Ruslan Salakhutdinov, Apple’s director of artificial intelligence research, shared some new details about the company’s AI developments that all seemed to revolve around self-driving cars, Wired reports. During a lunch talk at tech conference, Salakhutdinov discussed how the company is creating detailed 3D maps, teaching vehicles to navigate in unfamiliar areas and using camera and sensor data to spot cars, pedestrians and other obstacles on urban streets. The system was even able to pick out pedestrians when they were partially obscured by cars, which Salakhutdinov credited to recent leaps ahead in machine learning. “If you asked me five years ago, I would be very skeptical of saying ‘Yes, you could do that,’” he said.
The overall autonomous car project encompasses smaller projects with all kinds of objectives, some of which provide detailed 3D maps with features like traffic signals and road markings to give the vehicle as much information as possible upfront, and others that give the software moving through the world its own sense of direction and give the vehicle the ability to make decisions in “dynamic situations,” when all the planning and mapping can’t account for something like a pedestrian entering the road at the wrong time. Salakhutdinov also revealed images to demonstrate how the car’s camera system worked even when raindrops landed on the lens.