Are autonomous vehicles ready for Canada?

November 9, 2021

For those of us who spend part of the year in winter driving conditions, autonomous vehicles may seem out of reach. It’s a sentiment shared by almost one-third of the Canadian public who are concerned about autonomous technology. How can you teach a car to drive under whiteout or black-ice conditions? However hard it may be, it’s a solvable problem. If we can teach robots to dance, surely we must be able to teach them to drive in winter. It’s not just a Canadian problem either, since around 70 percent of the US population lives in the snow belt. 

What are the worst problems autonomous vehicles face in winter and how can we solve them? 

Inability to see lanes 

The first problem autonomous winter vehicles must handle is snow-covered lane markings. Without the ability to sense the road edge or the divisions between lanes, autonomous cars will be off the road or veering into adjacent lanes in no time.  

One solution for this is to use centimetre-level accurate GPS, which can help the vehicle distinguish lane boundaries even when it’s unable to see them. The vehicle can then maintain proper distance from the road edge and other lanes with a combination of rapid and precise GPS measurements and dead reckoning to interpolate vehicle position between measurements. (We have three of these advanced GPS solutions at Area X.O available for testing purposes.) 

Precise GPS locations can also help autonomous vehicles keep lanes properly under light snow situations. However, that solution may not work optimally when there is enough snow that previous traffic has created ruts. When driving in heavy snow, a human driver naturally follows the ruts, even when they don’t line up with the lane centres. This helps maintain vehicle stability, preventing it from sliding out of control. For an autonomous car, this goes from being a GPS-precision problem to a computer-vision problem. There are two ways to solve this latter problem. 

If a computer-vision system is camera based, it must be able to see fine graduations in a bright white background; if it’s radar based, it must be able to recognize very short gullies that could otherwise look like sensor noise. Both could solve the problem, but another solution might be wheel angle sensors that detect resistance when trying to steer out of a rut. In any case, these conditions are difficult challenges that require more research. 

Inability to see environment 

Worse yet, heavy snow conditions can prevent lidar- and camera-based autonomous systems from seeing, period. Lidar systems bounce laser light off objects to measure their distance. Those measurements are confused if a laser beam hits a tumbling snowflake. Similar issues exist for cameras that can capture images oversaturated with bright white snowflakes and fail to properly detect object boundaries. 

Part of the problem may just be that we need more people from cold regions, who are highly motivated to solve snowy-autonomous driving, to work on the problem. Canadian climates are proving to make great testing grounds for companies wanting to test their tech against the worst that winter can throw at us. A Finnish self-driving car named Martti can drive on snow-bound roads by having trained filters in snow and having several redundant sensor systems. And MIT researchers are looking at ground-penetrating radar to see, even when people cannot. 

Coated sensors 

Canadian drivers know that if you’re driving in the snow, you better be sure your windshield fluid is topped up – windshields can ice up, build up snow, get too streaky to see, or become coated in salt and dirt. This shows how important keeping sensors clean can be: sight is a person’s main “sensor,” and we certainly operate at lower capability when it’s difficult to see through the windshield. 

However, for an autonomous car, the problem is multiplied – the car may rely on multiple cameras, lidar, radar, and ultrasonic sensors that can get crusted over in ice, slush, and dirt. If those sensors are covered with frozen slush or grime, they won’t work. (This is a problem with today’s adaptive cruise systems too.) 

To solve this issue, researchers are working on different ice-phobic coatings – a layer over the sensor that repels ice, prevents it from sticking, or deforms slightly to shatter any stuck ice. Right now, according to Ontario Tech University’s Automotive Centre of Excellence, these coatings aren’t yet optimized to handle every situation. Many other solutions for general dirt may also work for winter conditions, including fluid sprayers, ultrasonic surfaces, air dams, and sensors behind windshields. 

Winter obstacles 

How do autonomous cars recognize common winter objects like snow drifts, snow flying off other cars, or snowy ridges left behind by snowplows? In this case, the answer is easy – better training of self-driving models with winter data. 

Researchers at the University of Waterloo and the University of Toronto teamed up to solve this problem. They created the Canadian Adverse Driving Conditions Dataset (CADC), an open-source data repository that covers wintery conditions, providing many kilometres of sensor data, videos, pictures, and annotations that capture the operation of an autonomous vehicle in a typical Canadian winter. With this data set, companies building autonomous vehicles have a ready-made way to train and test their models against real-life winter conditions.  

Staying home 

Even humans reach a point when we feel winter conditions aren’t safe for us to drive. If a human can’t drive the roads, should we expect cars to be able to manage them? Although this is perhaps a last resort, an autonomous vehicle may have to be programmed with the ability to recognize that conditions are beyond its capability to drive the vehicle safely. This “sanity switch” may be needed in cases where snow is too deep, temperatures are too cold, or blinding blowing snow could endanger the occupants should the vehicle become stuck or slide off the road. 

Let it snow 

It’s clear that fully autonomous systems will be available in sunny climates far sooner than they’ll be available in Canada or the Northern US. However, that isn’t stopping researchers from developing and testing solutions to Old Man Winter’s challenge at Area X.O with our 1,860-acre facility covered in snow, sleet and ice for almost half the year. Testing in harsh weather conditions is important to the future of autonomous vehicles in Canada – and everywhere else in the world that experiences snow. Learn more about Area X.O’s four-season capabilities and be part of the autonomous conversation by attending our upcoming CAV Canada 2021 event – register today. 

CAV Canada 2021 is proudly hosted by Area X.O, Invest Ottawa, and KNBA.

Connect with us on social to stay up-to-date on real-time announcements!
Facebook | LinkedIn | Twitter

Share this: