Just a Few Self-Driving Cars Could Fix Phantom Traffic Jams

A new study suggests they can help get rid of stop-and-go traffic on highways.

Autonomous car test
One autonomous car in this group was able to reduce stop-and-go traffic flow. John de Dios & Alan Davis

Chances are that if you’re driving any distance over the Memorial Day weekend, you will be confronted with one of the great mysteries of modern travel. It’s the phantom traffic jam, when the flow of cars on the highway alternately speeds up and then slows to a stop, and you find yourself cursing your fate and the brake lights in front of you.

But now there may be answer for how to rid the highways of this phenomenon: Mix in cars without human drivers.

That’s what a team of scientists is suggesting after two days of testing in Arizona. Based on computer simulations, they believed that adding just one autonomous vehicle to a pack of cars might be just the thing to reduce the notorious stop-and-go traffic waves. Having one driverless car that was able to maintain a consistent speed could be enough to help induce human drivers to follow suit. Or so their data indicated.

“We had a good idea from our simulations that a single car would be sufficient,” said Daniel Work, an assistant professor at the University of Illinois, Urbana-Champaign. “But the proof is actually doing it with real human drivers, because quite frankly, that’s the hardest thing to model.”

So, the team, which also included lead researchers from Temple, Rutgers and the University of Arizona, recruited human drivers to follow each other around a circular track.  Roughly 20 cars were involved—it varied from 19 to 22 throughout the testing—and that included one that a person steered, but otherwise operated autonomously.

While the humans knew that a self-driving model would be in their pack, they didn’t know why, according to Work. They were simply given these instructions: “Drive as if you were in rush hour traffic. Follow the vehicle ahead without falling behind. Do not pass the car ahead. Do not hit the car ahead. Drive safely at all times. Do not tailgate. But put an emphasis on catching up to the vehicle ahead if a gap starts opening up.”

To make that easier, vehicle speed was kept low, about 15 miles per hour. Still, the tendency of the humans was to speed up and slow down. By contrast, the driverless car, equipped with a laser scanner that enabled it to track both the speed of the car in front of it and the distance to its back bumper, was unfailingly consistent. And, ultimately, as the human drivers adjusted to that even pace, the stop and go flow diminished and, in some tests, was actually eliminated.   

“The driverless car not only can choose a speed to dampen the wave, it has the discipline to actually stick to it,” Work said.  

Here’s a video of one of the tests. At the beginning and end of the video, the car's acceleration is controlled by a human driver, and traffic starts to bunch up, but when the computer takes over, traffic smooths out:

Self-driving cars experiment demonstrates dramatic improvements in traffic flow

There was another benefit, one that Work admits pleasantly surprised him.  By adjusting to a more constant pace, the pack’s fuel consumption dropped by as much as 40 percent.  “When I first saw the numbers, I thought this can’t be,” he said. “It’s almost too good to be true.”

As positive as the results were for the study, which received funding from the National Science Foundation, Work knows there’s a big difference between what happens in controlled conditions on a track and what happens on the highway.

For starters, the cars in the tests were restricted to one lane, meaning there was none of the lane-hopping or merging that can tangle traffic flow. Adding those variables to traffic tests would significantly ratchet up their complexity and cost. But based on what the researchers learned, Work feels more confident in concluding that even if only five percent of vehicles on the highway were autonomous, it would still improve both traffic flow and fuel consumption.

“What motivated this study was how much advancement there’s been in the self-driving landscape, and the questions about what are we going to be able to do in the near future with this technology,” he said. “There’s been a lot of research about what happens when the entire fleet of vehicles is automated, but depending on who you listen to, that may take from a few years to a few decades.

“That’s great, but we wanted to see if there are benefits in the immediate future when you just have a few of those cars on the roadways. And what we found was that just one self-driving car made all the human drivers more efficient, too.” 

Get the latest stories in your inbox every weekday.