Raquel Urtasun, founder and CEO of autonomous trucking startup Waabi, has spent the past twenty years developing AI systems that may think like a human.
The AI pioneer previously served as chief scientist at Uber ATG before launching Waabi in 2021. Waabi was launched with an “AI-first approach” to speed up the business adoption of autonomous vehicles, starting with long-haul trucks.
“If you possibly can construct systems that may actually do this, suddenly you wish quite a bit less data,” Urtasun told TechCrunch. “You need quite a bit less calculations. If you're capable of draw the conclusions efficiently, you don't have to have fleets of vehicles everywhere in the world.”
Tesla's vision-first approach to autonomous driving is attempting to construct an AV stack with AI that perceives the world like a human and reacts in real time. The difference – except for Waabi's experience with lidar sensors – is that Tesla's fully autonomous driving system uses “imitation learning” to learn to drive. To do that, Tesla must collect and analyze hundreds of thousands of videos of real-world driving situations to coach its AI model.
The Waabi driver, however, has done most of its training, testing, and validation using a closed-loop simulator called Waabi World, which mechanically creates digital twins of the world from data, runs real-time sensor simulations, creates scenarios for stress testing the Waabi driver, and teaches the motive force to learn from its mistakes without human intervention.
In just 4 years, this simulator has helped Waabi launch business pilots (with a human driver within the front seat) in Texas, lots of them through a partnership with Uber Freight. Waabi World can also be enabling the startup to realize its planned business launch with fully driverless vehicles in 2025.
But Waabi's long-term mission involves far more than simply trucks.
“This technology is incredibly, extremely powerful,” said Urtasun, chatting with TechCrunch via video interview, a whiteboard covered in hieroglyphic-like formulas behind her. “It has this amazing ability to generalize, it's very flexible, it could possibly be developed in a short time. And it's something that we will expand in the longer term to far more than simply trucking… That could possibly be robot taxis. That could possibly be humanoids or warehouse robots. This technology can solve any of those use cases.”
The potential of Waabi's technology – which can initially be used to deploy autonomous trucks at scale – has enabled the startup to shut a $200 million Series B funding 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. With this round, Waabi's total funding stands at $283.5 million.
The size of the round and the strength of its participants are particularly notable considering the setbacks the AV industry has faced lately. In the trucking space alone, Embark Trucks was forced to shut down, Waymo decided to pause its autonomous freight business, and TuSimple closed its U.S. operations. In the robotaxi space, Argo AI faced its own shutdown, Cruise lost its operating licenses in California after a serious safety incident, Motional cut nearly half of its workforce, and regulators are actively investigating Waymo and Zoox.
“The strongest firms are built by raising money in really difficult moments, and the AV industry particularly has seen lots of setbacks,” Urtasun said.
However, AI-focused players on this second wave of autonomous vehicle startups have secured impressive capital raises this 12 months. UK-based Wayve, which can also be developing a self-learning reasonably than rule-based autonomous driving system, closed a $1.05 billion Series C led by SoftBank Group in May. And Applied Intuition raised $250 million at a $6 billion valuation in March to bring AI to the automotive, defense, construction and agriculture industries.
“In the context of AV 1.0, it's very clear today that it's very capital intensive and really slow to maneuver forward,” Urtasun said, noting that the robotics and autonomous driving industry has been slowed down by complex and fragile AI systems. “And investors, I’d say, usually are not very captivated with that approach.”
But what's exciting investors today is the promise of generative AI, a term that wasn't exactly in vogue when Waabi was launched, but still describes the system Urtasun and her team have developed. According to Urtasun, Waabi is a next-generation gen AI that may be utilized in the physical world. And unlike today's popular language-based gen AI models like OpenAI's ChatGPT, Waabi has discovered easy methods to construct such systems without counting on huge data sets, large language models, and all of the computing power that comes with them.
The Waabi driver, Urtasun says, has the remarkable ability to generalize. So as an alternative of attempting to train a system on each possible data point that has ever existed or could ever exist, the system can learn from just a few examples and cope with the unknown in a protected way.
“That was a part of the design. We built these systems that may perceive the world, create abstractions of the world, after which use those abstractions and think, 'What might occur if I do that?'” Urtasun said.
This more human, reasoning-based approach is way more scalable and capital efficient, Urtasun says. It's also critical for validating safety-critical systems running at the sting. You don't need a system that takes just a few seconds to reply, otherwise there's an accident, she said. Waabi terminated a partnership Integrating Nvidia's Drive Thor into its self-driving trucks, giving the startup access to automotive-grade computing power at scale.
On the road, it looks just like the Waabi driver understands that there’s something solid ahead and he should drive fastidiously. He may not know what the something is, but he knows he must avoid it. Urtasun also said that the motive force is capable of predict the behavior of other road users without having to be trained in specific situations.
“It understands things without us having to inform the system anything in regards to the concept of objects, how they move on the planet, that various things move in another way, that there’s occlusion, that there’s uncertainty, easy methods to behave when it’s raining heavily,” Urtasun said. “It learns all of this stuff mechanically. And since it is exposed to driving scenarios straight away, it learns all of those skills.”
She noted that Waabi's streamlined, unified architecture may be applied to other autonomy use cases.
“If you expose it to interactions in a warehouse, picking things up and dropping them, it could possibly learn that easily,” she said. “You can expose it to multiple use cases and it could possibly learn to perform all of those skills directly. There are not any limits to what it could possibly do.”
