Researchers at the University of Toronto (U of T) in Canada worked with colleagues in Uber’s Advanced Technologies Group on an algorithm that combines object detection, tracking over time, and prediction to improve safe maneuver planning by driverless vehicles.
“We try to optimize as a whole so we can correct mistakes between each of the tasks themselves,” notes U of T’s Wenjie Luo. “When done jointly, uncertainty can be propagated and computation shared.”
Uber first culled a dataset of several North American cities via LiDAR scanners that captured more than 1 million frames. The collaborators then developed a special artificial intelligence tool to cope with large volumes of scattered, unstructured data.
Up to 10 times faster than current methods, the algorithm uses a sparse computation based on which regions are most important.
“The car sees everything, but it focuses most of its computation on what’s important, saving computation,” says U of T’s Raquel Urtasun (who also is head of the Uber Advanced Technology Group Toronto).
From U of T News
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA