Should Self-Driving Be Rethought?

Kara Miller:

Welcome to Instigators of Change, a Khosla Ventures podcast, where we take a look at innovative ideas, the people who come up with them, and those who invest in them.

I'm Kara Miller, and this week, you know how self-driving cars were going to change everything 10 years ago? What the heck happened? Well, today two experts weigh in: an investor who was once an inventor.

Sven Strohband:

We were also wrong in what we thought could be done in a decade. It's a hard and thorny problem that has an issue with the remaining corner cases.

Kara Miller:

Then, we'll hear from the CEO of the self-driving car company, Waabi, who talks about the immense problem she thinks needs solving.

Raquel Urtasun:

I think there is a big opportunity to have a big impact in safety. There is two million people that die every year on the roads. I think that people expect perfection from robots often times, so I think it's a matter of really educating the population.

Kara Miller:

How to make the dream of self-driving a reality? That's coming up on today's show.

Kara Miller:

Okay, so rewind for a second here. It's 2004 out in the Mojave Desert, roughly somewhere between L.A. and Las Vegas, and a bunch of robotic cars have just proved they weren't quite ready for prime time. DARPA, the Defense Advanced Research Projects Agency, which is basically where the military makes big bets, they realize their self-driving car challenge has been a flop, but they're not willing to give up.

Speaker:

Welcome to the DARPA grand challenge. The objective? Create a self-navigating autonomous vehicle that can safely drive 132 miles through a harsh desert environment without human intervention.

Kara Miller:

The next year, 2005, DARPA is back pushing teams of young techies to answer the question of what self-driving cars can actually do.

Speaker:

The technology we see here will be used in the military within the next few years and, eventually, robots like these may be taking you to and from work. But it's not easy. 43 teams from around the country began with dreams of building a robot capable of conquering this course. Only 23 teams earn the right to run in the $2 million DARPA Grand Challenge final.

Kara Miller:

Carnegie Mellon University had been the most successful entrant in 2004 and they're back, but so are lots of others. Nearly 200 teams applied to be part of the challenge and a couple dozen finalists make it to the ultimate race.

Sven Strohband:

We formed a team that we very euphemistically called the Stanford Racing Team. It was a small group of folks that wanted to build a vehicle, and I was the lead engineer for that team responsible for a vehicle called Stanley, which was our entry in this race.

Kara Miller:

That's Sven Strohband, who's now a partner and managing director at Khosla Ventures.

Sven Strohband:

I'll be the first one to say that Stanley was quite primitive, relative to what I would call a modern self-driving car.

Kara Miller:

The teams were told at about four in the morning on the day of the race exactly where the cars would travel on this desert course, and in the darkness, they prepped their vehicles and their code. Then, the front runners, Highlander and Sandstorm from Carnegie Mellon, and was Stanley from Stanford, set out.

Speaker:

To finish the grand challenge, robots must travel on tight desert roads, over dry lake beds, through clouds of dust, under three tunnels, and finally at the 123 mile mark, through the toughest part of the course: beer bottle pass, a narrow, winding road with a mountain wall on one side and a 200-foot cliff on the other. Early on, the big three showed why they were the favorites to win this race.

Kara Miller:

As time went on, Cornell's car crashed, CalTech went up over a barrier, and one by one, UCLA, Princeton, Virginia Tech, Ohio State, they all fell out of the race.

Speaker:

With five miles to go, Stanley steps on the gas racing toward the finish line on the dusty desert trail. As Stanley crosses the finish line, the Stanford racing team has made its way into the history books.

Sven Strohband:

Then, Stanley ended up in the Smithsonian Museum, and it was one of the early times, I would say, when you had an autonomous vehicle, pushed a button, and it left early in the morning and came back sometime in the afternoon. There was no driver in the car in this off-road race, and that was a really interesting moment for everybody on the team.

Kara Miller:

Strohband, along with a couple dozen teammates and project lead Sebastian Thrun, they understood they were part of a big shift in transportation. Here is Thrun on the day that Stanford won.

Sebastian Thrun:

I didn't anticipate this to become a race for speed. It was one of the most thrilling races ever. We've been trailing red team for a long time, actually for three hours it was kind of clear to me we would actually come second. And all of a sudden, a turn of events, a little thing had happened, a little glitch. All of a sudden, we got the lead and came in ahead of them. But the most important thing for me is it actually doesn't matter who comes first. It matters that we as a community achieve it.

