Computer scientists working for Apple have published a research paper that gives us a look at exactly what sort of tech the company is cooking up for the autonomous car realm. Apple had originally been working on an actual autonomous car, but reportedly hit the reset button on those plans to focus on software for self-driving cars. The research paper outlines a new method that could be used to allow self-driving cars to detect pedestrians and cyclists.
The paper, which was published November 17 in an online journal called arXiv, and authored by Tin Zhou and Oncel Tuzel, focuses on advances made in the capability of detecting 3D objects. Currently, autonomous cars use LiDAR systems to shine light onto a surface and measure how long that light takes to return to the sensor. The difference in the time it takes the light to return to the sensor allow the creation of an image of the shape of whatever is in front of the car.
The catch with current LiDAR systems is that it's hard to detect small objects that are further away without using an additional camera. The Apple scientists fear that this sort of LiDAR system could be difficult to deploy and could be more prone to sensor failure, which could be disastrous in an autonomous car. The scientists have a new system they have dubbed "VoxelNet" which is a LiDAR system that doesn’t need a camera.
The system uses computer vision and AI to carry out necessary sensing functions. So far, Apple has only tested the system in a computer simulation rather than a real car. However, the scientists say their VoxelNet "outperforms the state-of-the-art LiDAR-based 3-D detection methods by a large margin."
While Apple has pulled back on its original vision of building its own autonomous cars for sale, it has notified the NHTSA that it plans to develop autonomous research vehicles. Apple also has a permit for testing autonomous cars in California. Apple intends to focus on autonomous autos from a "core technology point of view."