Sign In

Communications of the ACM

ACM TechNews

Research Data Puts Self-Driving Cars to the Ultimate Test: Canadian Winter


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
Driving in snowy conditions.

A new dataset will train future autonomous vehicles to drive in winter conditions.

Credit: Steven Waslander

Researchers at Canada's universities of Toronto (U of T) and Waterloo collaborated with San Francisco-based artificial intelligence (AI) infrastructure firm Scale AI to create a dataset for training future autonomous vehicles to drive in winter conditions.

The Canadian Adverse Driving Conditions dataset uses real-world scans of icy, snow-covered Canadian roads as a virtual training course for self-driving cars' algorithms.

U of T's Steven Waslander said most driving datasets are collected in summer, and self-driving algorithms trained on such data tend to be confounded in adverse conditions.

Waslander and Waterloo's Krzysztof Czarnecki compiled the new dataset over the past two winters, using a Lincoln MKZ hybrid equipped with cameras, a LiDAR scanner, and a global-positioning system tracker to record conditions across more than 1,000 kilometers (621 miles) of roads.

Scale AI labeled the data through computer and human image recognition, and further analysis and processing converted the data into a software-parsable format.

From U of T News
View Full Article

 

Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA


 

No entries found