VoxelNet could inform AI engineering of driverless vehicles.
Scientists from Apple have published a report on driverless car technology that describes an object-detection system called VoxelNet, which the authors claim “outperforms” the similar technology LiDAR.
Authors Yin Zhou, an AI researcher and Oncel Tuzel, a machine learning scientist, suggest “many” applications for their network, including “autonomous navigation, housekeeping robots, and augmented/virtual reality”. Zhou and Tuzel classify their work as “a major breakthrough”.
VoxelNet entails “point cloud based 3D object detection” technology. The scientific report, published on Cornell University’s website on November 17, details “a generic 3D detection network that unifies feature extraction and bounding box prediction into a single stage, end-to-end trainable deep network.”
The scientists claim VoxelNet is an improvement on existing technology as it removes the need for feature engineering. The Apple researchers report “encouraging results” from testing its network which “learns an effective discriminative representation of objects with various geometries.”
The network supposedly surmounts the “information bottleneck” of inferior LiDAR point clouds, which the authors term “sparse” with “highly variable point density due to factors such as non-uniform sampling of the 3D space.”
VoxelNet has achieved a “major breakthrough in recognition and detection tasks on images”, according to its authors, owing to its machine-learning features. It is hoped the advanced 3D sensor technology could inform AI engineering of driverless vehicles, such as the truck released by Tesla last week. Elon Musk, owner of the Silicon Valley firm, has predicted autonomous automobiles will become the norm within two years.
When contacted for comment on the research Apple declined to comment.
Apple launched its Machine Learning journal in July, publishing one article a month “written by Apple engineers about their work using machine learning technologies”. This blog has included posts on face-detection and audio tech recognition. However, the tech giant has never published any research relating to driverless technology.