When Snowflakes Fall, Are Self-Driving Cars on Slippery Ground?

After January’s snow and ice storm pounded Maryland, Yaobang Gong popped his Tesla Model Y into self-driving mode, backed the new car out of his driveway, and immediately got stuck in a snow mound. 

Most Tesla drivers likely would have shrugged off the hiccup, but Gong is a University of Maryland assistant research professor of civil and environmental engineering who studies traffic safety. After shifting into manual drive and spinning his tire out of the slush, he wondered how his car would have coped when the stakes were higher. 

“If I were on the highway, it would be totally different,” he said.

Gong investigates these issues as part of a UMD team analyzing self-driving car performance and risk in wintry conditions. Led by Associate Professor of civil and environmental engineering Terry Yang and funded by the National Science Foundation, the team has built an animated simulator that considers varying weather conditions, road bends and autonomous vehicle algorithms to determine which combinations could cause a self-driving car to misjudge its ability to safely maneuver. 

In running the simulator for the past year, the team observed many collisions, said Yang. In a worst-case scenario, with slick, curvy mountain roads, he says, “we’ve seen the vehicle slide and fall off a cliff.”

Like Gong, Yang’s everyday driving has shaped his work. His Acura isn’t self-piloting but does feature an automatic braking system. During one snowy commute without any cars ahead, he said, the system needlessly activated, jerking Yang forward.

Yang said self-driving sensors aren’t programed to reliably detect snow and ice, especially black ice, and that thickly falling snowflakes can confuse the system on occasion, raising the risk of rear-end collisions. His team seeks to help car manufacturers mitigate risk by targeting the rarest of scenarios.

”Full self-driving mode works very well 99.99% of the time, but what about that 0.01%?” he said. 

Companies have taken differing positions about autonomous vehicle safety on snow and ice. The self-driving taxi company Waymo, which manages several fleets in the Sun Belt and San Francisco Bay Area, has recently begun testing its technologies in northern cities with snowy conditions but has yet to validate them. Tesla, meanwhile, maintains that its vehicles are safe in the winter. 

To establish baseline driving behavior on a typical highway in the simulator, Yang’s team flew drones above a snowy stretch of Interstate 695 in Baltimore County, which captured, in one-second intervals, cars braking, accelerating, changing lanes and maintaining distance.

The researchers introduced varying, manipulatable friction levels to mimic different wintry mixes. They trained the computer to capture the exact moments of skids or spins through established literature on pavement texture and tire treads. Finally, they added an algorithm to simulate a self-driving car.

The team hopes to conclude data collection and publish their results soon, aiming to create a report for car manufacturers with data that could help them adjust their algorithms for greater safety.  

In the meantime, Yang’s lab, the Maryland Transportation Research & Artificial Intelligence Laboratory, plans to acquire its own Tesla later this year and begin testing it in a campus parking lot. Currently the team uses a self-driving minivan with lidar sensors and hybrid batteries to power onboard computers. Occasionally it tests the van in rain, but not snow. “We’re definitely not going to crash our car,” Yang said.

Since backing his Tesla into the snow pile earlier this winter, Gong has observed his car rolling over mounds of snow in self-drive mode several times. “Thankfully in all cases I made it safely back,” he said.

Until spring is in full swing, however, he’ll think twice before shifting into self-driving mode, he said.