Open-road Testbed Explores Effect of Automated Vehicles on Traffic
Phantom traffic jams can be a major pain. These slowdowns are not caused by accidents, construction, or other impediments — rather, they are simply caused by the interactions of a large volume of cars and drivers on the road. Previous experiments have found that automated vehicles are able to reduce congestion in these phantom jams. In one trial, a single automated vehicle was able to smooth a wave of congestion in a ring of 20 cars that were traveling in a circle. However, a similar penetration of 4.8 percent automated vehicles into the roadway system is essentially impossible to implement on highways at present.
The I-24 Mobility Technology Interstate Observation Network (MOTION) open-road testbed—led by the Tennessee Department of Transportation on Interstate 24 near Nashville, Tenn.—provides a way to measure the impact of automated vehicles on real highway traffic. During a minisymposium presentation at the 2023 SIAM Conference on Applications of Dynamical Systems, which is taking place this week in Portland, Ore., Sean McQuade (Rutgers University) gave a talk on behalf of Daniel Work (Vanderbilt University) on the implementation of the I-24 MOTION testbed. McQuade is the leader of the Scenarios Team for the collaboration, and the presentation’s additional coauthors were William Barbour (Vanderbilt University) and Benedetto Piccoli (Rutgers University).
Construction of the I-24 MOTION testbed involved erecting 40 poles with heights ranging from 110 to 135 feet, which were outfitted with around 300 4k cameras that record much higher-quality videos than the Tennessee Department of Transportation’s existing cameras. The cameras cover over four miles of highway, with higher coverage near on and off ramps. “We need to be able to monitor vehicles to see the effect on the traffic around it,” McQuade said.
There have been some prior projects to create large sets of traffic data, like the well-known 2006 Next Generation SIMulation program in California and the 2018 Highway Drone dataset from Germany that logged 25,000 vehicle miles. But the I-24 MOTION dataset is a step up: it records 200,000,000 vehicle miles annually, and will continue to add more data year after year. “What’s being designed now is an application programming interface to look at the data,” McQuade said. “This is a huge amount of data, and users might just want to look at one thing.”
The I-24 testbed was completed in November 2022, after facing various construction hurdles and delays to ensure the safety of the project. The team also elected to delay its launch until after the end of daylight saving time to ensure daylight for recording during peak hours. Now, the cameras produce about 40 terabytes of video data every single day, which is processed in real time. A trajectory processing algorithm generates high-precision three-dimensional bounding boxes for each vehicle in each video frame from each camera; these trajectories then undergo a coordinate transformation and are stitched together. The I-24 MOTION team has a centralized compute cluster to process the video streams and convert them into trajectory data, which the group intends to freely share with registered users across the world.
McQuade then proceeded to discussing an experiment with connected and automated vehicles (CAVs) that the Congestion Impacts Reduction via CAV-in-the-loop Lagrangian Energy Smoothing (CIRCLES) consortium recently organized on the I-24 testbed. This large-scale experiment used 100 Nissan vehicles with adaptive cruise controls that the team enhanced themselves, designed specifically to dampen phantom traffic waves (see Figure 1). “Can we even do this with the simplest highway?” McQuade asked. “That is the first question, and that’s what we were out to answer.”
A Raspberry Pi plugged into the central console of each car overrode their internal cruise control systems and recorded positional and other data. The cruise control system was able to sense and respond to the cars in front, but was not as adept at sensing objects in the rear. As such, the I-24 MOTION cameras observed how the CAVs manipulated the pockets of space around the car.
On the days of the test, a fleet of 100 control vehicles took laps along the I-24 testbed with the goal of smoothing traffic. They were driving amongst an average of 7,000 cars on the road per hour and took approximately 32-minute laps. “Knowing whether or not we could do this was a big ‘if’ with our penetration rate,” McQuade said.
The team first considered a route for the control cars that involved making two left turns at traffic lights that other cars on the road used infrequently, but this was ultimately deemed too disruptive. Instead, they decided to break the route into two segments that partially overlapped — half of the cars would drive on one segment, and the other half would drive on the other.
Ultimately, this effort was able to produce a time-space diagram of the propagating traffic waves with the presence of automated cars. The project was made possible through the support of the Department of Energy and National Science Foundation, and involved collaboration and support from the Nashville Department of Transportation, Tennessee Department of Transportation, Nissan, GM, Toyota, and C3 Ai, as well as numerous universities and other organizations. In particular, the CIRCLES consortium partnered with Metro Nashville for their field headquarters (see Figure 2). “People would walk out, get in their cars, and do some laps,” McQuade said. “We trained 150 drivers for a total of 100 cars. The training of the drivers was nontrivial because we really wanted to emphasize safety… Tennessee can be a bit grueling in terms of traffic.”
About the Author
Jillian Kunze
Associate editor, SIAM News
Jillian Kunze is the associate editor of SIAM News.