Open Navigation

Waabi’s Raquel Urtasun explains why it was the right time to launch an AV technology startup

'This is a combination of my 20-year career in AI as well as more than 10 years in self-driving'

Rebecca Bellan 3 years

Raquel Urtasun, the former chief scientist at Uber ATG, is the founder and CEO of Waabi, an autonomous vehicle startup that came out of stealth mode last week. The Toronto-based company, which will focus on trucking, raised an impressive $83.5 million in a Series A round led by Khosla Ventures. 

Urtasun joined Mobility 2021 to talk about her new venture, the challenges facing the self-driving vehicle industry and how her approach to AI can be used to advance the commercialization of AVs.

AI pioneer Raquel Urtasun launches self-driving technology startup with backing from Khosla, Uber and Aurora

Why did Urtasun decide to found her own company?

Urtasun, who is considered a pioneer in AI, led the R&D efforts as a chief scientist at Uber ATG, which was acquired by Aurora in December. Six months later, we have Waabi. The company’s mission is to take an AI-first approach to solving self-driving technology. 

I left Uber a little bit over three months ago to start this new company, Waabi, with the idea of having a different way of solving self-driving. This is a combination of my 20-year career in AI as well as more than 10 years in self-driving. Thinking about a new company was something that was always in my head. And the more that I was in the industry, the more that I started thinking about going away from the traditional approach and trying to have a diverse view of how to solve self-driving was actually the way to go. So that’s why I decided to do this company. (Timestamp: 1:21)

The limiting factors of large-scale commercial AV deployments

The self-driving vehicle industry hasn’t yet unlocked the code to full autonomy. Some companies, like Cruise and Waymo, have set out small-scale robotaxi deployments, but the tech behind the model isn’t yet strong enough to scale. 

Commercial deployment is very simple operational design domains. One of the reasons for this is the utilization of what they call a traditional approach, which hasn’t really utilized the full power of AI in order to solve such a difficult task as self-driving and realizing complex manual tuning. So this makes the scaling of this technology, and particularly handling the long tails of the scenarios and things that might happen when you drive, particularly difficult. (Timestamp: 2:55)

Finding the sweet spot between deep neural networks and rules-based AI

Urtasun’s approach to autonomy involves a combination of deep nets and rules-based AI, the latter of which is the general approach for self-driving vehicle development. That’s in large part because when we’re talking about transporting humans in an autonomous vehicle, developers tend to shy away from deep neural networks because of the associated black box effect, wherein developers are unable to verify or validate why the system did what it did.

On one side, you have this more traditional approach, where AI is used to solve small problems within this more complex system. And on the other hand, you have this AI black box approach. So Waabi is doing something that really takes advantage of these two approaches, but without the inherent disadvantages. So we have a new generation of power grids that combine deep learning with probabilistic inference and complex optimization and provide us with three particular characteristics that are important. It’s end-to-end trainable, so the system can actually learn the entire software stack from data. It produces interpretable representations so we can explain why the system decides to do a specific maneuver. And at the same time, the AI system is able to do very complex reasoning for this probabilistic inference on complex optimization. So you can think of it as really the best of both worlds. (Timestamp: 4:17)

Using Waabi’s high-tech simulator to reduce road testing

Since the dawn of autonomous driving, putting vehicles out onto the road to test how they function in real-life circumstances has been part and parcel of reaching higher self-driving levels. Navigating both the roads and the law has been a cornerstone for a lot of startups to be able to develop and scale up, but Waabi is also relying on simulating road conditions to test its vehicles.

We have a breakthrough simulator, which is a closed-loop simulator that enables us to test the entire software stack. As a consequence, we can simulate all the generic scenarios as well as all the edge cases, and we can test that at scale, as well as train the AI system in simulation. However, it’s always important that you also test the system in the real world, not just in the simulator, to understand whether there are any gaps between how it performs in simulation versus the real world. Now, this new simulator has super, super high fidelity; it’s in real time, and it has enabled us to really bridge that gap. And, you know, when we say that works well in the simulation, we can mathematically say that it will actually work similarly in the real world, which is a game-changer that really enabled us to reduce the amount of miles that we need to drive on the road, and, therefore, develop this technology in a safer manner. (Timestamp: 5:40)

Could partnering with other companies create conflict?

Uber and Aurora are both investors in Waabi, and Aurora specifically is going after autonomous long-haul trucking, which is the route Waabi is taking with its self-driving tech stack. Might this present competition or a conflict of interest for the young company?

Ultimately, we all want to solve self-driving, and long-haul trucking makes a lot of sense from an application perspective. Why? Because of the driver shortage that we have, as well as the need for safety. And at the same time, from a technical perspective, driving on highways is simpler than driving in our cities, where there is still a pretty complex problem to solve. (Timestamp: 7:26)

It’s a hard problem to solve that Urtasun believes requires collaboration from different industries. 

We plan to have a very open partnership, a friendly approach. And in particular, in working with, for example, OEMs, hardware providers, sensor providers, compute providers, because there have been amazing developments in the R&D space on those fronts. And we want to capitalize on that. (Timestamp: 8:12)

Can diversity of thought thrive in a consolidating industry?

As the autonomous driving industry grows, it also consolidates, with a few major players acquiring promising new ones. Though they may have fresh ideas, newer startups don’t always have the resources to scale commercially, which could hinder innovation and progress. 

When you have a very capital-intensive approach, consolidation makes sense, but not everything is good with consolidation. One of the things is that with the cross-pollination of people and ideas, the industry sort of goes in one approach. We need a diversity of approaches to solve something as difficult as what we’re trying to do here. So I think it is the perfect time, actually, for new companies to come to life and, in particular, with the luxury of starting with an angle or a view on the technology from 2021 with no baggage of anything that we saw before. (Timestamp: 9:12)

You can read the entire transcript here.

Related sessions from TechCrunch Mobility 2021:

Check out all the Sessions from Justice 2021.