This startup wants to create self-driving car software at breakneck speed

For the last A year and a half ago, two white Tesla Model 3 sedans loaded with five additional cameras and a palm-sized supercomputer drove quietly around San Francisco. In a city and age filled with questions about the capabilities and limits of artificial intelligence, the startup behind modified Tesla cars is trying to answer a simple question: How quickly can a company build software for self-driving vehicles today?

The startup, which is announcing its activities to the public for the first time today, is called HyprLabs. Its team of 17 people (only eight of them full-time) is split between Paris and San Francisco, and the company is run by Zoox co-founder Tim Kentley Clay, who abruptly exited the now Amazon-owned company in 2018. Hypr has taken in relatively little funding, $5.5 million since 2022, but its ambitions are wide-ranging. Eventually, she plans to build and operate her own robots. “Think of the love child of R2-D2 and Sonic the Hedgehog,” says Kentley Clay. “You will define a new category that does not currently exist.”

Currently, the startup is announcing its software product called Hyprdrive, which it sees as a leap forward in how engineers train vehicles to drive themselves. These kinds of leaps are present throughout the robotics field, thanks to advances in machine learning that promise to reduce the cost of training self-driving vehicle software, and the amount of human labor involved. This development in training has brought a new movement to a space that has suffered for years from a “trough of disappointment,” as technology makers fail to meet their own deadlines for operating robots in public spaces. Now, taxis are ferrying paying passengers in more and more cities, and automakers are making ambitious new promises about bringing autonomous driving to customers’ personal cars.

But using a small, agile, and cheap team to go from “driving well” to “driving much more safely than a human” is a tall hurdle in itself. “I can’t tell you, honestly, that this is going to work,” Kentley-Clay says. “But what we’ve built is a really strong signal. It just needs to step up.”

Old technology, new tricks

HyprLabs’ software training technology is a departure from other robotics startups’ methods of teaching their systems how to drive themselves.

First, some background: For many years, it seemed like the big battle in self-driving vehicles was between those who only used cameras to train their software – Tesla! – And those who relied on other sensors too – Waymo and Cruise! – Including sensors and radar that were previously expensive. But beneath the surface, larger philosophical differences emerged.

Camera-only followers like Tesla wanted to save money while planning to launch a huge fleet of robots; For a decade, CEO Elon Musk’s plan has been to suddenly turn all of his customers’ cars into self-driving cars through 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 was fed into a so-called “holistic” machine learning model through reinforcement. The system takes pictures —bike-And issues command orders-Move the steering wheel to the left and slow down to avoid hitting it. “It’s like training a dog,” says Philip Koopman, a researcher in autonomous vehicle software and safety at Carnegie Mellon University. “In the end, you say, ‘Bad dog’ or ‘good dog.’”