Kajitani, Larry L Light, Lisa Amini, Marco Laumanns, Mary E Helander/Waltham, Milind Naphade Ofer, Sebastien Blandin, Takayuki Osogami, Tony R Heritage, Ulisses Mello, Wei CR Ding, Wei CR Sun, Xiang XF Fei, Yu Yuan, Bipin Joshi For discussions, ideas and contributions. Apologies to anyone unintentionally missed. IJCAI 2013 Tutorial, Beijing, China
  • 7 Outline 1.  Traffic Management Ecosystem Overview 1.  2.  3.  2.  Analytics 1.  2.  3.  4.  3.  Problem Perspective Basic Concepts Key Performance Indicators Journey Planning Traffic Light Coordination Carpooling Sensor Optimization Supporting topics 1.  2.  3.  Competitions, Datasets Practical considerations Case Studies IJCAI 2013 Tutorial, Beijing, China
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Traffic Management Ecosystem Overview [Problem Perspective] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 8
  • We All See Traffic Daily. An Illustration from Across the Globe Characteristics New York City, USA New Delhi, India Beijing, China Moscow, Russia Ho Chi Minh City, Vietnam Sao Paolo, Brazil 1 How is traffic predominantly managed Automated control, manual control Manual control Automated control, manual control Automated, manual control Manual control Automated, manual control, Rotation system (# plate based) 2 How is data collected Inductive loops, cops, video, GPS Traffic surveys, cops Video, GPS, cops GPS, some video, cops Traffic surveys, cops Video, GPS, cops 3 How can citizens manage their resources GPS devices, alerts on radio, web, road signs (variable) Alerts on radio alerts on radio, road signs (variable), mobile alerts GPS, radio, road signs, mobile alerts Alerts on radio GPS devices, alerts on radio, web 4 Traffic heterogeneity by vehicle types(Low: <10; Medium 10-25; High: >25) Low High Low Low Medium Low 5 Driving habit maturity (Low: <10 yrs; Medium: 10-20; High: > 20) High Low Low Low Low Medium 6 Traffic movement Lane driving Chaotic Lane driving Lane driving Chaotic Lane Driving Source: Google map for New York City and New Delhi; Search done on Aug 20, 2010 IJCAI 2013 Tutorial, Beijing, China
  • Case Study: Boston Smarter  Ci*es  Challenge  –  Boston:  IBM,  in  collabora*on  with  the  City  of  Boston  and  Boston   University,  over  a  3-­‐week  period  in  June  2012     Report:   hIp://www.cityoLoston.gov/Images_Documents/IBM_SCC_Boston_Report_Final_LR_tcm3-­‐36399.pdf         Paper:   A  General  Approach  to  Exploit  Available  Traffic  Data  for  a  Smarter  City,     Biplav  Srivastava,  Raymond  Rudy,  Jane  Xu,  Brent  Miller,  Alberto  Giacomel,  Steven  Wysmuller  –  IBM;  Vineet   Gupta,  Nigel  Jacob,  Chris  Osgood,  Kevin  Parker  -­‐  City  of  Boston,  MassachuseIs,  USA;  Conor  Gately,  Lucy   Hutyra  -­‐  Boston  University,  USA,   20th  ITS  World  Congress  2013,  Tokyo,  Japan           10 IJCAI 2013 Tutorial, Beijing, China  
  • Conges*on  is  the  daily  pain  of  ci*es   The  costs  of  traffic  conges*on  are  enormous.   The  choices  that  drivers  make  affect  roadway  conges*on  and  air  quality  at  the   neighborhood,  city,  and  metropolitan  levels.   Vehicles  idling  in  traffic  cause  substan*ally  more  air  pollu*on  than  if  they  were   moving  at  op*mal  speeds.   In  Greater  Boston   117  million  hours  in  excess   Metro  aof  change   travel  *me   Drivers   rea  alone…     Exploding  popula*ons,  urbaniza*on,  globaliza*on  and   technology  are  driving   ost   $102   $2.3  billion  annual  cchange.   million  poten*al     opera*onal  savings   This  creates  unique  challenges  and  opportuni*es  for   transporta*on  providers.   IJCAI 2013 Tutorial, Beijing, China
  • Problem:  What  Conges*on  Means  to  Boston   Urban   Area   "    Travel  delays   "    Excess  fuel            consumed   "    Truck            conges*on            cost   Travel  Delay   (1000   Hours)   Los  Angeles   521,449   Area   Excess  Fuel   Consumed   …   Total  Conges;on  Cost   Rank   (1000   Gallons)   Rank   ($million)   Rank   1   278,318   1   10,999   1   10   51,806   11   2,393   11   …   Boston   Area   117,234   Source: 2011 Annual Urban Mobility Report. See http://mobility.tamu.edu/ums/ IJCAI 2013 Tutorial, Beijing, China
  • Poten*al:  Savings  Achievable  with  Known   Techniques   "    Significant  savings          possible  with  known          treatment  measures   "    In  greater  Boston  area,  much                    scope  for  savings  in            opera*ons     Urban   Area   Opera;onal  Treatment  Savings   Treatments   Los   Angeles   Area   Delay     (1000   Hours)   Rank   Cost   ($Million)   r,i,s,a,h   63,652   1   1,342.6   i,s,a   4,988   14   101.8   …   Boston   Area   i: freeway incident management s: street signal coordination a: arterial street access management Source: 2011 Annual Urban Mobility Report. See http://mobility.tamu.edu/ums/ IJCAI 2013 Tutorial, Beijing, China
  • Project  High-­‐Level  Goals  &  Objec*ves   •  Major  Goals   •     Reducing  CO2  emissions  to  support  Mayor’s  Climate  Ac*on  Plan   •     Analyzing  and  reducing  vehicle  miles  traveled  (VMT)   •     Providing  transporta*on  data  for  residents  to  make  choices            and  increase  their  quality  of  life   •  Major  Objec*ves   •   Data  needs  to  be  unlocked,  shared  and  analyzed   •  Prototype:  induc*ve  loops,  manual  counts   •  Future:  cameras,  GPS,  others   •  Show  what  is  feasible  in  near-­‐term  and  ar*culate  roadmap   for  future   IJCAI 2013 Tutorial, Beijing, China
  • Boston  Transporta*on  Data  State   Lots  of  Instrumenta*on…   Video   GPS   Not  enough  interconnec*on…   Unexploited  Intelligence…   Much  Data   Isolated  in   Silos   Mul*ple   Disconnected   Camera   Networks   Manual   Road   Sensors   Inaccessible   Data   Regional   IJCAI 2013 Tutorial, Beijing, China Insufficient   Data   Manual   Opera*ons   "  Boston  is  forward-­‐          thinking  &  progressive   "    Boston  recognizes        climate  &  traffic  goals        are  interconnected       Boston  is  na4onally   recognized  for   innova4on  
  • Result  1:  Publicly  Available  Data  for  Mul*ple  Consumers   "    Many  data  sources,  various  loca*ons  &  *mes     "      Stakeholders  can  access  data  easily  &  intui*vely     Researchers   Prac**oners   Planners   Engineers   Residents   "    Locate  available  data  sources     "    Zoom  in  to  areas  of  interest     "    Obtain  data       "    Drill  down  to  traffic  paIerns     "    Assess  environmental  factors       "    See  what  happens  in  real  *me     IJCAI 2013 Tutorial, Beijing, China
  • Result  2:  Street  Classifica*on  Based  on  Traffic  Volume   •   Assign  different  traffic        light  paIerns  for        different  streets,  *mes   •   Schedule  public  works        projects  to  minimize        traffic  impact   •   Detect  changes  in        traffic  paIerns  to  drive        policy  changes        (parking,  lanes,  street)   •   Assess  traffic  impact  of        new  landmarks   •   Inform  businesses,          developers   Commuting Early-Bird Going Home Night Owl Anomaly IJCAI 2013 Tutorial, Beijing, China Busy
  • Much  More  is  Possible…   As-­‐is  visualiza4on      Sensing  &  Assimila*on   "   From  Induc*ve  Loops,   " Tube  Counts,  Manual   Counts…     …to  Vehicle  GPS,   transporta*on  demands,   weather   So  What?   ü Locate  sensors  &  counts   Analy4cal  visualiza4on   Traffic  Awareness   Modeling  &  Simula*on   "   Real-­‐*me  traffic  informa*on   "   Hourly  conges*on  forecast   Smarter  City  Planning   making   planning   This  enables…   ü Find  peak  hours,  anomalies       ü Extract  traffic  paIerns   ü Verify  loop  accuracy   ü Es*mate  CO2  emission     IJCAI 2013 Tutorial, Beijing, China "   Traffic  predic*on   "   What-­‐if  traffic  decision   "   City  transporta*on   ü Iden*fy  faulty  loops   ü Verify  tube  count  accuracy     What-­‐if  visualiza4on   ü Predict  traffic  paIerns   ü Evaluate  traffic  policies  
  • Smarter  Transporta*on  Leads  to  a  Smarter  Boston   Traffic  informa*on   Traffic  predic*on   Dynamic  naviga*on   Route  planner   Road  warnings   Conges*on  alert   Driver  status   Parking  locator   Visitor information Location-based advertising m-Commerce Traffic & navigation e-Intelligent driving Environment protection Smarter City Boston Emergency & safety Emergency  assistance   Safety  alert   IJCAI 2013 Tutorial, Beijing, China Public  transport   Carbon  calculator   Vehicle  usage   Industry solutions Communication/ collaboration Group  tracking   Family  &  friends   Collabora*on   Fleet  management   Taxi  management   Emergency  management  
  • SCC Boston team with Mayor on June 27, 2012 (L  à  R)  Biplav  Srivastava,  Brent  Miller,  Steve  Wysmuller,   Alberto  Giacomel,  Raymond  Rudy,  Jane  Xu   Team at work – Source: Boston Globe article Press on the IBM SCC Boston team work: 1. Boston Globe, June 29, 2012 http://www.boston.com/business/technology/articles/2012/06/29/ ibm_gives_advice_on_how_to_fix_boston_traffic__first_get_an_app/ (Alternative: http://bostonglobe.com/business/2012/06/28/ibm-gives-advice-how-fix-bostontraffic-first-get-app/goxK84cWB9utHQogpsbd1N/story.html) 2. Popular Science, 2 July 2012 http://www.popsci.com/technology/article/2012-07/bostons-ibm-built-traffic-app-mergesmultiple-data-streams-predict-ease-congestion 3. Others: National Public Radio (USA), and a range of local TV stations on the work. IJCAI 2013 Tutorial, Beijing, China
  • What is Transportation’s Objective? •  Removing congestion? •  Moving people fast? •  Reducing city’s cost? •  Reducing environment impact? IJCAI 2013 Tutorial, Beijing, China 21
  • Understanding Who Wants to Move Source: http://www.vtpi.org/ tdm/tdm103.htm IJCAI 2013 Tutorial, Beijing, China 22
  • (a) 3 km trip (b) 6 km trip Car (car + motorcycle) takes least time regardless of distance unless there is congestion on the road. See speaker notes for more details. Which Mode of Transport is Best for a Particular Distance – Doorto-Door Trip Times (c) 12 km trip IJCAI 2013 Tutorial, Beijing, China (d) 24 km trip Source: Mythologies, Metros & Future Urban Transport , by Prof. Dinesh Mohan, TRIPP, 2008 Data used: •  international data on things like access time to public transportation was taken • Studies were done for Delhi Metro •  Minimum assumptions were made. • E.g., walking time for metro access (5 mins) is widely acceptable • See speaker notes too
  • Rela*onship  Between  Conges*on  and  Greenhouse  Gases       "    Very  low  and  very            high  traffic  speeds            lead  to  higher              emissions   "  Moderate  speed  has            low  emissions   Source: Traffic Congestion and Greenhouse Gases, by Matthew Barth and Kanok Boriboonsomsin From: http://www.uctc.net/access/35/access35_Traffic_Congestion_and_Grenhouse_Gases.shtml IJCAI 2013 Tutorial, Beijing, China
