Luxury vehicle maker Audi recently sent its self-driving prototype model for a 550-mile test drive from San Francisco to Las Vegas.
However, the autonomous vehicle, nicknamed “Jack,” faced some performance issues along the way that might delay its rollout to the consumer market.
According to a March 23 Motrolix article, Audi had expected the A7 Piloted Driving Concept test drive to go off without a hitch. During the drive, a piece of tumbleweed got lodged in “Jack”‘s grille, blocking one of its sensors for about 10 to 15 miles.
While the car was ultimately able to stay on course, the tumbleweed highlights a major issue faced by self-driving car developers. It’s literally impossible to program a car to perform every behavior and response possible, or to be able to anticipate unexpected events.
Rather than program autonomous vehicles to do every behavior possible, Jen-Hsun Huang, CEO of software company Nvidia Corp., explained that programmers are now working to instill a sense of “deep learning” into these cars, making their computers function more similarly to the human brain, learning continuously the more they’re used.
Nvidia Corp., a company whose clients include Audi and Tesla, plans to introduce its first autonomous vehicle computer outfitted with this “deep learning technology” sometime this May. The Drive PX automotive computer will cost approximately $10,000, Automotive News reports.
It’s far from being the only company to embrace deep learning. Israel-based Mobileye’s EyeQ3 chips, which are used for automatic braking, will now include deep learning technology as well as more traditional algorithms. On February 6, Microsoft announced that its supercomputer was the first ever to beat a human at recognizing some 1.5 million images from a database.
With this technology, cars could soon be better than even humans at detecting obstacles and obstructions like baby deer — or even tumbleweeds.