Kara Miller:

The question, though, of what that race achieved is still up for grabs. Strohband on has kept studying transportation and investing in it, but for now, most of the cars driving past you on the road, they are not self-driving. We, or at least I, am not able to sit in the back answering emails while some very competent AI driver chauffeurs me around. Today, a couple of conversations that are mostly about transportation, though, you'll see will veer off a little bit. To start, I asked Strohband, have things moved as quickly as you had hoped back when you competed at the DARPA challenge 17 years ago?

Sven Strohband:

It has absolutely moved slowly, but to be fair, that was an off-road race under relatively defined conditions, and driving, let's say in Manhattan, that's a very different technological challenge to accomplish. But yes, I think there was even a decent set of folks on our team where we had an internal bet that if we meet in 10 years, how many of us will still be driving to this meeting? It was about 50/50, if I remember correctly, because 10 years seemed such a long time to actually solve the problem. It turns out we did meet 10 years later and everybody drove him or herself, so we were also wrong in what we thought could be done in a decade. It's a hard and thorny problem that has an issue with the remaining corner cases. I now have a startup [inaudible 00:06:59] that's actually addressing some of those things, but yes, I thought it was disappointing that we don't already have reasonably widespread use of autonomous vehicles.

Kara Miller:

I remember a few years ago talking to the robotics pioneer, Rodney Brooks, and he talked about those edge cases and how... I think his concern was that the reality is that the media attention to any mistake that a vehicle that's driven by a computer, the attention that will get from the media is substantial. Versus, of course, people are in car accidents all the time, and that doesn't get that much attention. I wonder what you think of that issue, of the autonomous driver may make few mistakes, but the ones that are made get a lot of spotlight.

Sven Strohband:

That's certainly true, and I think that's true with all next-generation technology. It has a little bit more of an eye on it, but I think there's also a fundamental technical issue that I think lots of people underestimated. The fundamental technical issue is that if I built a self-driving car with all of its components, the sensors, the motion planner, the perception stack, and all of that, and I now would like to incorporate yet one more corner case into it. The further I am and the development, the more of my code and of my design this corner case touches, so I get slower and slower solving these corner cases, the more sophisticated and the more code I actually have. This slow down is a genuine problem because you want to clobber down of those corner cases as efficiently and as fast as possible, but the larger the code base gets, the more code these corner cases touch, so that becomes an issue.

I think there's a second one, which is, when do you know you're safe enough? Typically the answer is, "Well, we just drive millions and millions of miles." Well, that's not quite the solution because it turns out a mile is not really equal to a mile. For example, a mile of highway driving might be actually quite trivial. A mile driving down Lombard Street in San Francisco might actually be a tricky thing. Or driving in Manhattan is an even trickier thing. A mile's not really equal to a mile, so creating lots and lots of test cases for these vehicles extremely efficiently, instead of having to wait for such a case to naturally occur in a fleet, that's another issue that has to be solved here.

Kara Miller:

You mentioned one of your investments at Waabi, and we're going to talk to their CEO in a few minutes, but they're trying to figure out fundamentally, how do you think about these things, all these things that can go wrong on the road? You've got people crossing the street, animals can be darting in here and there, rain, snow, and the permutations of all of this are endless. So, when you think about that bet you made, who will come back here in their own car, who will come back here in a self-driving car, if we were to make that bet again now, I don't know. What do you think?

Sven Strohband:

Yeah, that's exactly the bet that I made with Waabi.

Kara Miller:

Okay.

Sven Strohband:

The approach is basically a fundamentally different approach, and it's approach that is rooted in having very, very good simulation and the capability to optimize the entire stack for a new corner case automatically. We are now in a position, and that wasn't really true back then, where we can have extremely good simulation and every self-driving car company has some simulation efforts, but here simulation is what I would call a first-class citizen. You record with your sensor stack what the car sees, but then you can replace cars, add pedestrians, change the sensor arrangement, and you can do all of that virtually. That has never happened, and you can run that in this simulation. So, you can create lots and lots of corner cases very, very quickly, and you can have your virtual driver run through all of those corner cases, and you can see the driver perform without having to have hundreds of thousands of cars on the road to collect enough corner cases. That's a fundamentally different way to look at the problem.

Kara Miller:

Just to be clear, so if we meet in 10 years, you're going to be using a self-driving car to get to that meeting, or...?

Sven Strohband:

I really hope so.

Kara Miller:

Okay. Okay. Another question of a different kind of transportation: you have made also investments in things having to do outer space. Rocket Lab, Varda, things that feel like the next stage in connecting people and Earth to space. What does that next stage feel like to you? I mean, to a lot of people, it feels like, "Wow, it's hard to imagine what that connection is like," but what is it like in your mind?