  • 25 Why Focus on Mere Congestion Leads to Anomalies §  The Pigou-Knight-Downs paradox •  States that adding extra road capacity to a road does not reduce travel time. •  This occurs because traffic may simply shift to the upgraded road from the other, making the upgraded road more congested. §  The Downs-Thomson paradox •  States that the equilibrium speed of car traffic on the road network is determined by the average door-to-door speed of equivalent journeys by public transport. •  This occurs when the shift from public transport causes a disinvestment in the mode such that the operator either reduces frequency of service or raises fares to cover costs. §  The Braess' paradox •  States that adding extra capacity to a network, when the moving entities selfishly choose their route, can in some cases reduce overall performance and increase the total commuting time. Details: Arnott, Richard and K.A. Small, 1994, “The Economics of Traffic Congestion,” American Scientist, Vol. 82, No. 5, pp. 446-455. Chengri Ding and Shunfeng Song , Paradoxes of Traffic Flow and Congestion Pricing, IJCAI 2013 Tutorial, Beijing, China
  • Focus Needed Current understanding stresses focus on Accessibility, the ability to reach desired services and activities, and not on mobility, the ability to physically move. If one solely focuses on congestion removal, it may lead to •  Promotion of more motorized travel •  Increase in pollution •  Not necessary reduction in traffic (due to paradoxes) Congestion removal is a side-product of good traffic management. IJCAI 2013 Tutorial, Beijing, China 26
  • Levers in Cities’ Hands for Traffic Management •  Physical infrastructure •  New flyovers, roads •  Expanding existing capacity •  New metros, … •  Policies •  Relocating businesses •  Incentivising public transport •  IT enabled technologies •  Intelligent Traffic/ Transportation Systems •  Social networking IJCAI 2013 Tutorial, Beijing, China
  • 28 Is Problem Just Supply-Demand Mismatch? •  Supply •  Roads and their capacity •  Personnel available •  Capital and operational budget •  Demands •  Travel needs of citizens •  Travel needs of organizations: businesses, governments •  Solution to mis-match? •  Keep building supply (roads, bridges, …) •  Keep reducing demand (restrict citizens, businesses, …) •  May work for some of the cities, for some of the time, •  but not for all of the cities and all of the time IJCAI 2013 Tutorial, Beijing, China
  • 29 Key Insights for a New Traffic Problem Look •  Two category of stake-holders putting their resources •  Public resources •  Example: $1M per month for a city with 5M population •  Example: $100M available for roads and bridges for the next 3 years. •  Private resources (often ignored) •  Example: Average time taken to travel 1 KM •  Example: Average $ needed to travel 1 KM •  Impact of time-frame •  In short-term, new physical supply cannot be created due to long construction time, time to hire personnel, etc •  Public resources have a short-term and a long-term cycle, while private resources for transportation need is time-frame insensitive •  Approach: optimize both public as well as private resources IJCAI 2013 Tutorial, Beijing, China
  • 30 Problem Statement for Traffic Management from Engineering Perspective •  Problem (Short Term) Match traffic demand to supply with optimal usage of available public resources and concomitant optimization of citizens’ private resources for travel needs •  Example: For a given day, •  minimize over-time payments to traffic personnel , while •  minimizing average commute time per km. •  Alternative: Service objectives can also be stated like average commute time per km be below 10 mins/ km. •  Implications: Conflicting goals: optimality of public resources (e.g., road, traffic cops) v/s optimality of individual’s resources (e.g., travel time) Details: Biplav Srivastava. A new look at the traffic management problem and where to start. In 18th ITS Congress, Orlando, USA, Oct 16-20, 2011. IJCAI 2013 Tutorial, Beijing, China
  • 31 Problem Statement for Traffic Management from Engineering Perspective •  Problem (Short Term) Match traffic demand to supply with optimal usage of available public resources and concomitant optimization of citizens’ private resources for travel needs •  •  Example: For a given day, minimize over-time payments to traffic personnel while minimizing average commute time per km. Service objectives can also be stated like average commute time per km be below 10 mins/ km. Implications: Conflicting goals: optimality of public resources (e.g., road, traffic cops) v/s optimality of individual’s resources (e.g., travel time) •  Problem (Long Term) For a known future period, match traffic demand to supply with optimal usage of available and planned public resources and concomitant optimization of citizens’ private resources for travel needs. •  Example: For the next 3 years, minimize city’s expenses (annual operational budget and capital investment in the period) while minimizing average travel time per km and average fare per km. Service objectives can also be stated towards citizens like average commute time be below 10 mins/ km and average fare be below $1/ km. •  Implications: Information is used to plan city’s regions (including businesses, communities) and roads, thereby influencing traffic patterns Details : Biplav Srivastava. A new look at the traffic management problem and where to start. In 18th ITS Congress, Orlando, USA, Oct 16-20, 2011. IJCAI 2013 Tutorial, Beijing, China
  • 32 Starting Point from Problem Statement for Traffic Management •  Problem (Short Term) Match traffic demand to supply with optimal usage of available public resources and concomitant optimization of citizens’ private resources for travel needs •  •  •  Example: For a given day, minimize over-time payments to traffic personnel while minimizing average commute time per km. Service objectives can also be stated like average commute time per km be below 10 mins/ km. Implications: Conflicting goals: optimality of public resources (e.g., road, traffic cops) v/s optimality of individual’s resources (e.g., travel time) Problem (Long Term) For a known future period, match traffic demand to supply with optimal usage of available and planned public resources and concomitant optimization of citizens’ private resources for travel needs. •  •  Example: For the next 3 years, minimize city’s expenses (annual operational budget and capital investment in the period) while minimizing average travel time per km and average fare per km. Service objectives can also be stated towards citizens like average commute time be below 10 mins/ km and average fare be below $1/ km. Implications: Information is used to plan city’s regions (including businesses, communities)and roads, thereby influencing traffic patterns •  Starting point for any solution requires traffic to be measured accurately at sustained, continuous basis, at a rate that helps effective traffic control Details: Biplav Srivastava. A new look at the traffic management problem and where to start. In 18th ITS Congress, Orlando, USA, Oct 16-20, 2011. IJCAI 2013 Tutorial, Beijing, China
  • Consulting Solution Approach “Average” City A Summary of where Global Cities are within the Transportation Maturity Model Top 3 City: range Leading Practice Level 1 Level 2 Level 3 Silo Centralized Partially Integrated Level 4 Level 5 Multimodal Integrated Multimodal Optimized Planning Integrated corridorbased multimodal planning Performance Measurement Minimal Defined metrics by mode Limited integration across organizational silos Shared multimodal system-wide metrics Minimal capability, no customer accounts Customer accounts managed separately for each system/mode Multi-channel account interaction per mode Unified customer Integrated multimodal account across multiple incentives to optimize modes multimodal use Limited or Manual Input Near real-time for major routes Real-time for major routes using multiple inputs Real-time coverage for major corridors, all significant modes System-wide real time data collection across all modes Data Integration Limited Networked Common user interface 2-way system integration Extended integration Analytics Ad-hoc analysis Periodic, Systematic analysis High-level analysis in near real-time Detailed analysis in real-time Multi-modal analysis in real-time Payment Methods Manual Cash Collection Automatic Cash Machines Electronic Payments Multimodal integrated fare card Multimodal, multimedia (fare cards, cell phones, etc) Network Ops. Response Ad-Hoc, Single Mode Centralized, Single Mode Automated, Single Mode Automated, Multimodal Multimodal Real-time Optimized Incident Management Manual detection, response and recovery Manual detection, coordinated response, manual recovery Automatic detection, coordinated response and manual recovery Automated preplanned multimodal recovery plans Dynamic multimodal recovery plans based on real-time data Demand Management Individual static measures Individual measures, with long term variability Coordinated measures, with short term variability Dynamic pricing Multimodal dynamic pricing Traveler Information real-time intervention capability Integrated agency wide planning (single mode) Data Collection real-time information creation capability Project-based Planning (single mode) Customer Management strategic planning Functional Area Planning (single mode) Static Information Static trip planning with limited real-time alerts Multi-channel trip planning and accountbased alert subscription Location-based, onjourney multimodal information Location-based, multimodal proactive re-routing Multimodal Network Management Maturity Model version 1.1 IJCAIMore details at: http://www-935.ibm.com/services/us/igs/pdf/transport-systems-white-paper.pdf 2013 Tutorial, Beijing, China Integrated regional multimodal planning Continuous systemwide performance measurement © Copyright IBM Corporation 2007
  • Technology Approach An Intelligent Transportation Integration & Analytics Framework Collection Integration & Analysis Command & Control Dissemination Transport Management Subsystems (Traffic Signals, Bridges, Parking, Mgt, etc) Roadside Infrastructure Integration Mapping Services (Creation and Maintenance of maps etc) Data Fusion Data Fusion Traffic Control Center Dashboard (Creation of Single View) (Creation of Single View) In-Vehicle Detectors (Probes etc) Data Warehouse (Reporting, Archiving) Geographic Geographic Information Information System System Third Party Information Feeds (e.g. Floating Cellular Data) Mass Transit Mass Transit Operational Operational Management Management Vehicle Identification Security (timetables, etc) (timetables, etc) Systems Management IJCAI 2013 Tutorial, Beijing, China Traveller Information Portal (Info Display, Route Planning) Route Guidance & Trip Planning CRM (Call Centre, Interactive Voice) Finance en& Paymts Pricing Models Pricing Models Integration (Loops, CCTV, Tag & Beacon etc) Integration Infrastructure Detectors Data Analytics (Forecasting , Journey Time Determination) Incident Management (Process Execution) (Variable Message Signs, Traffic Signals etc) ITS Asset Management, Maintenance & Optimization Traveler Traveller Information Information Gateway (B2B info Feeds, Gateway etc). (B2B info feeds etc) In - vehicle Devices Mobile Devices Web Browsers Third Party Systems Self - service Traveler Portal (view bill, pay charge, etc) External Systems (Vehicle Licensing, Electronic Payments, Bill Printing, etc) Public Access Devices (e.g. kiosks)
