How Waterloo researchers are giving self-driving cars 'Canadian vision
Engineers compiled thousands of snowy day images to tell self-driving cars how to navigate winter
A team of engineers at the University of Waterloo has made it its mission to Canadianize self-driving vehicles.
It's the first group in the world to create a data set for winter driving. In other words, it has compiled 10,000 video frames and 56,000 images to better teach self-driving vehicles how to navigate snow and ice.
"We're basically telling the world, 'Hey, we need to start addressing this because there are places that get bad winters,'" said Krzysztof Czarnecki, an electrical and computer engineering professor who leads UW's autonomous driving team.
The group of engineers has gathered the images from a laser and eight cameras fastened to its self-driving vehicle, which they fondly refer to the "autonomoose." They've spent the last couple years driving around Waterloo region capturing images of what roads look like under a layer of snow.
The team has released the information to the public so companies and other universities can create and innovate for all driving conditions. The only ask from the Canadian engineers is self-driving innovators share what they learn to keep the data in the public realm.
Buckle up for a bumpy ride
The autonomous vehicle industry is big business. Companies, such as Google and Uber, are spending billions of dollars a year on technology and prototypes.
The University of Waterloo has been ahead of the curve. It built its self-driving car in 2016 and by the summer of 2018, the Lincoln MKZ had logged 100 kilometres of public road driving.
But let there be no confusion: The autonmoose's winter trek is not a smooth ride.
As it loops around a track, the car slids and swerves, its brakes automatically press down and the steering wheel frantically spins to set itself back on course.
"One of the biggest challenges is for the car to actually make the decision whether this snow accumulation is something we can just drive through or is it something we should be avoiding," Czarnecki said.
The car learns by example, so over time — like any driver — it will improve.
The future is far
Don't expect to pick up your own all-season autonomoose anytime soon though.
Czarnecki emphasizes we're likely a decade away from commercial self-driving vehicles that are safe in all weather conditions, and widespread use would be even further down the line.
But the possibility is there. It's the ember that fuels the research and testing at the University of Waterloo.
Artificial intelligence would become the driver that doesn't get tired, distracted or abuse substances.
"Eventually the technology will be, I would say better than a human," Czarnecki said. "It has the potential."
ABOUT THE AUTHOR
