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