  • 35 References •  Mythologies, Metros & Future Urban Transport , by Prof. Dinesh Mohan, TRIPP, 2008 •  A new look at the traffic management problem and where to start, by Biplav Srivastava, In 18th ITS Congress, Orlando, USA, Oct 16-20, 2011. •  Arnott, Richard and K.A. Small, 1994, “The Economics of Traffic Congestion,” American Scientist, Vol. 82, No. 5, pp. 446-455. •  Chengri Ding and Shunfeng Song, Paradoxes of Traffic Flow and Congestion Pricing, 2008 •  Online Transport Demand Managament (TDM) Encyclopedia, http:// vtpi.org/tdm/ IJCAI 2013 Tutorial, Beijing, China
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Traffic Management Ecosystem Overview [Basic Concepts] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 36
  • Basic Terms •  Delay: Extra travel time compared to some standard. •  Roadway Capacity: The “supply” of road transportation; measured in vehicles per hour per lane. •  Travel Time Index: The ratio of the travel time during the peak period to the time required to make the same trip at free-flow speeds. IJCAI 2013 Tutorial, Beijing, China 37
  • 38 Sensed Traffic Metrics •  Traffic  flow  –  number  of  vehicles*  that  pass  a  certain  point  during  a   specified  time  interval  (vehicles/hour)   •  Speed  –  rate  of  motion  in  distance/time  (mph)   •  Density  –  number  of  vehicles  occupying  a  given  length  of  highway  or   lane  (vehicles  per  mile  per  lane) •  Spacing  –  the  distance  between  successive  vehicles  in  a  traffic  stream  as   they  pass  some  common  reference  point  on  the  vehicles   •  Time  headway  –  the  time  between  successive  vehicles  in  a  traffic   stream  as  they  pass  some  common  reference  point  on  the  vehicles   * What is a vehicle? There are 48 types in India! So measurement can be non-trivial! IJCAI 2013 Tutorial, Beijing, China
  • Traffic  Flow  and  Time  Headway   Traffic  Flow  given  by:       q= traffic flow in vehicles per unit time t= duration of time interval n= number of vehicles passing some designated roadway point during time interval t IJCAI 2013 Tutorial, Beijing, China n q= t
  • Space  Mean  Speed   •  Time  necessary  for  a  vehicle  to  travel   some  known  length  of  roadway     l us = t us= space-mean speed in unit distance per unit time l = length of roadway used for travel time measurements of vehicles t = vehicle travel time 1 n t = ∑ ti n i =1 ti= time necessary for vehicle i to travel a roadway section of length l IJCAI 2013 Tutorial, Beijing, China
  • Time  Mean  Speed   •  Measure  of  speed  at  a  certain  point  in  space   Arithmetic  mean  of  vehicles  speeds  is  given  by:     n ut = ∑u i i =1 n ut=time-mean speed in unit distance per unit time ui=spot speed of the ith vehicle n=number of measured vehicle spot speeds IJCAI 2013 Tutorial, Beijing, China
  • Direct v/s Indirect Measures - A Feel For Traffic Sensing No.   I2:  20km/hr   Ability for using in sensing   Collected by manual method   Document   Can be used right after they are integrated, cleaning and conversion   I2   Mobile data   CDR   Data will be converted to transport data in standard format during project implementation   I3   I2:  20km/hr   Format   I1   I3:  28km/hr   I1:  25km/hr   Data type   Data from mobile which have GPS device   Follow format of data traffic   Data will be collected directly during project deployment and switch to the format needed   I2:  4  km/hr   I1:  25km/hr   Situation: Sensor readings from different type Problem: What is the various locations; for some links, no sensor present overall view of traffic? IJCAI 2013 Tutorial, Beijing, China
  • Reference •  Mobility  related  terms,  Annual  Urban  Mobility  Report, http://mobility.tamu.edu/ums/media-information/glossary/ IJCAI 2013 Tutorial, Beijing, China 43
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Traffic Management Ecosystem Overview [Key Performance Indicators ] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 44
  • Performance Indicators •  Process-centric view of indicators •  Effectiveness indicators – measures that the users of a system see •  Efficiency indicators – measures that the performers in a system see •  Activity-centric view of indicators •  Input – resources invested in a particular activity •  Outputs – •  Primary: direct result of a particular activity •  Secondary, also called outcomes: cumulative results •  Others – related to categorization of the process or activity IJCAI 2013 Tutorial, Beijing, China 45
  • Example: New York City Indicators Source: IJCAI 2013 Tutorial, Beijing, China http://www.vtpi.org/tdm/ tdm131.htm 46
  • Measuring Modes for Comparison Source: http://www.vtpi.org/tdm/ tdm131.htm IJCAI 2013 Tutorial, Beijing, China 47
  • Common Indicators •  Awareness – the portion of potential users who are aware of a program or service. •  Response – the number of people who respond to an outreach effort, such as asking for information on alternative •  •  •  •  •  •  •  •  travel modes in response to a promotion campaign. Participation – the number of people who use a service or alternative mode. Utilization – the number or portion of trips that use a travel service or alternative mode. Mode split – the portion of travelers who use each transportation mode. Mode shift – the number or portion of automobile trips shifted to other modes. Average Vehicle Occupancy (AVO): Number of people traveling in private vehicles divided by the number of private vehicle trips. This excludes transit vehicle users and walkers. Average Vehicle Ridership (AVR): All person trips divided by the number of private vehicle trips. This includes transit vehicle users and walkers. Vehicle Trips or Peak Period Vehicle Trips: The total number of private vehicles arriving at a destination (often called “trip generation” by engineers). Vehicle Trip Reduction – the number or percentage of automobiles removed from traffic. •  Vehicle Miles of Travel (VTM) Reduced – the number of trips reduced times average trip length. •  Energy and emission reductions – these are calculated by multiplying VMT reductions times average vehicle energy consumption and emission rates. •  Cost Per Unit of Reduction – these measures of cost-effectiveness are calculated by dividing program costs by a unit of change. For example, the cost effectiveness of various TDM programs could be compared based on cents per trip reduced, or ton of air pollution emission reductions. However, cost-effectiveness analysis that only considers direct impacts and a single objective may overlook additional costs and benefits to participants and society. For example, two TDM programs may have the same direct costs per unit of emission reduction, but differ significantly in terms of consumer costs, consumer travel options, traffic congestion, parking costs, crash risk and land use impacts. Source: http://www.vtpi.org/tdm/ IJCAI 2013 Tutorial, Beijing, China tdm55.htm 48
  • Reference •  Transportation Demand Management (TDM) Encyclopedia, http://www.vtpi.org/tdm/tdm12.htm •  Performance Measurement, http://www.vtpi.org/tdm/tdm131.htm IJCAI 2013 Tutorial, Beijing, China 49
  • 50 Outline 1.  Traffic Management Ecosystem Overview 1.  2.  3.  2.  Analytics 1.  2.  3.  4.  3.  Problem Perspective Basic Concepts Key Performance Indicators Journey Planning Traffic Light Coordination Carpooling Sensor Optimization Supporting topics 1.  2.  3.  Competitions, Datasets Practical Considerations Case Studies IJCAI 2013 Tutorial, Beijing, China
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Analytics [Journey Planning] Speaker: Akshat Kumar IBM Research IJCAI 2013 Tutorial, Beijing, China 51
  • Outline •  Theory: Route finding/ planning •  Deterministic world model •  In the presence of uncertainty •  Case studies at the end IJCAI 2013 Tutorial, Beijing, China 52
  • Route Planning: Basics 1 •  Given a graph G=(V,E) •  Source S, destination T 8 2 10 •  Edge length 5 •  Goal: Find shortest path between S and T •  Dijkstra’s algorithm 9 7 3 S 5 IJCAI 2013 Tutorial, Beijing, China T 4
  • Using Classical Planning in Route Finding Settings •  Logistics domain in planning competitions test different variations of basic route finding •  Routing vehicles on a road network maximizing throughput •  Jimoh, Falilat, McCluskey, T.L, Chrpa, Lukas and Gregory, Peter (2012) Enabling Autonomic Properties in Road Transport System. In: 30th Workshop of the UK Planning And Scheduling Special Interest Group PLANSIG 2012, 13th and 14th of December 2012, Teeside University. •  Routing vehicles that need to travel in a convoy •  Representation, Modelling and Planning Routes for Multiple Entities with Non-Negligible Length Travelling in a Shared Transportation Network, Shri. Anand Kumar R, M.Tech., IIT Madras, 2013 IJCAI 2013 Tutorial, Beijing, China 54
  • Route Planning: Basics 1 •  Given a graph G=(V,E) •  Source S, destination T 8 2 10 •  Edge length 5 •  Goal: Find shortest path between S and T •  Dijkstra’s algorithm 9 7 3 S 5 •  Stochastic world, uncertain edge travel time •  Decision-theoretic optimization, generalized cost •  Adaptive plan execution, closed loop policy IJCAI 2013 Tutorial, Beijing, China T 4
  • Characteristics of Route Planning Stochastic Adaptive Time Dependent Uncertainty Deadline Generalized Cost Function Stochastic Route Planning Y N N Y Y Canadian Traveler Problem Y Y N N N Multimodal Route Planning Y Y Y Y N Stochastic OPs Y N Y Y Y IJCAI 2013 Tutorial, Beijing, China 56
  • Roadmap: Stochastic Route Planning •  Decision theoretic planning with deadline •  Adaptive route planning [Nikolova et al. ICAPS06] [Provan Networks03, Nikolova et al. AAAI08] •  Markov decision processes framework •  Canadian traveler problem (CTP) •  Special cases of CTP, optimality criterion •  Adaptive planning with time varying uncertainty [Botea et al. ICAPS13] •  Route planning for theme-park experience [Lau et al. UAI12, Varakantham et al. ADT13] IJCAI 2013 Tutorial, Beijing, China 57
  • Planning Under Uncertainty Setting •  Given a directed graph, source S, dest. T •  Travel time on edges random variables •  A given deadline, a cost function C(t) to assess the cost of arriving at time t. x5T 5 x15 2 xS2 S x4T x25 1 xS1 T xS3 4 x 24 xij ⇠ fij (·) x34 3 IJCAI 2013 Tutorial, Beijing, Planning Under Uncertainty. E. Nikolova, M. Brand and D. Karger. ICAPS 2006 Optimal Route China
  • Cost Function •  Obvious solution: Replace edge length by expected length •  Decision theoretic setting: Associate a cost function for arriving at time t relative to a deadline C : < ! <+ •  E.g. Assume deadline is d, C(t) = (t d)2 •  What is the optimal start time for a given deadline? •  For a given start time, what is the optimal route? •  Choice of different cost functions IJCAI 2013 Tutorial, Beijing, China 59
  • Problem Definition i t t S x1 ECP (t) = xij ⇠ fij (·) EC(t) = j i Z j 0 r 1Y i=1 | fei (xi ) C t + {z } r X Z 1 fij (x)C(t + x)dx 0 xr T xi . . . dxi . . . i=1 | {z } Joint prob. of •  Find optimal path, start time: min ECP (t) Total P,t duration edge durations •  Find optimal path for a given start time: min ECP (t0 ) P IJCAI 2013 Tutorial, Beijing, China 60
  • Quadratic/Exponential Cost Function •  How to compute convolution ECP(t)? •  Closed form expression in certain cases •  C(t) = t2 •  Penalize early as well as late arrivals equally ✓ ECP (t) = t + r X µi i=1 C(t) = t2 + ekt , •  ◆2 + r X 2 i i=1 0 •  Higher penalty for being late IJCAI 2013 Tutorial, Beijing, China
