AutoMatrix: A large-scale traffic congestion modeling tool to investigate Anytime Algorithms for Multi-core Computing Architectures
|University:||University of Pennsylvania|
|Principal Investigator:||Rahul Mangharam|
|PI Contact Information:||200 S. 33rd Street, 203 Moore Building, University of Pennsylvania, Philadelphia, PA 19104.|
|Project Manager:||Courtney Ehrlichman|
|Funding Source(s) and Amounts Provided (by each agency or organization):||UTC|
|Agency ID/Contract/Grant Number:||DTRT12GUTC11|
|Start and End Dates:||1/1/12 – 12/31/13|
|Subject Categories:||Algorithms for Multi-core Computing Architectures|
|Abstract:||We are designing AutoMatrix, a traffic congestion simulation platform for large-scale traffic modeling, routing and congestion management. AutoMatrix is currently capable of simulating over 16 million vehicles on any US street map and executing traffic estimation, prediction and route assignment algorithms with high-throughput. This research has the potential to extend real-time scheduling on massively parallel Graphics Processor Unit (GPU) architectures to attack a variety of data-driven, interactive and dynamical algorithms with timely operation.
In collaboration with the Delaware Valley Regional Planning Commission (DVRPC), we will also focus on simulating large traffic networks for disaster response and evacuation. The goal is to evaluate evacuation approaches from traffic signal adaptation and with support of information dissemination using vehicle-to-vehicle networking. AutoMatrix now includes both simulation-based and optimization algorithms for vehicle traffic evacuation on real street maps. The software will be open source for the transportation community to use freely. The PI is also on the Organizing Committee for the Delaware Valley Regional Planning Commission for Disaster Response and Evacuation. The experience from this will contribute a real-world perspective on traffic evacuation to the project.
|Describe Implementation of Research Outcomes (or why not implemented):|
|Impacts/Benefits of Implementation (actual, not anticipated):||Delays due to traffic congestion cost Americans $78 billion in the form of 4.2 billion lost hours and 2.9 billion gallons of wasted fuel, and 35–55% of these delays are caused by point-based traffic incidents rather than recurring congestion. As the density of vehicles increases, there is a need for large-scale traffic congestion management such that real-time “eco-routing” can be provided to prevent, avoid, and alleviate traffic back-ups. Models and tools for nationwide traffic congestion management, with networked streaming vehicle data, are required to compute the fastest and most eco-friendly routes without new infrastructure costs.
The design of the AutoMatrix platform will enable the scaling of traffic network operations to handle data processing for millions of vehicles, estimate and predict congestion, and facilitate route assignment as well as to model traffic operations and disaster response during congestion.
In the current year, we are extending the traffic modeling to semi-autonomous driving systems. Anytime Algorithms for Autonomous Vehicles. In semi-autonomous vehicles, algorithms for trajectory control, obstacle avoidance and path planning/navigation are very compute-intensive and require a lot of processing. These algorithms must run in an on-line and real-time manner within the closed-loop context of the moving vehicle. Currently, the computer vision processing bottleneck for sensor data capture (by cameras, position sensors and laser range finders) is the bottleneck and restricts fast vehicle responsiveness and faster velocities. This project if focused on the development of approximate and imprecise computation algorithms that take the large amount of data generated by these sensors and provide the best possible answer within the deadline, so the vehicle is always safe and responsive. We are developing this architecture to run on graphics processors (GPUs) and will demonstrate the safety and efficacy on both modeled vehicles and full-scale vehicles. This project has large impact in making low-cost sensing more viable by ensuring the processing is more effective and appropriate for the situation the vehicle is in.