Online Competitive Algorithms and Reinforcement Learning for Traffic Management
The management of traffic at intersections, both regulated and unregulated, has significant impact on delays experienced. Traffic signal cycle times in urban centers are generally quite long leading to inefficiencies in traffic flow and increased environmental impact. Traffic flow coordination relies on partially outdated technology, which is rigid and not based in contemporary algorithmic models. As an example, a number of papers show that the well-known Webster formula, now almost 70 years old and widely used to estimate the minimum delay optimal traffic signal cycle length, indeed overestimates the cycle length for high degrees of traffic saturation – a degree of saturation now common in most urban areas. In the Las Vegas Valley the RTC follows an approach based in traditional transportation science, which tries to synchronize lights along corridors and which use long cycle length, i.e. when a car encounters a red light the wait can substantial. The study team proposes to use computer science approaches, namely reinforcement learning and online competitive analysis to substantially improve the state of the art. In fact, as multi-modal traffic systems, ride-share systems and autonomous vehicles are becoming more prevalent vehicle traffic becomes more of a distributed system resembling internet traffic. With the use of deep learning techniques a system is envisioned that can analyze the large trove of data now available. The system will mine for systemic inefficiencies, and then give algorithmic solutions to eliminate such inefficiencies. When vehicles accumulate at an intersection this sequence of vehicles forms a platoon. Vehicles in a platoon all experience the same stopped delay and are subject to the deceleration and acceleration delay at that intersection. The situation is similar to an area of research called “batch scheduling” and the proposer proposes to study the problem of delays in the framework of batching. Batching problems, both offline as well of online have been studied extensively (including by the proposer, see yet, to the proposers’ knowledge a connection has not been made in this area.
- Record URL:
Language
- English
Project
- Status: Programmed
- Funding: $51871
-
Contract Numbers:
69A3552348309
-
Sponsor Organizations:
Office of the Assistant Secretary for Research and Technology
University Transportation Centers Program
Department of Transportation
Washington, DC United States 20590 -
Managing Organizations:
METRANS Transportation Consortium
University of Southern California
Los Angeles, CA United States -
Project Managers:
Hong, Jennifer
Bruner, Britain
- Start Date: 20240815
- Expected Completion Date: 20250814
- Actual Completion Date: 0
- USDOT Program: University Transportation Centers
Subject/Index Terms
- TRT Terms: Machine learning; Signalized intersections; Traffic control; Traffic signal control systems
- Geographic Terms: Las Vegas Metropolitan Area
- Subject Areas: Data and Information Technology; Highways; Operations and Traffic Management;
Filing Info
- Accession Number: 01928983
- Record Type: Research project
- Source Agency: Pacific Southwest Region University Transportation Center
- Contract Numbers: 69A3552348309
- Files: UTC, RIP
- Created Date: Aug 27 2024 5:42PM