  • Complexity Results •  Optimal route and start time with quadratic cost: min ECP (t) P,t •  Deterministic shortest path algorithm •  Optimal route and start time with quadratic+exponential cost •  Deterministic shortest path algorithm •  Finding the optimal route for a given start time: min ECP (t0 ) P •  NP-Hard even with quadratic cost function •  Pseudo-polynomial time dynamic programming IJCAI 2013 Tutorial, Beijing, Planning Under Uncertainty. E. Nikolova, M. Brand and D. Karger. ICAPS 2006 Optimal Route China 62
  • Summary •  Stochastic route planning •  Planning with expected cost not enough •  Decision theory + route planning: an attractive alternative •  How to chose appropriate cost function •  Design optimal paths and start time •  Computational complexity Vs. expressive power tradeoff •  Sophisticated cost functions make problem NP-Hard •  How to design approximate algorithms? •  How to make planning adaptive? IJCAI 2013 Tutorial, Beijing, China 63
  • Characteristics of Route Planning Stochastic Adaptive Time Dependent Uncertainty Deadline Generalized cost function Stochastic route planning Y N N Y Y Canadian Traveler Problem Y Y N N N Multimodal Route Planning Y Y Y Y N Stochastic OPs Y N Y Y Y IJCAI 2013 Tutorial, Beijing, China 64
  • Canadian Traveler Problem (CTP) •  Generalization of shortest path problem to partially observable setting 5 T x25 =10 1 2 4 x 24 S 3 IJCAI 2013 Tutorial, Beijing, China Route Planning Under Uncertainty: The Canadian Traveler Problem. E. Nikolova and D. Karger. AAAI 65 2008
  • CTP: Setting •  Initially edge cost unknown •  Traveler observes cost for edges incident at the node •  Input: Random variables modeling edge cost •  Output: Policy mapping observations to actions •  Goal: Least cost policy to reach from source to destination Features: •  Stochastic optimization •  Exploration Vs. exploitation tradeoff •  Sequential decision making IJCAI 2013 Tutorial, Beijing, China 66
  • CTP and Sequential DM •  Map CTP to a Markov decision process (MDP) •  Exploit MDP algorithms based on dynamic programming a s World s s s 0 5 T 10 1 2 S IJCAI 2013 Tutorial, Beijing, China C(s, a) 4 3 4 •  Actions: Which edge to traverse next •  State: <Nodeid, <edge_cost_history>> ⌦ N 2, hxS2 = 5, x25 = 10, x24 = 4i ↵
  • Solving CTP Using MDP Algorithms •  State space exponential even for small graphs Ø MDP algorithms no longer feasible •  Focus on special cases (easier versions, approximations) Ø CTP with resampling Ø Directed acyclic graphs, disjoint path graphs Ø Heuristic algorithms •  Open questions, further research IJCAI 2013 Tutorial, Beijing, China 68
  • CTP With Resampling 5 T 10 1 2 S 4 4 ⌦ State: N 2, hxS2 = 5, x25 = 10, x24 = 4i 3 With Resampling: 5 T 12 10 1 S 2 7 4 4 ⌦ ↵ State: N 2 Significant reduction in state-space! 3 IJCAI 2013 Tutorial, Beijing, China 69 ↵
  • Solving CTP With Resampling V (N5 ) 5 T 10 V (N4 ) 4 1 V (Ni ) : Expected cost to destination T 2 from Ni 4 ✓ aN2 = min 10 + V (N5 ), 4 + V (N4 ) S 3 IJCAI 2013 Tutorial, Beijing, China How to compute V(.) for nodes? 70 ◆
  • Solving CTP With Resampling V (N5 ) 5 T 10 V (N4 ) 4 1 2 S 4 •  Goal: Compute V(Ni) for each node •  Bellman optimality equations V (NT ) = 0 V (Ni ) = E 3  min Nj 2Nb(Ni ) xij + V (Nj ) •  Can computed efficiently with a Dijkstra-like algorithm A Tutorial, Beijing, Algorithm to Find Shortest Paths With Recourse. J. Provan. Networks 2003. IJCAI 2013 Polynomial Time China 71
  • Special Cases •  Directed acyclic graph Ø Compute value function in a similar fashion as with resmapling •  Disjoint-path graphs IJCAI 2013 Tutorial, Beijing, China 72
  • Open Questions •  Open questions Ø Find a compact policy representation for general CTP problems Ø Exploit advances in factored MDPs for the CTP Ø Handle richer partial observability using partially observable MDPs (POMDPs) Ø Uncertainty about location as well as travel duration •  Reuse plan from simpler setting to more complex planning problems •  Finding plans in simple setting is easier, faster; more complex assumptions may not hold during executions •  Plans of simple setting may be used for bootstrapping in complex setting IJCAI 2013 Tutorial, Beijing, China 73
  • Characteristics of Route Planning Stochastic Adaptive Time Dependent Uncertainty Deadline Generalized cost function Stochastic Route planning Y N N Y Y Canadian Traveler Problem Y Y N N N Multimodal Route Planning Y Y Y Y N Stochastic OPs Y N Y Y Y IJCAI 2013 Tutorial, Beijing, China 74
  • Planning In the Real World •  Dynamically changing distribution of travel time •  Multimodal journey planning •  Adaptive decision making •  Risk sensitive plans IJCAI 2013 Tutorial, Beijing, China 75
  • Multi-Modal Journey Planning •  Combine multiple transport in a trip •  Practical and commercial interest •  Increasingly available data—static public transport timetable, dynamic GPS data •  Availability on smart phones 76 IJCAI Multimodal Journey Planning in the Presence of Uncertainty. A. Botea, M. Berlingerio, E. Nikolova. ICAPS 2013 2013 Tutorial, Beijing, China
  • Problem Setting IJCAI 2013 Tutorial, Beijing, China 77
  • Algorithms and Planning Complexity •  Heuristic search in AND/OR search space •  Robust plan: minimizing the maximum arrival time •  Computational complexity of worst case plan •  Stationary travel durations—polynomial complexity •  Dynamically varying travel time distributions •  E.g. peak and off-peak distributions •  NP-Hard, reduction from subset-sum IJCAI 2013 Tutorial, Beijing, China 78
  • Characteristics of Route Planning Stochastic Adaptive Time Dependent Uncertainty Deadline Generalized cost function Stochastic route planning Y N N Y Y Canadian Traveler Problem Y Y N N N Multimodal Route Planning Y Y Y Y N Stochastic OPs Y N Y Y Y IJCAI 2013 Tutorial, Beijing, China 79
  • Planning Theme Park Experience •  Theme park route guidance •  Stochastic and time varying queue times at attractions •  Differing utilities for various attractions •  Main difference from route planning •  Risk attitude w.r.t. a deadline IJCAI 2013 Tutorial, Beijing, China 80
  • Dynamic Stochastic OPs (DSOPs) R(N5 ) •  A graph G=(V,E) t •  Edge length xij •  Travel and queuing time 5 xt 15 R(N1 ) •  Vertex reward R(.) xt s1 ✏ •  Goal: Simple path from S to T that maximizes total reward and P (aT  H) R(N2 ) 1 •  Deadline H •  Risk preference 1 T S 4 2 xt s2 xt s3 3 R(N3 ) ✏ 81 IJCAI 2013 Tutorial, Beijing, China Orienteering Problems for Risk Aware Applications. H. C. Lau et al. UAI 2012 Dynamic Stochastic R(N4 )
  • Chance Constrained Optimization •  DSOP an instance of chance-constraint optimization prob. ⇥ ⇤ min{g(x) := EP G(x, W ) } x2X s.t. prob F (x, W )  0 1 ✏ •  Sample average approximation (SAA) N 1 X min G(x, W i ) x2X N i=1 s.t. pN (x)  ✏ ˆ Samples from p(W) Approximate objective function •  Quality bounds, solve using MILP Approximate probability of constraint vilation Optimization Approaches for Solving Chance Constrained Stochastic Orienteering Problems. P. Varakantham and A. 82 Kumar. ADT 2013 IJCAI 2013 Tutorial, Beijing, China
  • Summary Stochastic Adaptive Time Dependent Uncertainty Deadline Generalized cost function Stochastic Route Planning Y N N Y Y Canadian Traveler Problem Y Y N N N Multimodal Route Planning Y Y Y Y N Stochastic OPs Y N Y Y Y IJCAI 2013 Tutorial, Beijing, China 83
  • Reference •  Optimal Route Planning Under Uncertainty. E. Nikolova, M. Brand and D. Karger. ICAPS 2006 •  Route Planning Under Uncertainty: The Canadian Traveler Problem. E. Nikolova and D. Karger. AAAI 2008 •  A Polynomial Time Algorithm to Find Shortest Paths With Recourse. J. Provan. Networks 2003. •  Multimodal Journey Planning in the Presence of Uncertainty. A. Botea, M. Berlingerio, E. Nikolova. ICAPS 2013 •  Dynamic Stochastic Orienteering Problems for Risk Aware Applications. H. C. Lau et al. UAI 2012 •  Optimization Approaches for Solving Chance Constrained Stochastic Orienteering Problems. P. Varakantham and A. Kumar. ADT 2013 •  Analysis of multimodal journey planners using a multi-criteria evaluation method, by Domokos Esztergr-Kiss*, Dr Csaba Csiszr, 19th ITS World Congress, Vienna, Austria IJCAI 2013 Tutorial, Beijing, China 84
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Analytics [Traffic Light Coordination] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 85
  • Traffic Lights •  First traffic signal system in the United States was implemented in 1912 to prevent traffic crashes. •  272,000 traffic signal systems in the United States in 2007. New role is better management of demand and supply gap. •  Case studies •  Denver case study: traffic signal system improvement program resulted in total delay reduction of ~36,000 vehicle hours per day and reduction in fuel consumption of 15,000 gallons per day between 2003-2008. (Source: Denver Regional Council of Government. The Denver Region Traffic Signal System Improvement Program. FHWA-HOP-09-046. Federal Highway Administration, Washington, DC, 2009.) •  Virginia case study: 30% reduction of travel times over uncoordinated approach; cost / benefit ration of 461.3 compared to non-coordinated automated traffic signal system (Source: Quantifying the Benefits of Coordinated Actuated Traffic Signal Systems: A Case Study, Byungkyu (Brian) Park and Yin Chen, 2010) Time Space Diagram. Source: City of Irvine Website - http://www.cityofirvine.org IJCAI 2013 Tutorial, Beijing, China 86
  • Basic Setting •  Each intersection is made up of a set of entry and exit roads •  The traffic light at an intersection cycles through a fixed sequence of phases, I. Each phase prescribes compatible movement of traffic from entry to exit on that intersection. •  The Signal Sequence is a sequence of phases along with its associated duration. •  Objective: minimize the total cumulative delay of vehicles traveling through the road network over a time period. •  Constraints •  The yellow light after each phase has to be of a pre-fixed duration •  Each phase duration is within a minimum and maximum range IJCAI 2013 Tutorial, Beijing, China 87
  • Traffic Signal Plans Example (Right-Hand Drive Roads) Source: Schedule-Driven Coordination for Real-Time Traffic Network Control Xiao-Feng Xie, Stephen F. Smith and Gregory J. Barlow, ICAPS 2012 IJCAI 2013 Tutorial, Beijing, China 88
  • Source: ICAPS 2010 Tutorial, Planning and Scheduling for Traffic Control, Scott Sanner IJCAI 2013 Tutorial, Beijing, China 89
  • SCATS: Key Details •  SCATS manages the dynamic (on-line, real-time) timing of signal phases at traffic signals •  Tries to find the best phasing for the currently sensed traffic situation •  Selection is automatic, but has access to a library of plans •  Works for individual intersections as well as for a road (sub-) network. •  Sensors can be •  Vehicles volume from inductive loop sensors, pneumatic tubes,… •  Pedestrians from road-side push buttons. •  Best results require assumptions about •  Vehicle types: car dimensions; roads: lanes; vehicle movement: in lanes •  Often the reason they do not work in chaotic traffic conditions Key Research Area: Manual selection of signal phases -> automatic selection IJCAI 2013 Tutorial, Beijing, China 90
