Yin Zhou and Oncel Tuzel submitted the paper to arXiv, the independent online journal, on November 17. It is difficult for the company to attract the product due to its future product confidentiality plan Artificial intelligence and machine learning researchers, so this paper can be described as significant.
Researchers have come up with a new software method called VoxeNet that helps computers detect 3D objects.
Apple declined to comment.
Academics are accustomed to sharing their research with their counterparts in other organizations, and in July this year Apple also launched "Apple Machine Learning Journals" for its researchers, and their results are rarely published Outside this journal, the journal has so far not published any papers on driverless cars.
Driverless cars typically use conventional 2D cameras and depth-sensing LiDAR modules to recognize the world around them, and while these modules provide depth information, they are less advanced due to the lower resolution, so without the real-time data provided by conventional cameras Difficult to detect long-distance small objects.
But Apple researchers say they can get "extremely encouraging results" with the new software, recognizing pedestrians and cyclists with LiDAR data alone, and they also said they can beat other only LiDAR-aware 3D object method.This experiment completely using computer simulation, not on the road test.
Although Apple CEO Tim Cook once referred to driverless cars as the "mother of all AI projects," they did not disclose the company's research in this area.
Apple told U.S. regulators in December last year that the company was excited about the technology and hoped regulators would not limit the tests.
Apple also submitted a test plan for driverless cars to California regulators in April of this year.