At a private workshop held last week for AI researchers, Apple gave a rare glimpse of some machine learning technology it is building for self-driving cars, reports MIT Technology Review. Speaking to researchers in the field, Ruslan Salakhutdinov, Apple’s director of AI, discussed several projects related to automated driving.
The talks were given during the Neural Information Processing Systems conference, the largest AI-focused academic event of the year, which was held in Long Beach, California. The event attracted thousands of researchers including many from rival tech companies. The talks were designed to showcase Apple’s technical prowess and to woo potential recruits.
Salakdinov, who joined Apple in 2016 but who still holds a post as a professor at Carnegie Mellon University, showed off a project previously disclosed in a paper posted online by Apple last month. This project trained a system to recognise pedestrians and other vehicles from 3-D point cloud information.
Other projects not previously revealed included a method for classifying different objects on the road using cameras placed on top of a vehicle, and a method of using camera footage to track its position precisely. This technique, known as visual SLAM (simultaneous location and mapping) could be used for autonomous driving but also augmented and virtual reality.