Raquel Urtasun, founder and CEO of self-driving truck startup Waabi, has been working for the past 20 years to develop AI systems that can reason like humans.
The AI pioneer was the chief scientist at Uber ATG before launching Waabi in 2021. Waabi was launched with an “AI-first approach” to accelerate the commercial deployment of autonomous vehicles, starting with long-haul trucks.
“If we could actually build a system that could do that, it would require a lot less data,” Urtasun told TechCrunch. “It would also require a lot less computation. If we could do the inference efficiently, we wouldn't need to have a fleet of vehicles around the world.”
Building an AI-powered AV stack that perceives the world like a human and reacts in real time is what Tesla has been trying to do with its vision-first self-driving approach. Apart from Waabi's familiarity with using lidar sensors, Tesla's fully self-driving system is different in that it uses “imitation learning” to learn how to drive. This requires Tesla to collect and analyze millions of videos of real driving situations to train its AI models.
Meanwhile, Waabi Driver does most of its training, testing and validation using a closed-loop simulator called Waabi World, which automatically builds a digital twin of the world from data, runs real-time sensor simulations, creates scenarios to stress-test Waabi Driver, and teaches the driver to learn from their mistakes without human intervention.
The simulator has helped Waabi launch commercial pilots (with a human driver in the passenger seat) in Texas in just four years, many of which have been made possible through a partnership with Uber Freight. Waabi World is also helping the startup achieve its goal of fully driverless commercialization, which it plans for 2025.
But Waabi's long-term mission is much grander than just trucks.
“This technology is extremely powerful,” Urtasun told TechCrunch in a video interview, standing in front of a whiteboard with hieroglyphic-like mathematical formulas written behind him. “This technology has an incredible ability to generalize, it's very flexible, it's very fast to develop. And in the future, it could potentially be expanded to a variety of uses beyond trucking. This could be robo-taxis. This could be humanoids or warehouse robots. This technology can address any of those use cases.”
Waabi's technology will first be used to scale self-driving trucks, and the startup was able to close a $200 million Series B round led by existing investors Uber and Khosla Ventures. Strong strategic investors include Nvidia, Volvo Group Venture Capital, Porsche Automobil Holding SE, Scania Invest and Ingka Investments. The round brings Waabi's total funding to $283.5 million.
The size of this round and strength of participants are particularly notable given the hits the autonomous vehicle industry has taken in recent years: In the trucking space alone, Embark Trucks closed, Waymo decided to suspend its autonomous freight business, and TuSimple closed its U.S. operations. Meanwhile, in the robotaxi space, Argo AI faces closure, Cruise lost its license to operate in California following a serious safety incident, Motional cut nearly half of its workforce, and regulators are actively investigating Waymo and Zoox.
“Raising capital during tough times actually builds the strongest companies, and the AV industry in particular has seen a lot of setbacks,” Urtasun said.
Still, AI-focused companies in this second wave of autonomous vehicle startups have raised some impressive funding this year. UK-based Wayve, which is developing a self-learning system rather than rules-based for autonomous driving, closed a $1.05 billion Series C round led by SoftBank Group in May, while Applied Intuition raised $250 million in March at a $6 billion valuation to bring AI to automotive, defense, construction and agriculture.
“From an AV 1.0 perspective, it's clear today that it's capital intensive and very slow to progress,” Urtasun said, noting that the robotics and self-driving industries are lagging behind with complex and brittle AI systems, “and I think investors are not very keen on this approach.”
But what investors are excited about now is the potential of generative AI, a term that wasn't in vogue when Waabi launched, but that still describes the system Urtasun and her team created. Urtasun says Waabi's system is the next generation of genAI, one that can be deployed in the physical world. And unlike today's popular language-based genAI models, such as OpenAI's ChatGPT, Waabi has figured out how to create such a system without relying on huge datasets, large language models, and all the computing power that comes with them.
Urtasun said Wabi Driver has a good ability to generalize, so rather than training the system on every data point that has ever existed or may ever exist, the system can learn from a few examples and safely handle the unknown.
“That was in the design: We built a system that perceives the world, makes abstractions of the world, and can reason based on those abstractions about 'what happens if I do this?'” Urtasun said.
This more human-like, reasoning-based approach is much more scalable and capital-efficient, Urtasun said. It's also essential for validating safety-critical systems that run on the edge. You don't want a system that takes seconds to react or the vehicle will crash, he said. Wabi announced a partnership to bring NVIDIA's Drive Thor to its self-driving trucks, giving the startup access to automotive-grade computing power at scale.
On the road, Waabi drivers seem to understand that there is something solid ahead and they need to drive carefully. They may not know what it is, but they know how to avoid it. Urtasun also said that drivers are now able to predict the behavior of other road users without having to be trained in different specific situations.
“It understands things without us telling it: the concept of objects, how objects move in the world, that different objects move differently, that there are occlusions, that there is uncertainty, how to act in heavy rain,” Urtasun said. “It learns all of this automatically, and now it learns all these capabilities as it is exposed to driving scenarios.”
She noted that Waabi's single, streamlined architecture can also be applied to other autonomous use cases.
“If you expose a robot to tasks in a warehouse — lifting things and dropping things — it has no problem learning to do that,” she says. “If you expose it to multiple use cases, it can learn all those skills together. There's no limit to what a robot can do.”