  • Online Planning Setting •  Each intersection has its own signal sequence which is valid for some period •  Sequence of phases and durations •  At the end of the period, it has to decide on extending the signal sequence for the next time horizon. Decisions: •  Whether to extend with same phases •  Whether to change phases and durations •  For online problem •  Incoming clusters of vehicles are jobs •  Assignment of clusters to phases for a known time horizon is the scheduling problem. Output is a signal sequence. IJCAI 2013 Tutorial, Beijing, China 91
  • Simple Illustration Source: Schedule-Driven Coordination for Real-Time Traffic Network Control Xiao-Feng Xie, Stephen F. Smith and Gregory J. Barlow, ICAPS 2012 IJCAI 2013 Tutorial, Beijing, China 92
  • Coordinating Sets of Intersections •  Decentralized, uncoordinated, control: optimizes flow through each intersection independently •  Information sharing by a central controller for loose coordination: use schedules of neighbors for future demand response •  Active mis-coordination prevention •  Incoming and outgoing flows are used to detect traffic state •  If downstream intersection has insufficient capacity, can block traffic from upstream •  Resolve by cutting phase at current intersection IJCAI 2013 Tutorial, Beijing, China 93
  • Other Systems •  SCOOT (Split Cycle Offset Optimization Technique) •  Uses another set of advance vehicle upstream of the stop line to provide a count of the vehicles approaching at each junction. •  Data is used to create “cyclic flow profiles” and updated very frequently (every 4 seconds) •  Key traffic control parameters are directly adjusted •  Split: the amount of green light for each approach •  Offset: the time between adjacent signals •  Cycle: time allowed for all approaches to a signaled intersection •  Result highly dependent on sensor placement, can be quite responsive •  Many variations of SCATS and SCOOT IJCAI 2013 Tutorial, Beijing, China 94
  • Reference •  Schedule-Driven Coordination for Real-Time Traffic Network Control, Xiao- Feng Xie, Stephen F. Smith and Gregory J. Barlow, ICAPS 2012 •  Behrisch, M.; Bieker, L.; Erdmann, J.; and Krajzewicz, D. 2011. SUMO Simulation of Urban MObility: An overview. In International Conference on Advances in System Simulation, 63–68. •  ICAPS 2010 Tutorial, Planning and Scheduling for Traffic Control, Scott Sanner •  SCOOT: •  SCOOT User Guide, Siemens Mobility, Traffic Solutions UTC System, 666/HF/16940/000, 2012 •  SCOOT summary: At http://www.scoot-utc.com/DetailedHowSCOOTWorks.php?menu=Technical •  Wikipedia: http://en.wikipedia.org/wiki/Traffic_light_control_and_coordination IJCAI 2013 Tutorial, Beijing, China 95
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Analytics [Car Pooling] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 96
  • Acknowledgement: Dilbert IJCAI 2013 Tutorial, Beijing, China 97
  • Terminology Car Sharing Car Shuttle IJCAI 2013 Tutorial, Beijing, China Car Pooling
  • Carpooling •  Working Definition •  A group of people travelling together from one region to another region regularly with individual's duty to take their vehicles in rotation (and possibly driving) •  Key points •  Group: Two or people form the group, and this leads to social incentive •  One region to another region: •  For any person, the individual distance would have been smaller than when they travel as a carpool •  The desire is to have a small region at source and destination •  Regularly: If travel is not regular, it becomes hard to share responsibilities, schedule departure and arrival •  Rotating vehicle duty: •  Not driving on some days is a big incentive to be in a carpool •  If someone is in a car-pool but constantly does not take their vehicle, they are using the medium as a shuttle service IJCAI 2013 Tutorial, Beijing, China
  • Incentives and Disincentives in Carpooling •  Advantages 1.  Reduces exhaustion as the person does not have to drive daily 2.  Reduces expenses as the person does not have to take his car daily Enjoys the company of a group and its privileges - office politics Other benefits of reduced cars is to employers (less parking), city (less pollution) 3.  4.  •  Disadvantages 1.  Takes more time to drive on the person’s day of driving; never less than the shortest time 2.  If group’s schedule is not compatible, may waste more time at source or destination IJCAI 2013 Tutorial, Beijing, China
  • Myths •  Myth #1: Finding drivers and riders will increase car pooling •  Myth #2: Giving money to drivers will promote car pooling •  Myth #3: Car pooling can be the low-cost alternative for car shuttle IJCAI 2013 Tutorial, Beijing, China
  • Suggested Actions •  Action #1: Identify compatible groups with similar commuting needs •  Help group formation of people who travel regularly from same regions, have similar schedule and travel long distances •  Action #2: Register to recognize car pools •  Action #3: Respect a car pool’s schedule •  Provide support if management makes someone miss their carpool •  Action #4: Give incentives to car poolers at source, destination and en-route •  Reserved parking slots •  Prioritized passageway on roads IJCAI 2013 Tutorial, Beijing, China
  • Snapshot of Ride Sharing Systems IJCAI 2013 Tutorial, Beijing, China 103
  • Ridesharing Formation •  Kamar and Horvitz (IJCAI 2009) describe ABC system on ride share formation •  System looks at traces of how people have travelled in the past and searches for coordinated plans (ride shares) that could have had most savings (e.g., in terms of cars saved) at the least cost in terms of additional time. Figure source: http://www.cra.org/ccc/files/docs/seesit/thu_pm/Horvitz.pdf‎ IJCAI 2013 Tutorial, Beijing, China 104
  • What If a Whole City Adopts? Seattle. 215 morningevening commute pattern; average duration of 26 Figure source: http://www.cra.org/ccc/files/docs/seesit/thu_pm/Horvitz.pdf‎ mins & average distance of 21km for morning; 29 mins and 24 km for the evening. IJCAI 2013 Tutorial, Beijing, China 105
  • Ridesharing Formation •  ABC system has three main components: •  User-modeling component, that accesses and represents the preferences of agents from GPS traces and online calendars •  Optimization component, that generates collaborative rideshare plans •  Payment component, that provides incentives to agents to collaborate. Authors tested with different incentive schemes. •  Characteristics •  Assumes simply having overlapping paths will lead people to share rides •  Could work when monetary incentives is key (e.g., organized service providers). •  May not work when security and familiarity of the co-travellers is overriding concern since it is not modelled •  System predicts rides based on past travels •  Assumptions may not apply when a ride-share is actually being planned since calendar may not be up-to-data, all preferences may not be time-based, delays on some days more acceptable than others but known at the last minute, others preferred may not be travelling on the particular day. •  It is unclear if ridesharing recommendations were actually taken; nevertheless, useful for ride share planning IJCAI 2013 Tutorial, Beijing, China 106
  • Adaptive Public Transport Routing •  Type of ride sharing and practices •  Fixed-route bus transit systems •  Demand responsive transit (DRT) systems, e.g., shuttle vans, taxis (Palmer, Dessouky, and Abdelmaguid, 2004) •  Fixed routes more cost efficient than adaptive responsive methods - passenger loading capacity of the buses and the consolidation of many passenger trips onto a single vehicle (ridesharing). •  Mobility Allowance Shuttle Transit (MAST) •  fixed base route covers a specific geographic zone; shuttles are allowed limited freedom to deviate from the fixed path to pick up and drop off passengers from preferred locations IJCAI 2013 Tutorial, Beijing, China 107
  • Reference •  Making car pooling work – myths and where to start, Biplav Srivastava, ITS World Congress, 2012 •  Collaboration and Shared Plans in the Open World: Studies of Ridesharing, by Ece Kamar and Eric Horovitz, IJCAI 2009 •  Service Capacity Design Problems for Mobility Allowance Shuttle Transit Systems, Jiamin Zhao and Maged Dessouky, At http://www-bcf.usc.edu/~maged/publications/publications/ MASTcapacity.pdf IJCAI 2013 Tutorial, Beijing, China 108
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Analytics [Sensor Optimization] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 109
  • Evolution of Traffic Flow Sensor Technology Goal: Get traffic information (speed, volume) for city region on sustained, continuous basis, at a rate that helps effective traffic control at low cost Latest Buzz: •  Social Media •  Mobile phones After 2000 Mobile phones as traffic probes 1990s GPS-based floating car 1970s Video image processor 1960s Inductive-loop detector 1920s Auto traffic signal Accuracy & reliability Source: IBM China Research Lab IJCAI 2013 Tutorial, Beijing, China 110 •  Multiple lanes and zones monitoring •  Rich traffic data •  Flexibility •  Network-wide info •  Enlarged coverage •  Increased flexibility •  Rich traffic data through XFCD •  High spatial coverage •  Cost effectiveness •  Network-wide info. •  24x7 availability •  Weather insensitive •  Rich traffic data •  High flexibility
  • No Single Panacea Technology Complementarity Inductive-loop detector Video image processor Timeliness Timeliness Analysis easiness Accuracy Rich traffic data Analysis easiness Reliability Network-wide information Security & Privacy Resource consolidation Temporal coverage Low operational cost Spatial coverage Low capital cost Trustiness (legal) Accuracy Rich traffic data Reliability Network-wide information Security & Privacy Resource consolidation Temporal coverage Low operational cost Spatial coverage Low capital cost Trustiness (legal) Floating car data Mobile traffic probe Timeliness Timeliness Analysis easiness Accuracy Rich traffic data Reliability Network-wide information Security & Privacy Resource consolidation Temporal coverage Low operational cost Low capital cost Spatial coverage Trustiness (legal) Source: IBM China Research Lab IJCAI 2013 Tutorial, Beijing, China Analysis easiness Rich traffic data Network-wide information Resource consolidation Low operational cost Low capital cost Accuracy Reliability Security & Privacy Temporal coverage Spatial coverage Trustiness (legal) 111
  • Sensor Optimization •  Sensor Subset Selection •  Composite Sensing IJCAI 2013 Tutorial, Beijing, China 112
  • Sensor Subset Selection Problem •  City authorities need to decide what sensors to use to get traffic data for traffic of their region •  Slew of techniques available varying in accuracy, coverage and cost to install and maintain •  Diversity in how they can be set up •  How sensing methods may complement each other •  City can make an initial decision but it will need to be re-visited over time •  Traffic patterns changes •  Sensing technologies changes Problem : find the subset of sensors from available types that give the best cost-benefit for a given traffic pattern IJCAI 2013 Tutorial, Beijing, China
  • Effect of Increasing Number of Sensors (a) (b) [Traffic Pattern 1 on a Grid] Effect of increasing number of sensors of different type. a) Shows that error (Root Mean Square Error/ RMSE) decreases with increase in percentage of sensors from 10 to 100%. b) Shows increase in cost is highest for manual sensors whereas it is lowest for Call Data Record IJCAI 2013 Tutorial, Beijing, China
