For the past 20 years, Raquel Urtasun, founder and CEO of self-driving truck startup Waabi, has been developing AI systems that can reason like humans.
The AI pioneer served as chief scientist at Uber ATG before launching Waabi in 2021. Waabi is launching with an “AI-first approach” to speed up the commercial deployment of autonomous vehicles, starting with long-haul trucks.
“If you can actually build a system that can do that, suddenly you need a lot less data,” Urtasun told TechCrunch. “It requires a lot less computation. “If we can do inference in an efficient way, we don’t need to deploy vehicles all over the world.”
Building an AV stack with AI that human-poweredly perceives the world and reacts in real time is what Tesla has attempted to do with its vision-first approach to autonomous driving. The difference, aside from Waabi's comfort with using LiDAR sensors, is that Tesla's fully self-driving system uses “imitation learning” to learn how to drive. This requires collecting and analyzing millions of videos of real-world driving situations that Tesla uses to train its AI models.
Waabi Driver, on the other hand, did 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. Perform real-time sensor simulations. Create a scenario for stress testing Waabi Driver. Teach drivers to learn from their mistakes without human intervention.
In just four years, the simulator helped Waabi launch commercial pilots (with a human driver in the front seat) in Texas, many of them through a partnership with Uber Freight. Waabi World is also helping the startup achieve a fully driverless commercial launch planned for 2025.
But Waabi's long-term mission is much grander than just a truck.
“This technology is very powerful,” Urtasun said in a video interview with TechCrunch. She had a white board behind her full of formulas that looked like her hieroglyphs. “It has incredible generalizability, is very flexible, and can be developed very quickly. And this is something that can be expanded to do more than just trucking in the future. It could be a robot taxi. This could be a humanoid or a warehouse robot. “This technology can solve all of these use cases.”
The promise of Waabi's technology, which will be used for the first time to scale autonomous trucking, helped the startup close a $200 million Series B round led by existing investors Uber and Khosla Ventures. Powerful strategic investors include Nvidia, Volvo Group Venture Capital, Porsche Automobil Holding SE, Scania Invest and Ingka Investments. This round brings Waabi's total funding to $283.5 million.
The size of this round and the strength of its participants are particularly noteworthy considering the hits the AV industry has achieved in recent years. In the trucking space alone, Embark Trucks closed, Waymo decided to exit its autonomous freight business, and TuSimple closed its U.S. operations. Meanwhile, in the robotaxi space, Argo AI is on the brink of closure, Cruise lost its license to operate in California after a major safety incident, and Motional cut nearly half its staff, while regulators is actively investigating Waymo and Zoox.
“Raising money in really difficult moments is what builds the strongest companies, and the AV industry in particular has seen a lot of setbacks,” Urtasun said.
That said, AI-focused companies participating in the second wave of self-driving car startups have secured impressive capital raisings this year. UK-based Wayve is also developing self-learning, rather than rules-based, systems for autonomous driving, and in May closed a $1.05 billion Series C led by SoftBank Group. And last March, Applied Intuition raised $250 million in funding at a $6 billion valuation to bring AI to the automotive, defense, construction and agriculture sectors.
“When you look at it in the context of AV 1.0, it’s very clear that it’s very capital intensive and the pace of progress is very slow,” Urtasun said, noting that the robotics and autonomous driving industries are being hampered by complex and fragile AI systems. I did. “And I would say investors are not very interested in that approach.”
What investors are excited about today is the promise of generative AI. This term wasn't popular when Waabi launched, but it nonetheless describes the system Urtasun and her team created. Urtasun says Waabi is the next generation of genAI that can be deployed in the physical world. Unlike today's popular language-based genAI models, such as OpenAI's ChatGPT, Waabi 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 says Waabi Driver has incredible generalization abilities. So, rather than trying to train a system for every single data point that has or could exist, the system can learn from a few examples and handle unknown data in a safe way.
“That was in the design. We’ve built a system that can perceive the world, create an abstraction of the world, and then take that abstraction and reason about, ‘What would happen if I did this?’” Urtasun said.
Human-like reasoning-based approaches are much more scalable and more capital efficient, Urtasun says. It is also important for validating safety-critical systems running at the edge. She doesn't want a system that takes seconds to respond. Otherwise, she said, the vehicle would crash. Waabi announced a partnership to bring Nvidia's Drive Thor to self-driving trucks. This will give startups access to automotive-grade computing power at scale.
On the road, it appears that the Wabi driver must understand that there is something solid ahead and must drive carefully. You may not know what it is, but you will know how to avoid it. Urtasun also said drivers could predict how other road users will behave in a variety of specific situations without the need for training.
Urtasun “understands things without telling the system about the concept of objects, how they move in the world, whether different things move differently, whether there is closure, whether there is uncertainty, how to behave when it rains a lot”. said. “All of this is learned automatically. And because they are exposed to driving scenarios right now, they learn all those skills.”
She noted that Waabi's single, streamlined architecture can be applied to other autonomy use cases.
“If they are exposed to the interaction of picking up and dropping objects in a warehouse, they can learn without problems,” she said. “You get exposure to different use cases and learn how all the technologies work together. “There are no limits to what you can do.”
