Raquel Urtasun, scientist, founder and CEO of autonomous vehicle technology company Waabi, launched her company in June 2021, a time when it seemed like the AV industry was consolidating.
Urtasun and her team of 40 in Toronto and California came out the gate swinging with an $83.5 million raise from a series of high-profile investors, including Uber, Aurora and Khosla Ventures.
Waabi uses an AI-first approach to commercialize autonomous freight faster and more efficiently than its competitors, Urtasun told TechCrunch. As a professor in the Department of Computer Science at the University of Toronto, a co-founder of the Vector Institute for AI, and former chief scientist at Uber ATG, the self-driving unit Uber sold to Aurora, she has acquired some insights into both the industry and the science backing it up. After all, despite consolidation and gains from a few major players, no one had really figured it out yet.
So what does an AI-first approach really look like?
In February 2022, Waabi launched Waabi World, a high-fidelity closed-loop simulator that doesn’t just virtually test Waabi’s self-driving software, but can also teach it how to drive. Waabi World automatically builds digital twins of the world from data, performing near-real-time sensor simulation, manufacturing scenarios to stress-test the Waabi Driver, and teaching the driver to learn from its mistakes without human intervention. This, Urtasun said, saves countless hours of human labor to train the Waabi Driver both in simulation and on the roads.
The entirety of Waabi World is powered by AI in a way that other companies’ simulators aren’t because it relies more heavily on deep neural nets, AI algorithms that allow the computer to learn by using a series of connected networks to identify patterns in data. Historically, developers haven’t been able to figure out the how and why behind an AI’s decision-making when using deep neural nets, which is very important when putting self-driving vehicles on public roads, so they’ve fallen back on machine learning and rules-based algorithms to tie into a broader system.
Urtasun said she’s found a way to solve the problem of the “black box” effect behind deep neural nets by combining them with probabilistic inference and complex optimization. The result? The developer can trace back the decision process of the AI system and incorporate prior knowledge so they don’t have to teach the AI system everything from the beginning again.
We sat down with Urtasun to discuss the pros and cons of starting a business after working for a larger company, the surprises of being a founder and why freight will be the first AV industry to commercialize at scale.
The following interview, part of an ongoing series with founders who are building transportation companies, has been edited for length and clarity.
After working for Uber and being an academic, what are your takeaways about what it’s like being a first-time founder?
When I decided to start Waabi, I didn’t necessarily know what being a founder meant. I’ve been working in industry and in this field and whatnot, but as a founder, you need to wear so many hats and there is so much going on. I didn’t expect that. And Waabi now is very different from what it was to start with, so there’s something that surprised me.
But it’s been an incredible ride. I have to say there is nothing like building what you really believe in with a team that you love to work with. There is nothing that can’t be done.
You’re wearing many hats now, but how was it working under someone at Uber and not being in control of the whole show, in comparison?
I was part of the executive team at Uber, so I had a lot of impact and, you know, a lot of say in many things. But it’s different when you’re building — and this is not just Uber, this is in general. If you are in a large company with 1,000-plus people who are going in one direction, even if you all agree that you need to steer to something else, it is so difficult and slow to actually do this process.
From that point of view, being in a startup, which is much more dynamic, is very exciting, but it’s not Uber versus not Uber. I think any big company would be similar. But I had a great time at Uber. I learned so many things and really discovered what it meant to really be part of a big problem, and really prepared me really well for what I’m doing today.
It’s different in so many ways, like, for example, I didn’t need to fundraise before, to driving the technical road map of the full organization, to partnerships and other things that we need to do. It’s been about trying to understand what do we need to do and when. That hasn’t been necessarily easy, and it’s a very different type of work. Before I was more focused on the technical problem. Now it’s about building the entire ecosystem around a technical problem.
I’ve heard that a bit from founders who’ve graduated from bigger companies that there’s the sense of learning a lot there, but feeling a bit unable to charge forward.
We did many generations of technology that went into production, but yes, there is nothing like going straight to where you want to go with your team.
Uber’s also a startup. How have you structured your company – is it the same or different at all?
In building this startup, what was very important was the culture. One of the things that I’ve loved about my teams in the past was that they really felt like a family, and I wanted to make that the core of Waabi, of our values. This respect for each other, we’re building this together, best ideas win. In our differences, we are all better.
Going from founder to Uber AI scientist to founder – what’s the biggest surprise you’ve had?
I’ve really had to step out of my comfort zone. I’ve spent so many years in AI, as an academic, and it’s been very successful, to the point that things were almost a given. And then suddenly, there was this new world that I didn’t know about, and I had to learn very, very quickly. So that definitely stretched me in many ways.
Everything went really well; we had a big fundraise in two months, but there is still this stress of, “Am I doing the right thing?” One of the things that really defines me, not just as a founder, but in general, is that I introspect everything I do all the time, and everything that the team does all the time, to always be better. It’s not about other people or the industry. It’s about, how can we be better? How can we move faster toward our goal?
How do you do that exactly? Do you have a structure that helps you evaluate progress that you find to be useful?
I am very strategic. Having a very clear technological road map aligned with a business road map for the next few years has been very useful. And then focusing on delivery, delivery, delivery, and trying to anticipate if there are any hiccups or anything along the way so that you can really progress.
I have always been very results-driven from my research and a very focused person, and so that piece has been useful to you being in the industry, as well.
What do you think is one of the biggest differences between the startup world and the academic world?
I think the one thing that is useful as an academic is that we are always stretched very thin. You need to wear many hats as an academic, as well. Oftentimes, in academia, you need to fundraise for grants and to pay your students or postdocs. You need to have a strategy. External presence is important. There are a few traits that are kind of common, perhaps once you join a big company it is very different, right? All that structure is given to you.
