Before founding Aurora, Urmson led the Google engineering team that designed a self-driving car. In 2007, when he was a professor at Carnegie Mellon University, he was a member of one of the earliest groups to build a robot car as part of a DARPA Grand Challenge to develop a self-driving vehicle that could traverse the Mojave Desert. This week, his company announced a new partnership with Chrysler to explore self-driving technology, as well as a new round of investment by Hyundai.
I recently interviewed Urmson for If Then, Slate’s technology podcast. In our discussion, which has been edited and condensed for clarity, we chatted about the barriers to the mass adoption of self-driving cars, whether it’s possible to test the technology safely in real-world conditions, and why the ideal shape for a self-driving car is a cone.
Read or listen to our conversation below, or get the show via Apple Podcasts, Overcast, Spotify, Stitcher, or Google Play.
April Glaser: Let’s start with the news this week, and maybe this will help explain what Aurora is, because it’s a little different than other self-driving car projects. You’re partnering with Chrysler to explore self-driving technology, but you’re not building cars with them. Can you explain what’s going on there, then maybe it’ll help us understand what you do?
Chris Urmson: Our mission is to deliver the benefit to self-driving technology safely, quickly, and broadly, and we do that by working with other companies. Early on we figured out, “let’s go do the things that we’re good at,” and we think that is building the driver. And so we work with a number of companies to bring our technology to market through their technology. In this case, we’ve signed this agreement to go explore with Fiat Chrysler how to bring this technology to market for commercial vehicles.
OK, but you’re not building cars?
No, we don’t build cars. That’s what Fiat Chrysler does, right? They’re really good at this. They’ve been doing it for 100 years, and they understand what it takes, and they’re excellent.
Looking back at the past now, when you started working on this technology 12, 15 years ago, what inspired you to do self-driving cars? Was it the idea that cars are bad and we should rethink them, or were you just fascinated by this cool tech? Why self-driving cars?
For me, I had been working on a project with NASA as part of a team, and we were down in the Atacama Desert in Chile, which is this crazy exciting place, and the robot that I was helping work on there moved at a slow walking pace, so 15 centimeters a second, and it was experimental, and so it didn’t work all the time. My Ph.D. adviser came down and said there was this DARPA Grand Challenge, and the idea was to drive it 50 miles an hour across the desert, and I just thought that sounded cool and like a really interesting problem. That kind of got me hooked, and then here we are, I don’t know, 15, 16 years later.
So it was a really cool idea, and then you started to realize that this could have applications.
Yeah, absolutely. In those early days, the competitions were funded by the Defense Department, and we were thinking about how do we keep our young men and women out of harm’s way in the supply lines, at the time thinking primarily about Iraq and Afghanistan. Then, as we worked on those competitions, and then eventually the urban challenge, I started to realize this technology has profound implications on-road. If you look at the 40,000 Americans that are killed every year on our roadways, 1.3 million people globally, the vast majority of those accidents—95 percent of them—are due to human error, and so the technology we’re developing we think can drive that number to zero over time. That’s a profound opportunity.
Then when you think about the number of people who can’t get around in the same way you or I can, whether they’ve lost the privilege of driving, or they went out for drinks tonight and they shouldn’t be driving home, or maybe they’re blind, or maybe they have other ways that limit their ability to get in cars, giving those people access seems like an incredible opportunity as well.
I hope it doesn’t make traffic worse, but we’ll get into that in a moment. I understand that cars will use computer vision and A.I. to drive and avoid obstacles, but what about navigation. I hope they’re not relying on Google Maps. How will navigation work?
Navigation, I think, will work similar to how it does today in that you get in a car and you figure out where you want to go. The underlying map technology will be a little bit different than Google Maps, because those are maps that have been designed for people to use, whereas when we have the vehicle using that map, it cares about different things.
Will we have to create new maps for each city where these types of vehicles are operating?
I expect so. There’s one layer of map which is for the people to look at, and that tells us where the Safeway is and where the place you’re going is, and you can understand and consume that. For the vehicles themselves, they use a different map, and it’s going to encode things that matter for driving and do that at very high resolution. So, where are the traffic lights, where are the lanes, what is the right of way at this particular intersection?
I’ve read that you think self-driving cars are about five to 10 years away from a small-scale rollout, but 30 to 50 years away from ubiquity, or a very large rollout. What would you say is the holdup at each plateau there?
