Most self-driving car companies use lidars, devices that emit laser beams to create 3D maps of the car's surroundings, to account for potential image-object detection mistakes made by deep neural networks.
Not Tesla, though. The company remains a vocal champion for a purely vision-based approach to autonomous driving. Tesla's Chief AI Scientist Andrej Karpathy explained why at this year's Computer Vision and Pattern Recognition (CVPR) conference.
"It's unscalable to collect, build, and maintain these high-definition lidar maps," he said. "It would be extremely difficult to keep this infrastructure up to date."
From TechTalks
View Full Article
No entries found