  • Subset Selection Approach •  Model sensor types based on cost, accuracy and coverage •  Create a sample space of sensor combination choices •  Use a traffic simulator (MATSIM) •  To measure the sensing error distribution entailed in each sensor combination •  To ensure physical characteristics of the city are taken into account •  Choose Pareto sensor combinations (non-dominated); Call this “Optimal Candidate Set (OCS)” •  Optional filtering steps Remove combinations above a give cost threshold •  Remove combinations above an error threshold •  Sensor Subset Selection for Traffic Management, R. Gupta and B. Srivastava, in 14th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2011), Washington DC, USA, Oct 5-7, 2011. Manual Video CDR GPS RMSE Cost Norm RMSE 100 0 0 0 1.2 4200 0 0 0 0 100 2.25 2520 0.028074866 0 0 100 0 6 840 0.128342246 0 0 40 0 20.55 335 0.517379679 0 0 10 0 38.6 82 1 •  For a given set of ‘k’ optimal combinations to be returned Select a preference function •  Use OCS selection using ICP approximation (Carlyle et al) •  Return ‘k’ optimal sensor combinations IJCAI 2013 Tutorial, Beijing, China RMSE •  0 0 0 0 45 0 40 0 0 35 0 30 0 0 25 0 20 10 0 15 0 10 0 0 5 0 0 10 10 20 30 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 20 30 40 50 60 70 80 90 100 100 100 100 100 100 100 0 0 10 2000 0 20 0 0 0 0 0 0 0 0 0 0 10 0 30 40 50 60 100 100 100 100 Cost 100 38.6 29.4 25.05 20.55 17.85 15.6 11.95 9.65 8.45 6 5.8 5.4 5.1 4.65 4.45 3.95 2.25 2.2 2.15 4000 2.05 2 82 167 247 335 414 500 586 668 753 840 1088 1175 1600 1835 2093 2356 2520 2855 2937 6000 3348 4005 Norm Cost 1 0.592034968 0.184069937 0.061437591 0 1.00 0.00 0.75 0.02 0.64 0.04 0.52 0.06 0.45 0.08 0.39 Non Selected 0.10 0.29 0.12 Pareto 0.23 0.14 Selected Pareto0.16 0.19 0.13 0.18 0.12 0.24 0.11 0.27 0.10 0.37 0.09 0.43 0.09 0.49 0.07 0.55 0.03 0.59 0.03 0.67 0.03 0.69 0.02 0.79 0.02 0.95
  • 116 Examples of City Preferences in Sensor Selection •  (Case A): a city which already has some sensors in place would want to add only those new sensor types which complement its investments •  Maximize incremental accuracy improvement •  (Case B): a city building the ITS infrastructure for the first time would want the highest accuracy possible within its budget •  Highest accuracy •  (Case C): a city may have low budgets and would want to have high coverage even at the loss of some accuracy (which it may compensate by putting more people on the field) •  Minimize cost •  (Case D): a city may want to minimize operational costs, even trading it off with higher capital expenditure, possibly because they come from a different budget •  Minimize operational cost IJCAI 2013 Tutorial, Beijing, China
  • 117 Key Results from Sensing •  Low-cost, noisy, CDR-based sensing complements existing sensors in a city due to easy coverage it provides •  Highly beneficial when other, more accurate, sensor types are few •  Benefit diminishes as the number of more accurate sensors increases •  Important result for traffic management of emerging countries •  Sensor type combinations can be suggested for a city’s traffic patterns using a simulator balancing accuracy and cost of sensing •  Returns non-dominated results •  Return few results approximating the search space IJCAI 2013 Tutorial, Beijing, China
  • Composite Sensing IJCAI 2013 Tutorial, Beijing, China 118
  • Situation •  Traffic is chaotic, heterogeneous •  Examples: Nairobi (Kenya), Delhi (India), Hyderabad (India) •  Sensors are coming up but are few, have low-resolution. Limited resources have to be stretched for maximum benefit. •  Videos are often first as they can be used for public safety as well •  Traditional sensors like inductive loops do not work due to vehicle and traffic diversity •  People may be willing to share traffic information •  Analytics role •  Complement limited data availability, noise in data •  Provide usable traffic management analytics that can scale with future sensor advances •  Approach: Detect traffic state rather than volume or speed •  As per DATEX 2, 5 states are necessary to manage: free flow, heavy, congested, impossible, Unknown IJCAI 2013 Tutorial, Beijing, China 119
  • Video + Simulation IJCAI 2013 Tutorial, Beijing, China Source: Frugal Innovation for Smarter Transportations in Developing Countries, Sachiko Yoshihama, April 2013 120
  • Image Processing, Network Analysis IJCAI 2013 Tutorial, Beijing, China Source: Frugal Innovation for Smarter Transportations in Developing Countries, Sachiko Yoshihama, April 2013 121
  • Audio + Video + Simulation Source:  Informa*on  fusion  based  learning  for  frugal  traffic  state  sensing,   Vikas  Joshi,  Nithya  Rajamani,  Takayuki  K,  Naveen  Prathapaneni,  L.  V.   Subramaniam,  IJCAI  2013   IJCAI 2013 Tutorial, Beijing, China 122
  • Sensing Fusion: Sample Results Overall classification results between 93 − 96% obtained in Delhi and Hyderabad. Source:  Informa*on  fusion  based  learning  for  frugal  traffic  state  sensing,  Vikas  Joshi,  Nithya  Rajamani,  Takayuki  K,  Naveen   Prathapaneni,  L.  V.  Subramaniam,  IJCAI  2013   IJCAI 2013 Tutorial, Beijing, China 123
  • Reference •  http://cacm.acm.org/magazines/2013/1/158775-human-mobility-characterization-from-cellular-network-data/fulltext •  Transportation Mode Inference from Anonymized and Aggregated Mobile Phone Call Detail Records, by Huayong •  •  •  •  •  •  •  •  •  •  •  •  Wang, Francesco Calabrese, Giusy Di Lorenzo,Carlo Ratti, 2010 13th International IEEE Annual Conference on Intelligent Transportation Systems, Madeira Island, Portugal, September 19-22, 2010 Mobile Landscapes: Using Location Data from Cell Phones for Urban Analysis, Carlo Ratti, Riccardo M. Pulselli, Sarah Williams, Dennis Frenchman, MIT, 2011 Handbook reference for : inductive loop detectors, magnetic sensors and detectors, video image processors, microwave radar sensors, laser radars, passive infrared and passive acoustic array sensors, and ultrasonic sensors, plus combinations of sensor technologies. http://www.fhwa.dot.gov/publications/research/operations/its/06108/ Ingram, J.W. The Inductive Loop Vehicle Detector: Installation Acceptance Criteria and Maintenance Techniques, Report No. TL 631387, prepared by California Department of Transportation, Sacramento, CA, for Federal Highway Administration, Washington, DC. U.S. Department of Commerce, National Technical Information Service, PB-263 948, Washington, DC, March 1976. N. Bansal, and B. Srivastava. On Using Crowd for Measuring Traffic at Aggregate Level for Emerging Countries. IIWeb workshop, WWW 2011, Hyderabad, India, March 28, 2011. H. Bischof, M. Godec, C. Leistner, B. Rinner, and A. Starzacher. Autonomous Audio-Supported Learning of Visual Classifiers for Traffic Monitoring. IEEE Intell. Sys., 25(3) pg 15–23, May/June 2010. M. Bramberger, R. Pflugfelder, A. Maier, B. Rinner, B. Strobl, and H. Schwabach. A Smart Camera for Traffic Surveillance. WISES03 pages 12, Vienna, Austria, June 2003. Y. Zhao. Mobile Phone Location Determination and Its Impact on Intelligent Transportation Systems. IEEE Trans. ITS, col. 1, NO. 1, Mar. 2000. M. A. Quddus, W. Y. Ochieng, and R. B. Noland. Current mapmatching algorithms for transport applications: Stateof-the art and future research directions. Transportation Research, Part C, vol. 15, no. 5, pp. 312–328, Oct. 2007. M. Prashanth, V. N. Padmanabhan, and R. Ramjee. Nericell: rich monitoring of road and traffic conditions using mobile smartphones. Proc. ACM Sensys, Pg. 323-336, 2008 Sensor Subset Selection for Traffic Management, R. Gupta and B. Srivastava, in 14th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2011), Washington DC, USA, Oct 5-7, 2011. V. Tyagi, S. Kalyanaraman, R. Krishnapuram. Vehicular Traffic Density State Estimation Based on Cumulative Road Acoustics, IEEE Transactions on Intelligent Transportation Systems, 2011 ``Horn-Ok-Please'', Rijurekha Sen, Bhaskaran Raman, Prashima Sharma, The Eighth International Conference on Mobile Systems, Applications, and Services, ACM MobiSys 2010, Jun 2010, San Francisco, CA IJCAI 2013 Tutorial, Beijing, China 124
  • 125 Outline 1.  Traffic Management Ecosystem Overview 1.  2.  3.  2.  Analytics 1.  2.  3.  4.  3.  Problem Perspective Basic Concepts Key Performance Indicators Journey Planning Traffic Light Coordination Carpooling Sensor Optimization Supporting topics 1.  2.  3.  Competitions, Datasets Practical Considerations Case Studies IJCAI 2013 Tutorial, Beijing, China
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Supporting Topics [Competition, Datasets] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 126
  • Traffic Datasets •  IEEE ICDM Contest: TomTomTraffic Predict for Intelligent GPS Navigation •  http://tunedit.org/challenge/IEEE-ICDM-2010 •  Predict future traffic based on historical data, •  Traffic reconstruction based on GPS traces •  Microsoft GeoLife GPS trajectories •  GPS data from 182 users •  1.2M Kms and 48K+ hours total trajectories •  http://research.microsoft.com/en-us/downloads/b16d359dd164-469e-9fd4-daa38f2b2e13/ IJCAI 2013 Tutorial, Beijing, China 127
  • Traffic Datasets •  511 SF Bay •  Provides traffic data feed to developers in SF Bay area •  Speed, travel time data for links in highways, freeway in Bay area •  Transit data (schedules, stops, fares etc) for 35 transit agencies •  http://511.org/ •  UK road traffic dataset •  Annual average daily flow count on major and minor roads •  http://data.gov.uk/dataset/gb-road-traffic-counts IJCAI 2013 Tutorial, Beijing, China 128
  • Electric Vehicles: Charge Car •  Optimizing energy efficiency of batteries in electric vehicles •  http://www.chargecar.org/ •  ChargeCar provides data about vehicle trips: •  GPS data •  Time •  Speed •  Acceleration •  Power •  ChargeCarPrize: Design an intelligent algorithms to decide how to use supercapacitor IJCAI 2013 Tutorial, Beijing, China 129
  • Comparison on Route Planners Transportation Community Analysis of multimodal journey planners using a multi-criteria evaluation method, by Domokos Esztergr-Kiss*, Dr Csaba Csiszr, 19th ITS World Congress, Vienna, Austria IJCAI 2013 Tutorial, Beijing, China 130
  • Comparison on Route Planners Transportation Community Analysis of multimodal journey planners using a multi-criteria evaluation method, by Domokos Esztergr-Kiss*, Dr Csaba Csiszr, 19th ITS World Congress, Vienna, Austria IJCAI 2013 Tutorial, Beijing, China 131
  • Reference •  Analysis of multimodal journey planners using a multi-criteria evaluation method, by Domokos Esztergr-Kiss*, Dr Csaba Csiszr, 19th ITS World Congress, Vienna, Austria IJCAI 2013 Tutorial, Beijing, China 132
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Supporting Topics [Practical Considerations] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 133