I think within the next five years we’ll see small-scale deployment. That’ll be a few hundred or a few thousand vehicles. Really this is the, it’s Silicon Valley speak, this is the zero-to-one moment of proving that the technology actually works, understanding how customers want to use it, convincing ourselves that—and when I say ourselves, I mean as a society—that these are sufficiently safe, that we trust them on the roadway, and that’s that first phase. Once we get through that phase, now it’s about how do we bring this to market at a commercial scale. Those partnerships that are so important to us as a company, working with those companies to get the technology integrated the vehicles, bring the price down, and actually start to manufacture and operate them like the vehicles we have today.
It seems like it’s also going to take an enormous amount of investment from other players, maybe even the government. You know the reason why cars reached ubiquity in general is because the government invested in roads. What kind of investment do you think would be required for cities to work coherently with self-driving cars?
I think one of the really interesting parts is that I don’t think we need a whole lot of investment from government in the near term, in the same way that if you look at the way the automobile came into use, we didn’t go out and build a bunch of roads and then hope cars turned up, or at least paved roads. What we had is footpaths, horse paths, cart paths, bicycle paths, and then as the car came into existence we realized, hey, it sucks to drive through the mud, and if only it were a little less muddy, then we could get between cities more easily, and that led to paved roads and eventually the interstate system.
I think we will see the same thing happen with automated vehicles. Fundamentally, roads that work, that are good, and easy for people to drive on, will be good and easy for automated vehicles to drive on, and so just kind of making it a little bit better for people is all we need right now. Then, when the technology actually starts to become scaled, then we can ask the question what have we learned, what are the ways that we can make this a little bit safer, a little bit incrementally more efficient, and that’s what I think local and state governments and federal government would invest in infrastructure.
Recently I’ve imagined that self-driving cars would require like barriers to keep pedestrians and bicyclists away. So you’re saying that type of infrastructure is not going to be needed at the outset?
I think that’s right, and to be clear, that’s certainly one way to solve the problem. And if we had a whole bunch of infrastructure money we wanted to spend, I think that would accelerate a certain class of self-driving vehicle. They don’t have to be quite as smart, because you’re right, if you put fences around them, you keep the people away, you keep the other vehicles away, it’s a much easier problem. In fact, you can see these kind of systems today. If you go to Heathrow Airport in London, there’s a people-mover that will take you from one of the terminals to a parking lot, and it just drives down this concrete barrier driveway.
Our approach, and the way I’ve been thinking about this for a long time, is that that’s too much to ask of a city to go and take the roadways or invest a whole lot to build roads special for these vehicles that don’t exist yet. And what we really need to do is take the technology and adapt it to work the way that we work and live today, and operate on the roads that exist today, because if we don’t do that, then I think this technology, it just won’t happen.
It won’t happen, and also some argue that cars have already been really bad for street-level life, like roads that have cut through cities and deepened segregation, and capped us from living more local or sustainable lives. What are your thoughts with self-driving cars and that? I mean, will they further tear us apart or isolate us?
I hope not. No, no, that certainly a thing I don’t want on my tombstone. But when I think about this, I actually have a much more optimistic view. Look at cities today, and a massive amount of time that people spend in vehicles is really about the way they utilize them. The statistic I heard was 30 percent of traffic in San Francisco is people looking for parking. I heard a more alarming statistic that was 80 percent of traffic in Paris was people looking for parking.
So imagine you have automated vehicles that take you to a location, you hop out, then it just drives down the block and picks up the next person and takes them where they’re going. Suddenly, you’ve alleviated a massive chunk of the congestion in a city. Similarly, if you look at the floor plan of a city today, somewhere between 30–40 percent of cities is dedicated to parking and roads. And so again, if you have automated vehicles operating as a transportation service, whether it’s private or public transportation networks in the city, you don’t need that real estate to be dedicated for parking. That real estate now can be recaptured, and it can be used for park space, it can be used for residential space, yeah, it can be used for mixed residential-commercial office space. There’s a real opportunity to take the heart of our city, which a lot of it is these urine-smelling concrete monstrosities, and turn that into something much more interesting.
Right, so not everybody would necessarily own a self-driving car once they start to roll out in larger scale.
Certainly for urban centers, I think it’s much more likely that this technology is a shared platform that people get on and get off. It’s an even more convenient version of a bus or of a taxi service.
So, ultimately, driving programs are going to have to make decisions about whose safety to prioritize, people inside the car or people outside the car, and I imagine it’s a continuum that might change depending on the context. How are you thinking about this?
This is a societal question, and that ultimately I think there’ll be some guidelines that we as a society come up with on that. I think it’s a little different than personal car ownership. If I buy a car, I want it to protect me. If these vehicles really are operating as a public service as part of the shared road network, I think we have to think a lot about the other folks on the roadway because they’re the ones who didn’t necessarily opt in to using that vehicle.