Sven Strohband:

I think in terms of human capabilities. We now have the capability to make relatively small satellites that have astonishing performance, and a lot of that has to do with miniaturization that has helped other industries now making its way into space. Rocket Lab enables them to fly those satellites with high precision, to the inclination that they want to, the orbit that they want. That's a fundamentally interesting capability because satellites on the ground make $0 and are kind of useless, so you needed to solve this... The Uber ride, so to say, to space. Rocket Lab successfully solved that, and then added capability to make the building of the actual satellite easier and easier. Let's say you built an Earth observation` satellite, and you're an expert in the sensor stack that actually records what's happening on Earth.

Before you can make use of that, you A, need to ride to space, and we talked about that earlier, but in addition to that, you need to build a satellite. That means you need to have solar panels, and you need to have star trackers, and you need to have reaction wheels, and you need to have the software that runs on the satellite, and a gazillion other things. Rocket Lab has basically made that entire process very easy now, because we can supply all of that to you. We can basically build the entire bus so that you can focus on the thing that is truly, truly distinguished, which is your sensor stack, your analysis of the data, all of that. That's a new capability that opens up space to folks that before, had no hope, really, having birds in orbit that do something useful, because they needed to be expert in a lot of different things and then find a ride that was highly restricted. Now, these barriers have been removed and I can't really wait what people all build here.

Kara Miller:

We talked a little bit about AI before, and then how it intersects with transportation, and I want to get back to that. But if we just back up, you have invested in lots of kinds of technology companies over the years, worked on lots of tech yourself. When you think very big picture now about the most exciting aspects of technology that get talked about to you, what do you feel like they are right now?

Sven Strohband:

I think some of the most astonishing things to me come, really, in two buckets.

Kara Miller:

Okay.

Sven Strohband:

Bucket one is AI. AI has a lot of hype associated with it and we already had an AI winter, so having too much hype is a bad thing, and I think the practitioners really are genuinely aware of this. But when I take a look at the capability of deep foundation models, now it is truly, truly astonishing. We can write text. We are starting to reason about problems that humans find hard to solve. For example, in mathematics, we are now capable of generating images by just describing the images and the AI creates the image. So, I think that is an entire area that I find absolutely astonishing, and I think some of the hype was genuinely warranted here. Some of it is silly, but there's definitely something there.

Sven Strohband:

The second bucket that I find fascinating now is that there was technology that was regarded as only being solvable by large governments actually solving them. For example, when we wanted to go to the moon, building a rocket was not something that a private enterprise would really attempt, but now that is a thing that is within the realm of a venture-funded startup that can actually do it. Another example is building a plane that flies mach five, or a drone that flies mach five, that would have been solidly within a set of defense contractors and lots of government funding. It is now a thing that a startup can undertake, and all of that has really a lot to do that our tools have gotten a lot better. That means the cost of entry here has dropped quite a bit. There's still issues with all of that, but I like deep tech investments and that often involves atoms as well, and now the bar to be in that space is considerably lower and it is accessible to startups. I find that bit also very exciting.

Kara Miller:

It's not just that there's a ton of money washing around. It sounds like it's more that, whereas during the NASA program you had... I think the government spent 2% of its money to try to get us to space. Some very large percentage. It sounds like it's just cheaper to do really big things.

Sven Strohband:

Yeah. We get to stand on the shoulders of giant, as one says so nicely. One is we know a lot more about it, the other part is our tools have gotten a lot better. I'll give you a concrete example. Going from a CAD design to verify the design in, let's say, computational fluid dynamics, so you can actually figure out if it does what you think it would do, to then having the part as a metal part, let's say, in a lightweight material like titanium. Well, that used to be a really, really big deal, where you might not even have been able to do the CFD back then, and if you wanted the part, it was an act of having highly talented machinists make this part and all of that.

Now, we can go from a CAD model, run the CFD, maybe even on your laptop, for sure in a data center relatively easily, and we can 3D print a lot. For example, Rocket Lab's engines are largely 3D-printed, and that also works in materials like inconel or titanium that are relevant to the problem. So, we really do stand on the shoulders of giant.

Kara Miller:

I'm sure you're pitched ideas all the time. Do you feel like there is an area where you think, "I wish more people were working in this area?" This deserves more attention, or this technology. I feel like there's a gap here. I just wonder if there's anything like that occurs to you.

