\n<\/p>\n
For the last<\/span> year and a half, two hacked white Tesla Model 3<\/a> sedans each loaded with five extra cameras and one palm-sized supercomputer have quietly cruised around San Francisco<\/a>. In a city and era swarming with questions about the capabilities and limits of artificial intelligence, the startup behind the modified Teslas is trying to answer what amounts to a simple question: How quickly can a company build autonomous vehicle software today?<\/p>\n The startup, which is making its activities public for the first time today, is called HyprLabs<\/a>. Its 17-person team (just eight of them full-time) is divided between Paris and San Francisco, and the company is helmed by an autonomous vehicle company veteran, Zoox<\/a> cofounder Tim Kentley-Klay, who suddenly exited<\/a> the now Amazon-owned firm in 2018. Hypr has taken in relatively little funding, $5.5 million since 2022, but its ambitions are wide-ranging. Eventually, it plans to build and operate its own robots. \u201cThink of the love child of R2-D2 and Sonic the Hedgehog,\u201d Kentley-Klay says. \u201cIt’s going to define a new category that doesn’t currently exist.\u201d<\/p>\n For now, though, the startup is announcing its software product called Hyprdrive, which it bills as a leap forward in how engineers train vehicles to pilot themselves. These sorts of leaps are all over the robotics space, thanks to advances in machine learning that promise to bring down the cost of training autonomous vehicle software, and the amount of human labor involved. This training evolution has brought new movement to a space that for years suffered through a \u201ctrough of disillusionment<\/a>,\u201d as tech builders failed to meet their own deadlines to operate robots in public spaces. Now, robotaxis pick up paying passengers in more and more cities<\/a>, and automakers make newly ambitious promises about bringing self-driving to customers\u2019 personal cars<\/a>.<\/p>\n But using a small, agile, and cheap team to get from “driving pretty well\u201d to \u201cdriving much more safely than a human\u201d is its own long hurdle. \u201cI can’t say to you, hand on heart, that this will work,\u201d Kentley-Klay says. \u201cBut what we\u2019ve built is a really solid signal. It just needs to be scaled up.\u201d<\/p>\n HyprLabs\u2019 software training technique is a departure from other robotics\u2019 startups approaches to teaching their systems to drive themselves.<\/p>\n First, some background: For years, the big battle in autonomous vehicles seemed to be between those who used just cameras to train their software\u2014Tesla!\u2014and those who depended on other sensors, too\u2014Waymo, Cruise!\u2014including once-expensive lidar<\/a> and radar. But below the surface, larger philosophical differences churned.<\/p>\n Camera-only adherents like Tesla wanted to save money while scheming to launch a gigantic fleet of robots; for a decade, CEO Elon Musk\u2019s plan has been to suddenly switch all of his customers\u2019 cars to self-driving ones with the push of a software update. The upside was that these companies had lots and lots of data, as their not-yet self-driving cars collected images wherever they drove. This information got fed into what\u2019s called an \u201cend-to-end\u201d machine learning model through reinforcement. The system takes in images\u2014a bike<\/em>\u2014and spits out driving commands\u2014move the steering wheel to the left and go easy on the acceleration to avoid hitting it<\/em>. \u201cIt\u2019s like training a dog,\u201d says Philip Koopman, an autonomous vehicle software and safety researcher at Carnegie Mellon University. \u201cAt the end, you say, \u2018Bad dog,\u201d or \u2018Good dog.\u2019\u201d<\/p>\n<\/div>\nOld Tech, New Tricks<\/h2>\n