Raquel Urtasun, featured on this year’s A.I. Power Index, is rethinking how autonomous vehicles are built and deployed. As founder and CEO of Waabi and a professor of computer science at the University of Toronto, Urtasun has challenged the prevailing belief that progress in A.I. is simply a matter of scaling bigger models with more data and compute. Instead, Waabi is building what she calls “next-generation AV 2.0,” an interpretable and verifiable end-to-end system capable of reasoning new situations on the road, something rule-based AV 1.0 models have consistently failed to achieve at scale. Since its founding in 2021, Waabi has attracted more than $200 million in backing from Uber, Nvidia, Porsche and others, signaling both industry validation and a bet on Urtasun’s vision. Waabi aims to deploy fully driverless trucks in Texas by the end of 2025, a timeline that would put it years ahead of rivals.
What’s one assumption about A.I. that you think is dead wrong?
There’s a widespread assumption that the next A.I. breakthrough will come simply from more compute and larger models. This has been proven to be untrue, and we’re seeing increasingly diminishing returns, yet this philosophy still underpins the industry’s broad approach to developing A.I.
If you had to pick one moment in the last year when you thought “This changes everything” about A.I., what was it?
The breakthrough in end-to-end A.I. models capable of reasoning changes everything, especially for Physical A.I.
At Waabi, we have pioneered an interpretable and verifiable end-to-end A.I. model for self-driving that can safely generalize to all possible scenarios on the road through reasoning, including those that are very different from what the model has seen during training. That is a true game-changer and will unlock the safe commercial deployment of AVs for the logistics industry, something that will be transformative for supply chains and economies.
The applications don’t stop there, however. The flexibility and generalization abilities of our end-to-end A.I. will empower robots with very different form factors, from other types of vehicles to humanoids and warehouse robotics. This will completely reimagine the physical world and empower humans in extraordinary new ways, unlocking unprecedented efficiencies, enhancing safety and creating opportunities we can only begin to imagine.
What’s something about A.I. development that keeps you up at night that most people aren’t talking about?
The current “more is more” development approach (bigger models, more data, more compute) remains something that needs to be urgently addressed. The environmental and societal costs of A.I. development are staggering. At the same time, the unequal distribution of the resources required for A.I. development risks deepening global inequalities, leaving much of the world behind.
In the interests of our world and future generations, we must prioritize sustainable A.I. over the brute force scaling of today’s models. This is the only way to unlock the true potential of A.I. while ensuring this technology benefits all of humanity.
You’re targeting the deployment of fully driverless trucks in Texas by the end of 2025—that’s incredibly ambitious, given the struggles of other AV companies. What makes your next-generation AV2.0 approach fundamentally different from the preprogrammed responses that haven’t worked?
The traditional AV 1.0 approach we’ve seen to date faces critical barriers to deployment. It’s reliant on hand-coded rules and vast amounts of real-world driving data, which is capital-intensive and doesn’t scale. Our next-generation AV2.0 approach leverages an end-to-end interpretable and verifiable A.I. model powered by the industry’s most realistic neural simulator.
This enables:
- Significant efficiency in terms of capital and time to market
- Enhanced safety performance that exceeds industry standards.
- Superior scalability for geographic expansion and OEM integration
- Direct delivery to customer distribution centers, eliminating costly autonomous terminals that do not meet customer needs
With an unparalleled development advantage and a product that is scalable, cost-efficient and safe, Waabi is uniquely positioned for driverless deployment in 2025, just four years after our inception and rapid scaling shortly after.
You hired Uber Freight’s CEO, Lior Ron, and raised $200M from Uber, Nvidia and Porsche. How do you balance being both their partner and potential disruptor in the trucking industry?
Our goal has always been to work closely with the entire industry ecosystem, unlocking opportunities and synergies with our various partners. Our technology delivers incredible value to transportation and logistics networks as well as to OEMs so we don’t see being a partner and a disruptor in the industry as being in conflict.
Your digital twins achieve 99.7 percent accuracy for training scenarios, but that last 0.3 percent could be life-or-death edge cases. How do you prove to regulators and the public that simulation-trained trucks are safe enough for highways?
To clarify, 99.7 percent refers to the realism score for our simulator, Waabi World. This astonishingly high degree of realism provides evidence that our simulator faithfully replicates the real-world driving experience. This allows us to simulate any real-world scenario and rigorously test system performance before deployment. Our simulator enhances every single mile we have driven in the real world, enabling us to deploy on the road with greater confidence since we already understand how the system will respond to various situations, including safety-critical citations that will be too costly or unethical to collect in the real world.
In addition, we conduct both mixed-reality testing, where we bridge the gap between simulation and the physical world on a test track, as well as on-road testing to ensure our trucks are safe and ready to handle any situation that might arise.
Want more insights? Join Working Title - our career elevating newsletter and get the future of work delivered weekly.
