When Raquel Urtasun joined Uber in May 2017, she walked into a battleground.
Urtasun, a leading artificial intelligence researcher who was (and remains) a professor at the University of Toronto and a cofounder of the Vector Institute for AI, had been hired as the new R&D lead for Uber ATG, the ride-hail giant's self-driving arm, and to lead a new artificial intelligence-focused lab.
But Uber was in turmoil. Less than three months before Urtasun joined, Waymo had launched its explosive lawsuit accusing the ride-hail giant of stealing its trade secrets. The suit centered on Anthony Levandowski, who had left Google and joined Uber the year before to lead the autonomy effort that CEO Travis Kalanick considered vital to his company's long-term survival.
Uber had sidelined (and would soon fire) Levandowski, who was ultimately convicted on a criminal charge of trade secret theft (only to be pardoned by an outgoing President Trump). Kalanick was himself under fire, as a series of scandals exposed Uber's toxic, often misogynistic corporate culture, and would resign in June.
Considering the chaos, Urtasun appeared to be a brilliant hire for ATG. Not only could her pioneering work in artificial intelligence be untouched by any claims of intellectual property theft, it could seriously propel the company's self-driving ambitions.
That didn't quite work out. While she was respected and well-liked, according to two former Uber engineers who interacted with Urtasun during her tenure, her impact on Uber's self-driving program was limited.
Part of the problem was that Urtasun's novel machine learning algorithms weren't easily plugged into a software stack that was several years old when she arrived. "She didn't have a big impact," one of the former engineers told Insider. "To her frustration and that of others. She clashed with the team because you can't just hand off an algorithm to a roboticist."
In response, Urtasun said she "had a significant impact on ATG's software stack," and that her time leading R&D at the ride-hail giant taught her what worked — and didn't — in using advanced AI in self-driving cars. "From my perspective, AI needs to be at the center of the solution in order for the technology to scale," she said. "It can't be peripheral or plugged into an old software stack."
The idea that Urtasun — or anyone — could rehab Uber's self-driving program fully fell apart in March 2018, when one of its self-driving cars killed a woman crossing the street in Arizona, in part because it failed to identify her as a pedestrian. Uber essentially iced the effort, letting it limp along with minimal on-street testing until December 2020, when it offloaded it to Aurora. Aurora did not make an offer to Urtasun and her team, TechCrunch reported.
A fresh start
Now, Urtasun is returning to the self-driving game with a startup of her own, free of baggage and loaded with cash. She is the founder and CEO of Waabi, the Toronto-based startup that emerged from stealth Tuesday morning, announcing it had raised $83.5 million in Series A funding. Khosla Ventures led the round, with participation from Uber, Aurora, 8VC, Radical Ventures, Omers Ventures, BDC, and AI bigwigs Geoffrey Hinton and Fei-Fei Li.
The startup will focus on autonomous trucks, where a clear commercial use case and a relatively simple driving environment make for an easier problem than piloting a robo-taxi through a city center. And rather than build its own vehicle from the ground up, Urtasun said Waabi will be "very partnership friendly," eager to team up with existing manufacturers, hardware providers, and so on.
Urtasun launched her own company because in a field dominated by a handful of players — Waymo, Cruise, Argo, and Aurora chief among them — she sees a whole lot of same.
"What's happening in the industry is that there is really one way of solving this problem," she told Insider, noting that most of the leaders of the field share a pedigree, having started off on the early Google Chauffeur team that sprang out of DARPA's Grand and Urban Challenges. Those competitions, held between 2004 and 2007, sparked today's self-driving industry and did much to set the recipe for how an autonomous vehicle is made. After more than a decade of work and billions spent, commercial self-driving deployments are rare and small.
"The world is trying to solve this very complex task in the same way," she said. "So there is a need for a diversity of solutions."
Urtasun's pitch is that by leaning heavily on artificial intelligence, namely deep learning, Waabi can move much faster than her established competitors, with many fewer engineers and much less investment. "If you want to do that wholesale, the best route is to start a new company," she said.
New blood
Still, the idea of launching an autonomy-focused startup feels very 2016. Over the past few years, as the sheer difficulty of the problem has become clear, investment in the space has cooled and consolidation has been the order. The prevailing wisdom is that making a car drive itself takes a certain amount of brute force.
"I can't think of another example of a situation where you have to get to the point where you have literally thousands of engineers, billions of dollars of capital, and no product yet," Cruise CEO Dan Ammann said in a 2019 interview. "This is not a problem you can solve with 50 or 100 engineers."
Urtasun is looking for the middle ground. She'll use her new funding to grow her current team of about 40 (many of them holdovers from her Uber days), but she isn't building a juggernaut. And that clearly has its appeal.
While some investors believe the likes of Waymo and Cruise will dominate, "others believe there's a need for new blood, smart people taking a new approach," said Shahin Farschi, a partner at Lux Capital who was an early investor in Zoox. Waabi's impressive Series A makes that clear.
But AV trucking is an already crowded space. If Waabi wants to form partnerships, Farschi said, "the question becomes, how do you integrate with those teams?"
And that's a problem Waabi will have to solve with more than artificial intelligence.