Sven Strohband:

Yeah. At Khosla, we have a couple of areas that we think about as areas of high impact. One of the thing that is universally shared in our shop, and definitely by me, is the urgency of climate change and what one can actually do about it. I know there's lots of people working in it, but man, we have a lot to do. So, we invested, for example, in a fusion company, and so we are trying things that might sound a little bit crazy, very much like a rocket company sounded a little bit crazy seven years ago, to try to see if technology can make a fundamental dent into the problem. That's an area where there's a lot more that can be done, be it on the carbon-capture side, be it on the energy generation or distribution side, that I think is, let's say, a passion area.

Kara Miller:

Yeah. Yeah. Let me ask you, finally then, I think that's a good lead in to this; climate change is one of those big, big, big things. I know you think some about business and government and working with each other and partnering with each other, and some things are so big that partnerships are really quite important. Give me a sense of how you see that intersection, businesses and government. What is the state of it? Is there enough guidance for businesses and how to create a partnership?

Sven Strohband:

I think there's a big irony here. Silicon Valley got really started with a lot of interaction between the US government and companies. Then, we forgot about this, and we had the semiconductor revolution, and then we had software, but that was really the origin of Silicon Valley. I now see a different trend where two things have happened: one is the government has gotten better at being open to startups, and I think that's a trend that has started probably something like three, four years ago, and I find very encouraging. The second part is startups have gotten more used to the idea that they can have the government as a reliable customer. Now, I will say this is still in its infancy a little bit here. Most of the government contracts are still within defense primes and so on, but there's very, very interesting technology that gets early adopted in startups, and giving the US government capability in those areas is becoming a more and more interesting thing, also, for the US government. So, I actually see both of those things moving a little bit closer together again.

Kara Miller:

Sven Strohband is a partner and managing director at Khosla Ventures. Sven, thanks so much for your time.

Sven Strohband:

My pleasure. Thank you.

Kara Miller:

Before we move on here and talk to the CEO of a self-driving car company, that topic we just touched on, government regulation and how founders should work with regulators, it's going to be the focus of a deep dive for us next week, when we'll be joined by the former head of the FAA, Michael Huerta. He offers insight into how founders can work effectively with the government, or sometimes, not so effectively.

Michael Huerta:

What you don't want to do is walk in, as some companies have sometimes approached regulators, with a whole army of lawyers sitting down and telling the regulators, "Here are all the reasons that your rules don't apply to us."

Kara Miller:

Stay tuned for that. In a moment, I'll be joined by Raquel Urtasun, CEO of Waabi, but first a message from Khosla Ventures.

Speaker:

Lots of people are thinking about making a career change right now. If you are one of them, take a look at one of the companies in the Khosla Ventures portfolio. KV companies seek to fundamentally change how industries work, from health, finance, future of work, to transportation, energy, and even space. Check out KhoslaVentures.com/jobs. That's KhoslaVenturess.com/jobs. Now, back to Instigators of Change.

Kara Miller:

Raquel Urtasun son has spent years fine-tuning the technology behind self-driving cars.

Raquel Urtasun:

Unfortunately, I know too many people that have had severe traffic accidents. I really want to change that, and provide safety to our roads.

Kara Miller:

She's the CEO of Waabi, a company rooted in AI. We talked about some of what the company's attempting to do with Sven Strohband, but as he suggested, the road to self-driving has been bumpy.

Raquel Urtasun:

Things are very, very far from where we would like it to be. In my opinion, it's really due to the technology that is employed to try to solve this very complex task. In particular, it's a technology that requires a lot of manual tuning, it requires a lot of human interaction, it requires a lot of driving millions of miles on the road to understand mistakes, et cetera. That's doesn't scale, so there is a need for something else, and I think that something else is what is going to bring, really, the solution that we can commercialize at the scale.

Kara Miller:

Urtasun is a professor of computer science at the University of Toronto, and from 2017 to 2021, she ran research and development for Uber Advanced Technologies Group, which works on self-driving cars. She says she's come to believe that there's a fundamental flaw in the technology on which self-driving is being built, which is why there's not much self-driving going on on the roads. The problem, as Strohband noted, is that the weird things that happen to you on the road, they don't tend to happen often. Most drives are pretty normal.

Raquel Urtasun:

You can think of it as, human drivers can be pretty erratic and you need to handle them. There might be debris on the road that have shape that you have never seen before. There might be weather conditions that make it actually quite difficult for some of the current sensors to see well. There might be combinations of many things that you have never seen before, but together form scenarios that are difficult for a particular autonomous system to handle. So, there's potentially many, many of these things, and the problem is really into, you have to handle all of them.