  • Sample Transportation Standards Source: Smarter city data model standards landscape, Part 2: Transportation Standard Examples of supported concepts Advanced Traveler Information Systems (ATIS), SAE 2354 Messages defined for traveler International standard (SAE) with traction in North AmericaGaining information, trip guidance, parking, momentum in the US; Nebraska, Washington and Minnesota Mayday (emergency information) planning support. See Washington State Department of Transportation Advanced Traveler Information Systems Business Plan for details. DATEX II Current deployments Traffic elements, operator actions, European standardVersion I deployed in several European impacts, non-road event countries. DATEX II deployment appears to be gaining momentum in information, elaborated data, and several European countries. See DATEX Deployments for details. measured data IEEE 1512 Common incident management International standardEarly deployments include Washington D.C., message sets for use by NYC, and Milwaukee. See emergency management centers, 1512 Public Safety Early Deployment Projects for details. traffic management, public safety, hazardous materials, and entities external to centers National Transportation Currently thirteen data dictionaries U.S. specific standardAccording to NTCIP web site, several cities Communication for ITS defined, including object definitions and states in the U.S. have NTCIP projects underway. See Protocol (NTCIP) 1200 for: dynamic message signs, CCTV NTCIP Deployment: Projects for details. Series camera control, ramp meter control, transportation sensor systems Service Interface for Real Information exchange of real-time European specific standardBased on best practises of various Time Information (SIRI) information about public national and proprietary standards from across Europe. transportation services and vehicles Traffic Management Data OwnerCenter, ExternalCenter, International standard (ITE and AASHTO) with traction in North Dictionary (TMDD) Device, DateTime, Dynamic AmericaIn early stages of deployment in the US. TMDD is backed by Message Sign, Event, Generic, US DoT and multiple vendors (Transcom, Siemens). Multiple state Organization, and RoadNetwork DoTs are planning to move to TMDD in their next rev (including Florida and Utah). See " A Report to the ITS Standards Community ITS Standards Testing Program, For TMDD and Related Standards as Deployed by the Utah Department of Transportation" for TMDD testing details from the Utah DoT. IJCAI 2013 Tutorial, Beijing, China
  • 135 Traffic Simulators Availability Licensing Credit: Table compiled by Raj Gupta, IBM Research Programming Code Development Current status language Extension Started Macro / micro Location traffic flow Dynamit Academia mit.edu/its/dynamit.html Freesim Open Source GNU java Possible 2004 last release 25-7-2011 both http:// www.freewaysimulator.com /index.html Sumo Open Source SourceForge C++ Possible 2002 upto date micro http:// sumo.sourceforge.net/ Microsimulation of Road Traffic Flow Academia java applet Possible Last release June 2011 micro http://www.trafficsimulation.de/ TraNS Academia Megaffic IBM MITSIMlab Open Source Free to use Transmodeler Single User licence 10 K Matsim Open source GNU 2007 http://lca.epfl.ch/projects/ trans 2008 http:// www.research.ibm.com/trl/ projects/socsim/project.htm IJCAI 2013 Tutorial, Beijing, China Possible No support since 2005 Possible 2003 last release 28-3-2011 micro mit.edu/its/mitsimlab.html micro Java 1999 http://www.caliper.com/ TransModeler/default.htm macro http://www.matsim.org/
  • 136 High-level Suggestions •  Follow and understand the data •  Problem characteristic depends on it •  Relevance of algorithms depends on it •  Result quality depends on it •  Prototype early •  Optionally, in a simulator, and •  Definitely, in real-world •  No solution works unless millions are using it •  Think long term •  Traffic systems last for decades, not 2-3 years (typical life of a computer) •  Support inter-operability in IT as well as in users •  Quantify benefits and costs using standard metrics •  Always think of people issues •  Good to have analogies with computer networks, but remember that packets can be dropped while travelling, but not people •  If people adopt, an imperfect solution will still be successful IJCAI 2013 Tutorial, Beijing, China
  • Reference •  Simulator links •  mit.edu/its/dynamit.html •  http://www.freewaysimulator.com/index.html •  http://sumo.sourceforge.net/ •  http://www.traffic-simulation.de/ •  http://lca.epfl.ch/projects/trans •  http://www.research.ibm.com/trl/projects/socsim/project.htm •  mit.edu/its/mitsimlab.html •  http://www.caliper.com/TransModeler/default.htm •  http://www.matsim.org/ IJCAI 2013 Tutorial, Beijing, China 137
  • AI-BASED ANALYTICS IN TRAFFIC MANAGEMENT Section: Supporting Topics [Case Studies] Speaker: Biplav Srivastava IBM Research IJCAI 2013 Tutorial, Beijing, China 138
  • 139 Theme: Increase Bus Information to Commuters to Promote Ridership Application: Application: Application: Journey Planning (Basic) Journey Planning (Condition Based) Medium: Medium: Medium: Information : Physical sensors info added - Information : 3rd party last mile info added - Location, schedules Location, schedules, bus personnel, 3rd party Integration: Integration: All phones, WWW Information : No new physical sensors Static bus schedules, inputs bus personnel Integration: Standards-aligned alerts Only smart phones WWW Standards-aligned alerts Technical / deployment complexity IJCAI 2013 Tutorial, Beijing, China Journey Planning (Door-to-door) All Phones, WWW Standards-aligned alerts
  • Case Study: Analytics with Little Instrumentation •  Scenarios •  S1: Journey Planning for Commuters •  S2: Planning Transportation in the City IJCAI 2013 Tutorial, Beijing, China
  • Promoting Public Transportation: Before and After We Seek Many cities around the world, and especially in India and emerging ones, are getting their transportation infrastructure in shape. –  They have multiple, fragmented, transportation agencies in a region (e.g., city) –  They do not have instrumentation on their vehicles, like GPS, to know about their operations in real-time –  Schedule of public transportation is widely available in semi-structured form. They are also beginning to invest in new, novel, sensing technologies –  Cities give SMS-based alerts about events on the road. Our approach seeks to accelerate time-to-value for such cities. Kind of Information Today Available to Bus User With IRL-Transit+ Benefit Bus Schedule (static) Available online and pamphlets Available from IT-enabled devices( low-cost phones, smart phones, web) Increase accessibility Bus Schedule Changes (dynamic) No information Infer from city updates Increase information Analytics (Bus Selection Decision Support) No information Will be available (Transit) Increase information Standardization of information No support Will be supported (Data Models, Transit) Increase information’s interoperability IJCAI 2013 Tutorial, Beijing, China
  • Prior Work •  Bay Area, USA has : http://511.org •  •  Multi-agency public authorities consortium, has advanced instrumentation It is the model to replicate §  Google has state-of-the-art from any non-public organization. It has separate services •  Maps for driving guidance •  Transit for public transport, more than 1 mode •  Gaps: •  Considers only time, not other factors like frequency, fare and waiting time •  Does not integrate across their services for different mode categories •  Does not publish their data •  Acknowledgement: We use their GTFS format to consolidate schedule data §  Many experimental systems with capabilities less than Google, •  Delhi: Disha on DIMTS website - http://61.16.238.196/disha/index.php •  Mumbai Navigator: http://www.cse.iitb.ac.in/navigator1/index.html •  Mumbai: Go4Mumbai (portal)- A http://www.go4mumbai.com/ §  Shortest route finding algorithms from mapping companies IJCAI 2013 Tutorial, Beijing, China
  • Problem •  Invariant Inputs: •  The person •  has a vehicle (e.g., car), and •  can also walk short distances •  The city has taxis, buses, metros, autos, rickshaws •  Buses and metros have published routes, frequency and stops •  Autos and rickshaws can be available at stands, or opportunistically, on the road •  Taxis can be ordered over the phone •  Input: •  A person wants to travel from place A to B •  Output •  Suggest to the person which mode or combination of modes to select •  Observation: Using preferences over factors that matter to users to keep commuting convenient, while making best use of available public and para-transit commute methods Paper: Making Public Transportation Schedule Information Consumable for Improved Decision Making, Raj Gupta, Biplav Srivastava, Srikanth Tamilselvam, In 15th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2012), Anchorage, USA, Sep 16-19, 2012. IJCAI 2013 Tutorial, Beijing, China
  • Background: Public Transportation Schedule Information •  Is widely available for public transportation agencies around the world •  Gives the basic, static, information about transportation service •  Usually in semi-structured format with varying semantics •  Can have errors, missing data IJCAI 2013 Tutorial, Beijing, China
  • Solution Steps •  Use the widely available schedule information from individual operators (agencies) •  Clean and consolidate it across agencies and modes to get a multi-modal view for the region •  Optionally: Convert it into a standard form •  Optionally: Enhance (fuse) it with any real-time updates about services for the region •  Perform what-if analysis on consolidated data •  Path finding using Djikstra’s algorithm •  Analyses can be pre-determined, analyses can also be user-created and defined •  Make analysis results available as a service •  On any device •  To any subscriber IJCAI 2013 Tutorial, Beijing, China
  • 1-Slide Summary: Multi-Mode Commuting Recommender in Delhi And Bangalore Highlights •  Published data of multiple authorities used; repeatable process • Multiple modes searched •  Preference over modes, time, hops and number of choices supported; more extensions, like fare possible •  Integration of results with map straigtforward IJCAI 2013 Tutorial, Beijing, China
  • Benefits of Our Approach Works with whatever data is available. The system can start with just published route information. Can quickly get started in a new city. Our process is fast and we can have a new city incorporated in days. Extendible a)  b)  c)  a)  b)  c)  d)  e)  Can scale based on dynamic information (e.g., GPS) Can work with GIS/ map information but that is not required. Support different types of preferences. Can make this information available on low-cost phones. We use a Research technology called Spoken Web. Can support context-based filtering of results. E.g., the user can only see reachable points IJCAI 2013 Tutorial, Beijing, China
  • 148 Technical Details – Factors Impacting Accuracy •  Quality of schedule published by public transportation operators (bus and metro for us) •  Names, spelling and conventions in stops by different agencies •  We correct and can do more - If we correct too much, we remove the traceability to original published schedules •  Lack of co-relationship across stop names and location •  Affects what the user sees when they select •  We can include geo-spatial analysis when we offer choice of locations •  Increase inter-operability across agencies •  Make traffic data into linked open data format •  Integrate with geo-spatial analysis like ESRI IJCAI 2013 Tutorial, Beijing, China IRL-Transit is only one type of decision support. One can build others for private para-transit agencies (e.g., radio taxis).
