‘Hey car, why did you do that?’
Automated driving startup Wayve—fresh off an announcement that Canada will house its third R&D hub, after those in the U.K. and Silicon Valley—wants you to be able to question your car’s choices.
The startup, which is backed by Microsoft and top artificial intelligence pioneers like Yann LeCun and Open AI’s Canadian chief scientist Ilya Sutskever, is setting up shop in Vancouver to further its research as it unveils Lingo-2, a “driving commentator” program that will provide people in automated vehicles with continuous commentary explaining what the car sees on the road and what action it plans to take.
The details: Wayve plans to hire about 10 people for the Vancouver office within the first year, focused on research and development, particularly for its AI models. It hasn’t yet pinned down a physical space.
While the company does have a fleet of vehicles in the U.K. that it uses for internal testing across six different platforms, including the Ford Mach-E, its primary focus in Vancouver is developing software and machine learning, chief scientist Jamie Shotton said in an interview.
(Which is good news, since B.C.’s government said earlier this month that vehicles will require provincial permission to drive with advanced systems that automate steering, braking and acceleration simultaneously.)
Wayve’s main hope is that the Canadian team will look for scientific breakthroughs, as Vancouver prepares to host one of the world’s most influential AI conferences, NeurIPS, later this year. Wayve also said its Vancouver office will serve as a hub to connect with its partners Nvidia and Microsoft, both of which are expanding in Canada.
The company’s business strategy is to start by selling advanced driver-assistance programs to automakers, and it hopes to build relationships with automakers.
The backdrop: Wayve’s expansion comes as automated-driving technology is being deployed more widely than ever—and as it’s under more scrutiny.
The new Wayve: Developing Lingo-2, the software that explains its decisions, is one way the company has tried to differentiate its products from others hitting the road. Shotton said Wayve is hoping it will be easier for drivers to get used to automation if they start with advanced driver-assistance technology alongside Lingo-2.
The company’s flavour of self-driving vehicles run mostly on data from cameras, and can operate without HD maps. The older approach to automated driving relies on mapping the world down to the centimetre, as well as giving vehicles a long list of rules to follow. That can make those systems “brittle” and easy to break when something unexpected happens, like a plastic bag flying across the windshield, Shotton said.
Wayve instead trains its models on the behaviour behind the wheel of humans well versed in the rules of the road, like fire-engine drivers and police officers, an approach that falls under the broader school of research called embodied AI. Whereas ChatGPT is cognitive AI—mimicking human-like thought—embodied AI focuses more on human-like decisions and actions, Shotton said.
“So as it’s driving around, it can explain to you what it’s doing, and say ‘I’m slowing down for the red light,’ or ‘I see the pedestrian in the zebra crossing, so I’m giving way,’” Shotton said. “It gives you a glimpse into what it’s seeing.”
Read Shift—The Logic’s authoritative weekly newsletter on automotive technology industry news—for more; and if you know someone who should be reading it, they can sign up here.
