Jay Gierak of Ghost, based in Mountain View, California, is impressed with Wayve’s demonstrations and agrees with the company’s overall stance. “The robotics approach is not the right way to do this,” says Gierak.
But he’s not sold on Wayve’s total dedication to deep learning. Instead of one large model, Ghost trains many hundreds of smaller models, each with a specialism. It then encodes simple rules that tell the self-driving system which models to use in which situations. (Ghost’s approach is similar to that of another AV2.0 company, Autobrains, based in Israel. But Autobrains uses yet another layer of neural networks to learn the rules.)
According to Volkmar Uhlig, co-founder and CTO of Ghost, breaking down the AI into many smaller pieces, each with specific functions, makes it easier to determine that an autonomous vehicle is safe. “At some point, something happens,” he says. “And a judge will ask you to point to the code that says, ‘If there’s a person in front of you, brake.’ That piece of code has to be there.” The code can still be learned, but in a large model like Wayve’s, it would be hard to find, Uhlig says.
Still, the two companies pursue complementary goals: Ghost wants to create consumer vehicles that can drive themselves on highways; Wayve aims to be the first company to place self-driving cars in 100 cities. Wayve is now working with UK supermarket giants Asda and Ocado, collecting data from their city delivery vehicles.
Yet both companies lag far behind the market leaders in many respects. Cruise and Waymo have driven hundreds of hours without people in their cars and are already offering robotic taxi services to the public in a small number of locations.
“I don’t want to narrow the scale of the challenge ahead,” Hawke says. “The AV industry teaches you humility.”