  • Incorporating Dynamic Updates •  Invariant Inputs: •  The person •  has a vehicle (e.g., car), and •  can also walk short distances •  The city has taxis, buses, metros, autos, rickshaws •  Buses and metros have published routes, frequency and stops •  Autos and rickshaws can be available at stands, or opportunistically, on the road •  Taxis can be ordered over the phone •  Input: •  A person wants to travel from place A to B •  [Optional] City provides updates on ongoing events, some may affect traffic •  Output •  Suggest to the person which mode or combination of modes to select •  Observation: Using preferences over factors that matter to users to keep commuting convenient, while making best use of available public and para-transit commute methods City Notifications as a Data Source for Traffic Management, Pramod Anantharam, Biplav Srivastava, in 20th ITS World Congress 2013, Tokyo, Japan IJCAI 2013 Tutorial, Beijing, China
  • Number of SMS messages for bus stops in Delhi for 2 years (Aug 2010 – Aug 2012)* •  344 stops with updates •  3931 total stops IJCAI 2013 Tutorial, Beijing, China * using Exact Matching
  • IRL – Transit in Aug 2012 Key Points • SMS message from city •  Event and location identified •  Impact assessed •  Impact used in search IJCAI 2013 Tutorial, Beijing, China
  • 152 Transportation Supply (Coverage) Analyses with IRL Transit Making Public Transportation Schedule Information Consumable for Improved Decision Making, Raj Gupta, Biplav Srivastava, Srikanth Tamilselvam, In 15th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2012), Anchorage, USA, Sep 16-19, 2012. IJCAI 2013 Tutorial, Beijing, China
  • Examples:  Supply-­‐side  Analyses   •  Q1: How well are stops connected in a city connected to each other, compared to other cities? •  Q2: What are the critical transit points? •  Q3: Which regions are under-served by the public transportation network? –  Requires additional information about demographics and region topology •  Q4: Where can new services be introduced? E.g., which regions are under-served by the public transportation network as well as well-connected by road network –  Requires additional information about the regions which are well- connected by road network IJCAI 2013 Tutorial, Beijing, China
  • A1:  City’s  Stop  Coverage  with  Route  Network   100 90 80 70 60 Del % 50 Ban % 40 30 20 10 0 0-H 1-H 2-H 3-H 4-H Method: Computed the % of stops that are covered in the city as the number of hops increase. The Result: Within 2 hops, most of the city’s stops are connected Delhi better connected (higher slope) than Bangalore IJCAI 2013 Tutorial, Beijing, China Hops   0-­‐H   1-­‐H   2-­‐H   3-­‐H   4-­‐H   Del  %   3.24   61.42   99.88   100.00   Ban  %   5.06   40.48   89.08   99.42   100.00  
  • A2: See Critical Transit Points in Cities ahinsa  sathal   aiims   ailway  colony   air  force  camp   air  force  h.qrs   air  force  sta*on   air  force  sta*on  bani  camp   ajmere  gate   akashwani   akashwani  bhawan   akashwani  bhawan/  ndpo   akbar  pur  majra  xing   akber  pur  vill   akshardham   alaknanda   alaknanda  /  narvada  appI   ali  pur   Delhi 8.635795   16.68753   1.835628   3.796412   2.336254   4.338757   4.58907   2.836879   1.627034   6.049228   1.168127   1.41844   1.084689   1.75219   0.70922   0.876095   6.758448   83.18732   98.99875   56.36212   68.58573   72.00668   62.0776   59.24072   76.76262   52.7743   88.65248   47.0171   44.13851   36.25365   45.8907   42.92866   39.50772   77.17981   bommanahalli  Bgl   bowring  ins*tute  Bgl   brigade  road  Bgl   bsk  3rd  stg  3rd  phase  Bgl   bsk  Bgl   bsk  bda  complex  Bgl   btm  layout  Bgl   byatarayanapura  Bgl   bydarahalli  Bgl   c.m.t.i  Bgl   c.v.raman  nagara  Bgl   cauvery  bhavan  Bgl   central  silk  board  Bgl   central  talkies  Bgl   Bangalore Major hubs (well connected) Medium hubs (somewhat connected) Method: • Compare only for 0- and 1- hops • Compare a stop’s coverage percentage with average of the city •  Decide a stop’s connectedness (high/ low) based on results for 0- and 1- hops IJCAI 2013 Tutorial, Beijing, China 20.19   88.82   4.66   41.93   7.14   30.43   1.55   37.89   6.83   48.76   3.73   47.20   14.60   86.02   4.04   38.51   3.73   35.09   3.11   36.02   1.55   15.22   1.55   17.70   8.39   68.94   4.04   39.44  
  • Sample of results for Delhi across Bus and Metro Network Case Study Number11 • Source S.P.DEPOT • Destination KRISHI BHAWAN Number of Solutions 64 A3:  Gaps  in     City’s  Mul*-­‐Modal   Coverage   Solution Number 1 Route Ids 170 [S.P.DEPOT, EAST PATEL NGR] 408 [EAST PATEL NGR, KRISHI BHAWAN] … Notes: •  ~15 kms, many choices •  Destination is government office district; not much traffic expected in nonworking time or days •  Taxi stand queues can be varied based on this Case Study Number 21 • Source MANSAROVER PARK • Destination WELCOME Number of Solutions 12 Problems with coverage Method: 1.  Query the system for commuting choices 2.  Make inference based on results, and background information like Maps and Demographics results Solution Number 12 Route Ids LINE-1 [MANSAROVER PARK, SHADARA] [Metro] 982EXT [SHADARA, WELCOME] [Bus] Case Study Number27 •  Source PANCHSHEEL CLUB •  Destination PUSHPANJLI ENCLAVE Number of Solutions 1 Solution Number 1 Route Ids 419EXT [PANCHSHEEL CLUB,ITO] 313 [ITO, PUSHPANJLI ENCLAVE] Notes: •  ~4 kms, only 1 choice •  Both are heavy residential areas •  Opportunity for taxis/ autos IJCAI 2013 Tutorial, Beijing, China
  • Sample of results for Delhi Example 1 A4:  Iden*fy  Scope  for   New  Services   Saket and Okhla Situation: Connectivity through metro and buses is not good enough. But road connectivity is high. Road Characteristics : Bridge and four lane road New Services: Direct bus services can be introduced Example 2 Source : Geeta Colony, Delhi Destination : Shahdara, Delhi Method: 1.  Query the system for commuting choice coverage 2.  Find road connectivity for query region 3.  Compare coverage and road connectivity to correlate IJCAI 2013 Tutorial, Beijing, China Solution Number Route Ids 622 [Geeta Colony, Krishna Nagar] 317 [Krishna Nagar, Shahdara] Note :- There is no one hop public transport is available whereas road distance is 4.5 km. Road Characteristics : No red Light
  • Case Study: Insights In Motion Contact: Milind Naphade Dubuque: http://www.cityofdubuque.org/index.aspx?NID=1496 Istanbul: Istanbul See 2 minute video. IJCAI 2013 Tutorial, Beijing, China 158
  • 159 City in Motion: Understanding People Movement for Improving Outcomes Information Sources Network Data (millions of events/day) Analytics & Models Time of Day Density Maps Business Services Planning Large Scale Events, Emergency Response Outcomes Reduce Congestion Smart Fare Card Data (millions of events/day) Reduce Journey Time Traffic Flow Optimizing Bus & Multimodal Transit Transit System & GIS Data Improve Customer Satisfaction Origin-Destination Optimal Facility/Warehouse Location Planning Census & Demographics Data 159 Improve Revenue Reduce Operating Expenses Deep Analysis Location-based Services, Traffic Alerts, Promotions IJCAI 2013 Tutorial, Beijing, China Reduce Carbon footprint Improve Quality of Life
  • 160 City in Motion – Bus Route Optimization in Dubuque Information Sources Network Data (millions of events/day) Smart Fare Card Data (millions of events/day) Analytics & Models Business Services Time of Day Density Maps Reduce Journey Time by 60% Traffic Flow Bus Route Optimization Transit System & GIS Data Origin-Destination Optimal Bus Stop Location Planning Census & Demographics Data 160 IJCAI 2013 Tutorial, Beijing, China Outcomes Deep Analysis Reduce Carbon footprint by 46% Reduce Operating Expenses by 40% Improve Quality of Service by reducing unment demand by 37%
  • 161 Bus Route Optimization to reduce operating costs, meet demand better Old Routes (Before Analysis) Optimal Routes (After Analysis) Clean sheet bus route optimization •  reduces operating costs by 40% •  reduces unmet demand by 37% •  reduces average travel time by 60% •  reduces GHG emissions by 45% IJCAI 2013 Tutorial, Beijing, China
  • 162 Optimal Stop/Location Placement to meet demand better Old Stops (Before Analysis) New Stops (After Analysis) Reduce unmet demand by 37% without increasing operating cost/travel time IJCAI 2013 Tutorial, Beijing, China
  • 163 City in Motion – Multimodal Transit Optimization in Istanbul Information Sources Network Data (millions of events/day) Smart Fare Card Data (millions of events/day) Analytics & Models Traffic Flow 163 Optimizing Bus & Multimodal Transit Reduce Congestion Reduce Journey Time Origin-Destination IJCAI 2013 Tutorial, Beijing, China Outcomes Time of Day Density Maps Transit System & GIS Data Census & Demographics Data Business Services Deep Analysis Reduce Carbon footprint
  • 164 Multimodal Transit Optimization to improve coordination, reduce congestion Metro Route IJCAI 2013 Tutorial, Beijing, China Feeder Bus Optimization for Metro
  • Reference •  Making Public Transportation Schedule Information Consumable for Improved Decision Making, Raj Gupta, Biplav Srivastava, Srikanth Tamilselvam, In 15th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2012), Anchorage, USA, Sep 16-19, 2012. •  City Notifications as a Data Source for Traffic Management, Pramod Anantharam, Biplav Srivastava, in 20th ITS World Congress 2013, Tokyo, Japan •  Milind R. Naphade, Guruduth Banavar, Colin Harrison, Jurij Paraszczak, Robert Morris: Smarter Cities and Their Innovation Challenges. IEEE Computer 44(6): 32-39 (2011) •  Ming Li, Jing Dai, Sambit Sahu, Milind R. Naphade: Trip analyzer through smartphone apps. GIS 2011: 537-540 IJCAI 2013 Tutorial, Beijing, China 165