Apple's work on self-driving cars has been more secretive than just about every other project in the autonomous car space — but now, two of the chartph.company's scientists have published some of their auto-focused research for the first time.?
The paper, authored by Apple engineers Yin Zhou and Oncel Tuzel and published in the independent journal arXiv, details a new chartph.computer imaging software technique called "VoxelNet" that could improve a driverless car system's ability to detect pedestrians and cyclists.
The scientists claim their new method could be even more effective than the two-tiered LiDAR and camera systems that have bechartph.come the industry standard for object detection in self-driving cars. Those expensive systems depend on cameras to help determine the small or faraway objects (like pedestrians or cyclists) detected by LiDAR sensors, which use light beams to detect and map 3D obstacles in the world around the the vehicle.?
The VoxelNet system — which was named after the "voxel" unit of value for a point in a three-dimensional grid — eliminates the need for a camera to help identify the objects detected by LiDAR sensors, allowing the autonomous platform to work on LiDAR alone. The scientists tested the software using models that showed pedestrians, cyclists, and other faraway objects.?