My personal take is that we need to think most about the vulnerable road users, so the cyclists and the pedestrians out there, and prioritize them first, and then it’s the other vehicles that are moving around because they’re in hard protective cases, right, and I think when you think about it that way, you can shape the problem in a way that makes sense.
It’s true that people in a car are more protected than people outside of a car. How do we balance the need to test these cars in very real-world conditions with the very real danger of doing so? Phoenix [where a pedestrian was struck and killed by one of Uber’s self-driving cars] has been kind of a testing ground, but it scares me to think that yes, we need to test these, but oh, they’re being tested where we are.
“We really need to take the technology and operate on the roads that exist today. If we don’t do that, then it just won’t happen.”
I think it is about having appropriate safeguards and process around it. We’ve been developing this technology we tested on the roads. It’s an essential part of learning, much like as people we get our learner’s permit first, and we’re expected to go out, and there’s constraints on how we operate. The same is true with this technology. It’s the only way we can gather the data we need to have confidence that we’re building something that’ll operate, and so for us, it comes down to having carefully trained and educated drivers. Our drivers go through a multiweek process of first observing, and then working in the right seat where they’re observing the technology, and then ultimately in the driver’s seat.
In our case, these are actually employees of the company. We think that helps make sure they’re even more invested in the outcome than contract employees that others might use. So really, it’s education policy and then a safety-first culture. It’s not coincidental that when I rattled off the company’s mission earlier about delivering the benefits of self-driving technology safely, quickly, and broadly, that safely is the first of those for us.
Is there a particular type of car or shape of car that’s best suited for self-driving technology? Is smaller better, or vans?
The driver we’re building we think works across all of them. In the same way that I can drive my car home, or when I land somewhere and rent a car, I can drive that, we think that’s a really important property of the system we’re building. Yes, I do think if there is kind of the optimal self-drivable car, it probably looks like a cone so that you can see all the way around it really nearby.
That’s what I want to know. Like in a dream world, it would be like a cone that you could see through?
Well, I think if you were building a car just to be self-driven, which I don’t think is a particularly interesting car, then I think yes, it probably looks like a cone so you can see right next to it, but in practice, our company isn’t about building technology for technology’s sake. It’s about building technology that can go help and serve people, and so for us, it’ll look like a vehicle that’s got a comfortable interior where people can rest or work on their way to where they’re going, or it’ll look like a vehicle that has multiple seats in it so that people can share rides and have a much more useful public transportation infrastructure, or it might be a package van, or it might be an electric 18-wheeler that’s moving goods between hubs. So it’s really about the purpose of the vehicle.
This might be a very obvious and possibly annoying question, but it’s something that kind of bugs me. Why is the focus on self-driving cars? Why not better buses, or trains, or something that could carry more people? Because cars are not that great in many ways.
Yeah, I think that’s a very reasonable question, and it’s how do we actually make moving people around safer and more efficient? Buses are great, trains are great when there really is a concentration of where you want to go from and go to, but it turns out we don’t live our lives mostly like that. We live our lives going from my house to my office, and there’s exactly one person who wants to make that trip every day. From my neighborhood to my office, maybe there’s a half-dozen people who want to make that trip, and so I think it’s really about finding the right platform, the right vehicle, that optimizes how many trips we’re making so that it can be efficient.
Where you’ve got a 50-person bus at 2 in the morning, and there’s a driver and one passenger, that’s actually incredibly inefficient. And similarly, one of the challenges with transit routes is where do you put the route, where does the bus drive. That’s kind of a gerrymandered process. It’s also suboptimal for almost everybody, whereas if we could have, say, a four- or five-person vehicle that’s automated, because now we can actually afford to have a four- or five-person transit vehicle that can actually go point to point to point, and then pull those people and take them to where they want to go if it’s nearby, that now starts to become a really compelling transit option, and with this technology, I think you can do that in a way that’s cost-effective.
You’ve been working on this for so long. What’s exciting to you still in this space?
I think actually almost all of it. As you say, I’ve been working at this for 15 years. It’s been incredible to see the technological advances. I’m an engineer at heart, I love that stuff, and seeing what our team is able to do today versus what we were able to do back in the day, it’s just, it’s incredible. Seeing and digging further into the opportunities of this technology, and the impact, transportation is fundamental to everything. It’s how we get around, it’s how the goods and the food that we consume or use gets to us.
Being able to touch that and help improve our national and global infrastructure in a fundamental way, that’s really compelling. And then when I get a chance to talk to people that are impacted by this technology, and move it from kind of an academic understanding of the impact on their lives to, like, wow, this will change my life, those moments are really special, and I feel like we have a tremendous opportunity, but also a tremendous responsibility as folks working in the space to push this forward in a safe and thoughtful way.