Kara Miller:

I have certainly skidded during winter weather, for example, but only a few times during the decades that I've been driving. So, the traditional practice often employed by folks who work on self-driving cars of sending out a car to drive back and forth across the country, that may or may not encounter some of the strange weather that I've seen. Urtasun became convinced we'd never get from almost there to actually there with self-driving if there wasn't a fundamental shift in our approach.

Raquel Urtasun:

I think if you look at the industry at large, there has been quite a bit of consolidation in terms of the approaches, and this is coming from a cross-pollination of people and ideas. Everybody's going towards a very, I would say, robotic-centric approach, and I think that the way to think about it is that it builds a small pieces of the puzzle, and then somehow you put them together to have the final system.

Kara Miller:

But, she argues, that's not going to work. This isn't a problem to break into small pieces, the pedestrian piece and the bicycle piece and the tree piece. This is a whole system that you have to think about all at once, and AI has to be front and center.

Raquel Urtasun:

It has become more and more clear that we are making, with the piece-by-piece approach, incremental progress, but from that incremental progress to the big step change that we need in order to really solve the self-driving, we are, in my opinion, never going to get there with that approach. It, for me, became clear that we need something else, and it's very clear for me as well, into what is that something else? This AI-first approach. That's why I decided to fund Waabi, to really execute on that vision, and really unleash that power of AI to serve this task.

Kara Miller:

It sounds like one of the approaches had been, "Make sure you send these cars out, these self-driving cars and log all the miles and see what happens." But it sounds like your approach is much more like the video game, simulate tons and tons of potential things that happen. Talk about, why have those two different approaches? Why have the Waabi approach?

Raquel Urtasun:

Yeah, yeah. I guess you're referring to the second big piece of Waabi's approach is we use this next-generation simulation, as well, in order to test as well as train the software stack. If you look at, again, a little bit of the industry in the past couple of years, we are starting to see that they're paying more attention to simulation, but current simulation approaches are very far from really being a substitute for years driving on the road.

Kara Miller:

Okay.

Raquel Urtasun:

The reason why is very similar to the brain technology or approach, is that the simulation approaches that exist, they are actually used to test a small pieces of the entire system. As a consequence, there is no way to really test or understand whether the full system will perform well. As a consequence, the only way to do it is that you need to drive many, many miles in the real world, and why many miles?

Because precisely many of the things are very rare, so in order to even see them once, you need to drive these millions, billions of miles. That's, of course, a lesson in a scale. We need something else, and this something else is really a new generation of simulation, where to this self-driving system, driving on the simulation is exactly the same thing as driving in the real world. If you're able to match these two things, then you can change the equation of how much miles in the real world you need versus how much miles on simulation in order to really advance and potentially deploy your systems. That's what we are doing with Waabi World, which is this next-generation simulation that, indeed, it gives you the ability to really reduce the number of miles of the real world to just be validation, not be the primary way of testing.

Kara Miller:

Does that mean that the idea is that before the technology is deployed, you're simulating, I don't know, a kid running out to get a ball that's bouncing across the street, or deer crossing the street? Or all these different things that happen not very often when you're driving, but in your lifetime, they might happen a few times?

Raquel Urtasun:

That's right. That's right. You simulate both, common scenarios because you want to still make sure that you can handle those, as well as the rare cases. The good thing with the simulator is that you control the world. In the real world, you don't know when these things are going to happen. You can't instrument them. But in simulation you can, and you can test all the safety-critical cases as well, which otherwise will be too dangerous to test in the real world. So, it gives you much more information in a much more efficient way, respect how your system is performing.

Kara Miller:

Talk me through the timeframe and the way in which you think self-driving does break through and becomes a big part of life. Is it trucking first? Is it in five years? Are we talking about 20 years? How do you foresee this happening as a process?

Raquel Urtasun:

Yeah, great question. I'm a big believer that trucking is really where this technology is going to have an impact, in terms of commercialization. This is why Waabi is really full on into self-driving trucks. This is twofold, is on one side, there is a chronic shortage of drivers. It's one of the most dangerous professions in North America. There is a need for increased safety. At the same time, automating highway driving is easier than automating driving in our cities, like I was mentioning, New York City or Toronto where I live. I definitely believe that this is where we are going to see commercialization first, but the thing to note is that commercialization is going to happen in phases. It's not that from one day to the next, you switch and then there is self-driving vehicles everywhere. You will see automation of particular, small geographies, and then it's a matter of who can scale that faster in terms of where you can operate safely

Kara Miller:

Hm. I asked this of Sven before, but I wonder if you worry that accidents that result from self-driving cars, even if there's a lot more accidents from people driving their own cars, that the self-driving accidents are going to get so much attention that people will be worried. Again, there might be 10 or a 100 times as many conventional accidents, but if they don't get the same kind of spotlight, maybe, that this fancy technology gets, people are going to think about them differently.

