1 of 16

Using Artificial Intelligence to Enhance Decision Support for Corridor Management

AJ Skillern

Manager – R&D

1

2 of 16

Introduction

  • Southwest Research Institute
    • Independent non-profit Research & Design (R&D) organization
    • Over 25 years of Intelligent Transportation Systems (ITS) experience
      • ATMS
      • ICM
      • Connected and Autonomous Vehicles
      • Big data/machine learning

​

  • AJ Skillern
    • Over 11 years of ITS experience
    • Work with 10 different Departments of Transportation (DOTs) for ITS software projects

​

2

3 of 16

Agenda

  • SwRI’s experience with corridor management and decision support
  • Modeling and Simulation
  • Machine Learning
  • The importance of data
  • Why AI?
  • The future of AI in corridor management
  • How to prepare

​

3

4 of 16

SwRI’s experience

  • Regional Integrated Corridor Management (R-ICMS)
    • FDOT D5
    • Caltrans D12
  • Modeling and simulation for evaluating effectiveness of response plans

​

​

​

​

​

4

5 of 16

SwRI’s experience (cont.)

  • TDOT AI-DSS / I-24 Smart Corridor
    • Machine learning and real-time data analysis
      • Response plan scores
      • Diversion route path analysis
      • Variable speed limits

​

​

​

​

5

6 of 16

SwRI’s experience (cont.)

  • VDOT RM3P AI-DSS
    • Predicted congestion and parking availability
    • Modeling and simulation for response plan analysis
    • Multi-modal analysis

​

​

​

​

6

7 of 16

Modeling and Simulation

  • Simulation, optimization, and statistics that incorporate complex network models
  • Derive insights and provide forecasting and prediction capabilities
  • Traffic model created prior to system implementation
    • Based off historical data from various sources
  • Simulates predicted effects of lane closures, diversion routes, etc.

​

7

8 of 16

Modeling and Simulation (cont.)

  • Pros
    • Network models can provide accurate representation of corridor
    • Can evaluate “what if” scenarios
    • Scores can assist in diversion route evaluation and selection
  • Cons
    • High cost to setup and maintain
    • Inflexible to changes in roadway infrastructure (recalibration)
    • Requires a lot of data

​

8

9 of 16

Machine Learning-based Decision Support

  • Enable processing, manipulation, and visualization of large amounts of data
  • Derive business insights that support decision-making
  • Business intelligence tools and machine learning technologies utilized to evaluate responses and select the “best solution”
  • Can enhance traditional traffic management devices
    • VSL
    • LCS
    • CCTV (computer vision)

9

10 of 16

Machine Learning-based Decision Support (cont.)

  • Pros
    • Lower cost compared to traditional modeling/simulation
    • Continual learning
  • Cons
    • Data is crucial
      • Quality & quantity
    • Unproven

​

​

10

11 of 16

The common denominator: DATA

  • ICM requires a lot of data, AI needs more
    • Traffic signals & other devices
    • Traffic conditions (speed, volumes)
    • Unified roadway segment network
  • Requires integration of multiple data sources and formats
  • Low quality data negatively affects the output
  • Large backlog of historical data may be needed

​

​

​

11

12 of 16

The common denominator: DATA

12

13 of 16

The Future of AI in ICM & DSS

  • Increased automation (Less operator review)
  • Increased device support and analysis
    • Lane Control Signals
    • Variable Speed Limits
    • Parking
    • Ramp meters
    • Connected / automated vehicles
    • CCTV Cameras
    • Traffic Signals

​

13

14 of 16

The Future of AI in ICM & DSS (cont.)

  • Dynamically generated response plans and diversion routes
  • Route/Mode choice options
  • Incentivization
  • Load Balancing
  • Enhanced Modelling / Simulation
  • Big Data Inclusion / Analysis
  • Predictive ICM / Change before the pain

14

15 of 16

Summary

  • Artificial Intelligence technologies are growing fast in the traffic management space
  • AI can be a valuable asset for ICM and DSS
    • Modeling & Simulation
    • Data-driven active traffic management backed by ML
  • These solutions require data, preparation, funding, and more research
  • Data is key
    • Is the system properly instrumented to provide the right data?
    • Does it meet the operational needs of the system?
    • Can disparate systems be correlated to combine data?

​

​

15

16 of 16

Questions?

16

Samuel Burnett

210-522-3586

samuel.burnett@swri.org

AJ Skillern

210-522-6207

ansley.skillern@swri.org