In academia, you’re looking for, “What is the toughest problem to solve and how can I solve that one problem?” In industry, this is not what you are after. You want to understand what is the product that people will actually really want to use, and what is the safest, fastest path toward it? And those are totally separate mindsets.
You’re still a professor at the University of Toronto. Do you have any work-life balance?
Work-life balance. I don’t know it very well.
Oftentimes people say that for researchers, university is a lifestyle — it’s more than just a job. So for me, solving this particular problem is more than just a job. It’s an obsession, and I have all the support I need to actually really push this passion to the next level. But work-life balance is definitely not my forte. But it’s a choice I take. For me, I believe self-driving is gonna happen, and I really want to be building that solution. So this isn’t the time to take shortcuts right now. Just focus on the solution.
Do you think that there’s anything you do outside of work that helps you stay focused at work?
Physical activity is very important. I love sports. They taught me many things. I used to play team sports, which taught me the importance of the team but also it taught me to always improve myself. At the same time, a healthy body is a healthy mind.
I love traveling, with the pandemic it’s hard. And I like to be with my significant other.
You raised about $80 million last year for your Series A. Are you raising more now?
I guess we raised a big quantity, but it was really a seed round because it was the first fundraiser we did, such that we could actually focus on the delivery of the technology we wanted. So we don’t necessarily need to raise right now. There is a lot of interest, so I’m not saying no, but I’m not necessarily actively fundraising either.
How do you think about investment?
Oftentimes people think about investment as just, “I want money in the bank.” I think that’s the wrong thought. It’s more about who do you want as partners for a long time that can really help you to be the best company? So it’s very important who you pick.
You want to take a diversity of investors that can help you on many different fronts. From that point of view, our investors are very strategically picked to give us deep tech and commercialization from, for example, Khosla. And then tapping into the Canadian ecosystem since we are a Canadian company, as well as the logistics bit with, for example, Uber, which potentially could be a partner in the future.
The investors in general really understood why there is a need for a different player like Waabi, and why Waabi is that player. There is no one else that is a Waabi.
Autonomous vehicles are obviously a big space right now. What do you think it says about a company that is fundraising but hasn’t announced a raise in a while?
When companies have issues fundraising it’s because it’s very hard to articulate why they are different, and why is this a good investment versus somebody else. I think if you have a clear differentiator, you have a great team to execute and the right road map, then you’re starting in a good spot.
More Transportation Founders
Waabi is kind of a new player to the space. Do you think there’s room for more AV players still to come out? Or do you think we’re moving more toward consolidation?
The industry was moving toward consolidation, in terms of the technology, because there is broad cross-pollination of ideas and people that move around these companies. But I think in a domain like self-driving, where the field has been entered into, in my opinion, the wrong place, I think there is room for new players. And I would be curious to see whether the Waabi story is gonna make more people actually try out.
Oftentimes people ask me, “Do you think you’re too late?” And the answer is absolutely not; I think it is the right time. Because we have really learned what works and what doesn’t work, and we learned that there is a need for something else. And if you know what that something else is, then you can really execute much faster. We showed you the Waabi simulator and how quick you can actually get to the next level, and we look forward to showcasing that also with our self-driving trucks.
Are there areas where you think more startups should be coming online in the AV industry in general?
We see a lot of things on sensors, we see a lot of things on compute. It is hard to partition the software into pieces that you can make a startup out of, right? Because the entry point to self-driving is high. It’s not as if you can do a little software run and then just sell it and life is good. There are a lot of complexities in the system. So it’s just very hard in the sense that it’s a single system and there are so many things built together to make it work.
Is it even fair to ask the question of what’s harder to solve – the hardware or software?
I think they’re very intertwined. There is still quite a bit to be done, for example, on the sensing and compute front. Now we’re seeing great road maps by many startups, and if those are going to be ready on schedule then we are in a great spot.
I think the software right now is a very hard problem to solve. And I think that people have underestimated, either publicly or in private, the difficulty of this. There is definitely a need for new approaches and new blood on this.
Why did you choose freight?
The three canonical applications are last-mile delivery, whether it is food or groceries, robotaxis and freight, and for me, it is very clear that the first commercial application of this technology at scale is going to be freight. And this is because there is an absolute need for drivers. The chronic shortage is just growing and the pandemic has made things even worse. Truck driving is one of the most dangerous professions in North America.
At the same time, highway driving is much easier than driving in cities, like you would with robotaxis. And you’re picking and dropping off people, which can be dangerous. With last-mile delivery, how do you deliver to the door and go up the stairs? There are more trivialities to these products. For me, freight is a very clear product where the market is in absolute need and the technology is much simpler.
Some companies like Waymo are using a transfer hub model that breaks down the journey into autonomous on the roads and then human-driven to get to the last mile. How are you thinking about tackling freight?
I’m a big believer of cut the problem into pieces, solve the pieces that you can and that are important from a business perspective, and then expand. You shouldn’t try to be everywhere but look for an operational domain where you can really solve this and can have a lot of freight, and then expand from there.
What’s the most important piece that needs to be solved first?
The most important piece is proving the safety of a system so that you can take the driver out. But this driver-out has to be out all the time. Not just a one-day demo, right? And the question is, can you find that operation domain where you can do that? And then expand from there and prove safety on that operation.
When you begin operations, are you focusing on the long-haul, the middle-haul? Where do you see the biggest opportunity for initial commercialization?
I like long haul, particularly from the perspective of the truckers. Young people don’t want to be long-haul truckers because you basically give up having a family down the road. And from the automation perspective, these corridors are easier to automate.
Final question: Where do you think Waabi will be in a year?
I think we will not be talking about, “Will Waabi do a fleet versus not?” The conversation will be, “When is Waabi deploying?” Maybe we can take a bet. Let’s see whether we will have that conversation a year from today.