Raquel Urtasun:

Yeah. I think there is a big opportunity to have a big impact in safety. There is 2 million people that die every year on the roads. I think that people expect perfection from robots often times, so I think it's a matter of really educating the population, working with regulators, building trust, and being more transparent, which is something that this industry has definitely not done on a good job, to being transparent and really showcasing what this technology does, can do, and what are the steps? All the things I need to do in order to get to the next generation. This is something that I'm very passionate about, to really build trust and be transparent about what we are doing at Waabi.

Kara Miller:

I think you said two million people die every year from car accidents. Is that right?

Raquel Urtasun:

Yes. Yeah, something like that.

Kara Miller:

That's a huge number. I think people don't realize that. That's huge.

Raquel Urtasun:

Yeah. Yeah, exactly, and numbers are not getting better, which is despite some of the advances on technologies. There is definitely a lot that we can do. We can save any of those lives. In my opinion, this is a great thing already.

Kara Miller:

Hmm. You've said, this is a quote, "Women still have to accomplish 10 times as much as men to be recognized." That's a lot more. 10 times is a lot more. I would hope we'd be farther along right now, but what's your experience been, as somebody who's started a company, obviously has gone out and tried to raise money, the whole thing?

Raquel Urtasun:

Yeah. Yeah. The world has evolved in a positive direction over the past couple of decades, but I think there is still a lot to be done. Unfortunately there is a lot of episodes of discrimination, of things that shouldn't be there, but they are. It's not just women, it's every minority, I will say, that still suffers from this. But definitely, as a woman in technology, the same as all my colleagues, we always need to do much more to be recognized. It's just, unfortunately, a fact. But what is important is to never give up, regardless of what that is. Just continue pushing and continue being a role model, and then changing that opinion, changing the world, and then making sure that we can mentor these amazing young women that will really take the torch from us and will lead the future. So, continue on changing this.

Kara Miller:

Do you think that it's particularly difficult for women in deep tech, in AI, maybe in areas where there just tends to be a lot of men? Versus maybe, I don't know if biology, for example, might be a little different.

Raquel Urtasun:

Yeah. Yeah. Obviously, some areas where there's definitely not enough, or the representation is very skewed in one direction. In the intersection where I am of robotics, transportation, AI, there's definitely a tremendous misrepresentation of women. But I think that oftentimes, what we see is people in the industry providing just general percentages, but then when it comes to, "Where are your women? Do they have the impact, the leadership opportunities that they should have?" It's very different story, and I think that's where we... It's not about hiring women, it's about how we can really help them to really achieve all their potential, and again, making positive impact. That's where there is a lot more that needs to happen, and the first step is to have a very diverse workforce where everybody's heard, where everybody's opinion matters, and where everybody can be themselves and be in a safe space. That needs to change in this industry, for example.

Kara Miller:

A final question for you, which is... I know you've had a big role at Uber, but how is running a company different from anything you've ever experienced?

Raquel Urtasun:

Yeah. It's definitely the next challenge in my career, I have to say. It's been an incredible ride, so I love multitasking and I think I'm pretty good at it, but this is definitely the next level of this. But I will say that the role of a CEO or founding a company is not easy, but it's incredibly rewarding. In my case, I founded this company with the amazing group of people that I've been working many years with, and there is nothing that can really substitute building the technology you believe in with the people you love to work with.

Kara Miller:

Raquel Urtasun is the founder and CEO of Waabi. Raquel, thanks so much.

Raquel Urtasun:

Thank you for having me here.

Kara Miller:

Next week, as I mentioned earlier, we will build on this conversation and talk more about where transportation goes from here with someone who's been on the inside, Michael Huerta, former head of the FAA during the Obama Administration. We'll also talk about how you work with the government if you're building a company. What should you to do, what should you not do? That's next week on Instigators of Change. Remember, you can pick up our podcast on Apple Podcasts, Spotify, Google. The show is produced by Matt Purdy. I'm Kara Miller. Talk to you next week.