Michigan Testing Shows Fairly Innocuous Weather Baffles Self-driving Car Systems

Steph Willems
by Steph Willems

Successfully operating self-driving cars on crowded, complex roadways in sunny, dry locales like Phoenix, Arizona is already enough of a challenge, but researchers in the cold, tempestuous climes of Michigan have revealed what the latest and greatest autonomous technology is really up against.

Rain, sometimes hard rain. But also light rain. Also: cold temperatures, and trees with leaves that fall off in the winter. Given that so few places in the world boast such extreme weather and vegetation anomalies as Michigan, this won’t pose a problem for the widespread proliferation of driverless cars, will it?

While adverse climatic conditions can wreak havoc on cars, decades of testing and development have yielded batteries that crank harder, wipers that turn on at the first touch of rain, and an endless list of other stress-relieving advances. For driverless cars, it’s still the Wild West. Early days.

The proximity of Detroit Three HQs and development centers means Michigan State University finds itself at the center of things on this new frontier. There’s already a driverless car test facility, and school created the CANVAS program (Connected and Autonomous Networked Vehicles for Active Safety) to study how well the driving systems handle the tasks of avoiding things they shouldn’t hit. Part of that research included analyzing how weather impacts the sensor arrays and cameras used to guide the vehicles.

While MSU’s findings will soon appear in a study, researchers gave the skinny to Automotive News. Basically, autonomous car systems need a lot more work before companies can be assured of reliability outside the temperate Southwest.

The worst culprit when it comes to pedestrian and cyclist safety and driverless cars is rain. Raindrops confuse the algorithms used to detect such objects, claims Hayder Radha, the MSU professor of electrical and computer engineering who oversaw the study. As radar and lidar can’t do all the seeing, cameras are also used to guide a vehicle’s path. What the camera sees must then be interpreted.

“When we run these algorithms, we see very noticeable, tangible degradation in detection,” Radha said. “Even low-intensity rain can really create some serious problems, and as you increase the intensity, the performance of what we consider state-of-the-art mechanisms can almost become paralyzed.”

The presence of rain has stymied testing of production automobiles fitted with the latest driver-assist systems in the past. Camera-based systems used for lane-keeping and other functions are susceptible to a rain-caused algorithm confusion. To gauge the impairment, MSU tested a variety of conditions. Raindrop size varied, as did the concentration of raindrops. Wind was factored in, too.

From Automotive News:

Using a scale that ran from clear weather to a blinding rainstorm, they found algorithms failed to detect as many as 20 percent of objects when the rain intensity was 10 percent of the worst-case scenario. When rain intensity increased to 30 percent, as many as 40 percent of objects could no longer be detected.

Other problems cropped up during the changing of the seasons. High-resolution mapping used to help guide self-driving cars becomes less useful when the landscape changes in winter. The removal of foliage means landscape simply doesn’t look the same.

“You can imagine in environments where there are a lot of leaves on trees or on shrubs close to the road, they are an essential part of the map,” Radha said. “So summer and winter are completely different. When they fall down in winter, you have nothing to work with.”

Cold temperatures also wreaked havoc on the high-tech systems, with temperatures of 10F and below bringing about a marked uptick in the amount of “noise” returned by lidar sensors. To accurately map out the landscape in front of the car, lidar can’t send back hazy returns or false positives. Compensating for this could lead to a dumbed-down system that’s less safe and responsive than intended. Uber learned this lesson in Arizona in March.

Radha said suppliers are attempting to develop sensors that operate effectively in cold conditions, knowing that driverless cars won’t be able to catch on until weather no longer poses a challenge. As for snow, well, we’d be very interested to know how companies plan to contend with that challenge.

[Images: Ford]

Steph Willems
Steph Willems

More by Steph Willems

Comments
Join the conversation
2 of 46 comments
  • Krhodes1 Krhodes1 on Dec 01, 2018

    Ultimately, I as I have said on here many times, I am of two minds when it comes to autonomous vehicles. On the one hand, humans SUUUCK at driving. It's really a minor miracle ONLY 30K people die a year in the US in motor vehicle accidents. On the other hand, I am in IT and I know full well that computers suck too - they just suck in different ways. So will we end up simply trading one set of dilemmas for another set of dilemmas? We COULD, if we had the will, make people better drivers... I am not so sure we can make computers suck less. :-) And for sure, leaving the computers out of the picture is a WHOLE lot cheaper. But that said, I would be perfectly happy with some limited autonomy that can handle just boring Interstate highway driving in good weather. I feel like THAT should be initial focus, not the "do-everything" autonomous car that really seems like a fool's errand at this point. It shouldn't take THAT much additional infrastructure to allow my car to run up and down I-75 (perhaps in a dedicated lane) while I take a nap - I will drive to and from the highway. I do think that the halfway solutions like Tesla's Autopilot are just stupid. They work just well enough to get you killed when they suddenly don't.

  • Robbie Robbie on Dec 03, 2018

    Too much rain: car parks itself in a safe spot, and tells owner it has to wait until the rain subsides. How big a problem can this be?

Next