Algorithm prepares self-driving cars for winter driving

Unlike most training datasets, the algorithm is based on scans of icy, snow-covered Canadian roads

Spotted: A collaboration between the University of Toronto, the University of Waterloo and Scale AI has released data that will help train future self-driving cars to handle the challenges of winter driving. The dataset acts as a virtual training course for the computer algorithms that enable cars to drive themselves.

Most training datasets are collected on sunny, summer days. Taking algorithms trained on those datasets and trying to use them in adverse conditions can be dangerous, as changes in sensor data caused by snowfall can lead to misclassifying objects (such as pedestrians and other vehicles), or missing them entirely. However, “The Canadian Adverse Driving Conditions dataset” is based on actual scans of icy, snow-covered Canadian roads. 

“Data is a critical bottleneck in current machine learning research,” said Alexandr Wang, founder and CEO of Scale. “Without reliable, high-quality data that captures the reality of driving in winter, it simply won’t be possible to build self-driving systems that work safely in these environments.”

The dataset was created with “Autonomoose,” a Lincoln MKZ hybrid equipped with eight onboard cameras, light detection and a GPS tracker. The Autonomoose is able to capture data at a rate of 10 images or scans per second. Over the past two winters, the teams have taken the Autonomoose around a snowy Ontario, recording data from more than 1,000 kilometres of driving.

Explore More: Mobility & Transport Innovations