They’re called “ghost cars,” and anyone watching them race will know why.
You can’t see or hear them — but to the real cars swerving around the track, weaving among them, those ghost cars are a constant threat.
Those invisible cars are part of new research at Clemson University that aims to create algorithms to help autonomous, wirelessly connected vehicles anticipate the behavior of other vehicles to reduce braking.
While that goal may sound oddly specific, Clemson mechanical engineering professor Ardalan Vahidi said reducing the number of times self-driving vehicles brake in traffic is a huge factor is reducing energy. The less a vehicle brakes, the less energy it wastes through heat — and as a result, the more energy-efficient it becomes.
“The big picture is that we’ll have more opportunities to save energy when autonomous cars that are connected to the internet and other wireless networks start talking to one another,” Vahidi said.
That’s where the ghost cars come in.
The team of researchers recently tested its algorithm by sending two separate autonomous cars — one gas-powered Mazda and one electric Nissan — out on a closed track in southern Greenville County. Only one car ran the track at a time — or at least, only one car that was visible to those watching.
But to the car itself, the track was a congested roadway of heavy rush hour traffic.
By using computer simulations, researchers were able to create “ghost” vehicles in front of and behind the Mazda and Nissan, making the cars think they were navigating the kind of traffic that might, say, be present on Woodruff Road as the after work rush heads home. This allowed the team to be more aggressive and try difficult scenarios, with the bonus that any “collision” would only be theoretical; the car would stop and think it had just hit another vehicle, but in reality it was unscathed. Some of the ghost cars were designed to be autonomous, while others were driven by computer-simulated human drivers, giving the vehicles a chance to deal with both scenarios at once — a possible reality on the roads in the coming years.
Through the use of ghost car testing, the team found its algorithms resulted in energy savings up to 23%.
“There are a lot of groups focusing on autonomous vehicles, but the focus on how they can be energy efficient is not as mainstream,” Vahidi said. “That’s our niche.”
One major bonus the team uncovered was the impact of reduced braking on traffic flow in general. Braking often caused “phantom traffic jams,” as the team observed, but having the test cars anticipate what the ghost vehicles were going to do smoothed out traffic flow significantly, an indication that stop-and-start traffic may be something future generations only read about in the history books.
Vahidi and his team are now writing a paper that will detail more of the results, bringing an end to three years of research funded with $1.16 million from the Department of Energy.