Andrej Karpathy, Tesla’s director of artificial intelligence, announced Wednesday he’s leaving the company only months before its anticipated release of its long-delayed “full self-driving” software to 1 million people. Tesla’s driver-assist features made significant advances in his tenure, but also have drawn increased scrutiny from regulators over their safety record.
“It’s been a great pleasure to help Tesla towards its goals,” Karpathy tweeted. “Autopilot graduated from lane keeping to city streets and I look forward to seeing the exceptionally strong Autopilot team continue that momentum.”
Karpathy said that he had no concrete plans for his next move, but he planned to revisit “technical work in AI, open source and education.”
“Thanks for everything you have done for Tesla! It has been an honor working with you” CEO Elon Musk tweeted Wednesday.
Neither Karpathy nor Musk provided an explanation for his departure. Karpathy’s five-year tenure was longer than many previous leaders on the Autopilot team. Tesla is notorious for a demanding culture that can lead to turnover, but it’s not clear if that played any role in his exit. Karpathy’s announced departure comes several months after Musk announced in March that Karpathy would be taking a roughly four-month sabbatical.
Karpathy said at the time that he was traveling in Europe and Asia with only a backpack. He praised European cities during his travels for being more unique, clean, safe and “pedestrian/bike friendly.”
It’s not uncommon for Tesla executives to make a soft exit from the company, departing shortly after going on a leave of absence, like executive Doug Field, who led the development of the Model 3 and is now at Ford, or chief people officer Gaby Toledano.
Tesla owners have described seeing improvements to Autopilot during Karpathy’s tenure. Many say it makes them feel less fatigued after long trips. But Autopilot has also been controversial. Drivers can become distracted while using it, which has at times been deadly. Tesla has since expanded its driver monitoring technology, with an in-car camera. The naming of Tesla’s driver-assist technologies, Autopilot and “full self-driving,” have also been criticized as inducing more trust from some drivers than is prudent, given the systems’ limitations.
The National Highway Traffic Safety Administration has spent nearly a year investigating reports of Teslas using Autopilot rear-ending first responder vehicles parked on highways to respond to crashes.
The investigation may lead to Tesla issuing a recall, or litigation between Tesla and the federal government.
Tesla already recalled “full self-driving,” an advanced version of Autopilot intended for city streets, earlier this year as it was not coming to complete stops at stop signs. The flaw was fixed with an over-the-air update, a recall fix that’s less onerous for automakers and consumers.
“Full self-driving” has been celebrated by many of Tesla’s ardent fans. But many of their friends and loved ones have been less enthusiastic, and the National Highway Traffic Safety Administration released data this month showing that Teslas using its driver-assist technology have been involved in 273 crashes.
“Full self-driving” can drive in a jerky, uncomfortable manner. In some cases, it steers, brakes and accelerates smoothly. Other times, its flaws can be unsettling. Tesla warns drivers that it “may do the wrong thing at the worst time.”
Karpathy was an advocate for Tesla’s unique approach to developing autonomous driving features. The automaker relies on cameras to perceive the world, rather than using additional sensors like radar and lidar, which are the norm among nearly all its competitors, including Alphabet’s Waymo and General Motors’ Cruise.
“Is that person distracted and on their phone? Are they going to walk into your lane?” Karpathy said in 2019. “Those answers are only found in vision.”
Karpathy was already highly regarded among artificial intelligence experts before joining Tesla in 2017. He left a position as a research scientist at OpenAI, a research laboratory that Musk was initially involved in. Karpathy also created Stanford’s first class on deep learning, a trendy type of artificial intelligence that’s seen huge gains in recent years. Karpathy studied artificial intelligence at the University of Toronto with Geoffrey Hinton, a legend in the field.
Karpathy and Tesla did not immediately respond to requests for comment.