clock menu more-arrow no yes mobile

Filed under:

Full transcript: Self-driving car engineer Chris Urmson on Recode Decode

His startup, Aurora, is building software to make cars fully autonomous.

If you buy something from a Vox link, Vox Media may earn a commission. See our ethics statement.

A bicyclist rides by a Google self-driving car at the Google headquarters in Mountain View, California.
A Google self-driving car
Justin Sullivan / Getty

Chris Urmson, the CEO of Aurora and former CTO of self-driving cars at Google, stopped by the Recode podcast studio to talk with Kara Swisher about the future of autonomous cars.

You can read some of the highlights here, or listen to the entire interview in the audio player below. We’ve also provided a lightly edited complete transcript of their conversation.

If you like this, be sure to subscribe to Recode Decode on Apple Podcasts, Spotify, Pocket Casts, Overcast or wherever you listen to podcasts.

Kara Swisher: Recode Radio presents Recode Decode, coming to you from the Vox Media Podcast Network.

Hi, I’m Kara Swisher, executive editor of Recode. You may know me as the person who thinks Google should buy Uber and rename it Guber, but in my spare time I talk tech and you’re listening to Recode Decode, a podcast about tech and media’s key players, big ideas and how they’re changing the world we live in. You can find more episodes of Recode Decode on Apple Podcast, Spotify, Google Play music or wherever you listen to your podcasts. Or just visit for more.

Today in the red chair is Chris Urmson, the CEO of Aurora. Chris was previously the CTO of self-driving cars at Google, and he co-founded Aurora last year along with Sterling Anderson and Drew Bagnell who ran their own self-driving projects at Tesla and Uber. Chris, welcome to the show.

Chris Urmson: Thanks for having me here.

No problem. I’m so excited to talk about something good about cars, not Uber ... I was joking with you then. I’m so tired of these people. In any case, I’m not tired of you, you’re fantastic. So just by way of background, Chris and I met when ... explain, you showed me the first really autonomous car essentially that Google had in a parking lot near Google.

Yeah, I think we had you out in one of the Priuses way back in the early days.

Well, the early days, right. That was when they were ... those were outfitted cars that we drove around the area, the Google campus area.

That’s right. I think we’d go buy them off the dealership and then we put lasers and radars and cameras on them, add our software to them and go and test them out on the roads.

Right exactly. But after that, it was the little clown car. Remember we did the clown car thing and I tried to have it run you over but it wouldn’t do it.

No, it was good. You know, I think about the koala car rather than clown car, but ...

Okay, what do we call it?

Koala car.

Koala car. All right, okay, I can call it clown car, but explain that. We showed it off at the Code Conference a couple years ago for the first time. But I came out there and Liz Gannes and I got in it. Explain that car, because that was different from the others we had done.

Yeah, so that was a car that we built from the ground up at Google, and the goal behind it was one, to learn what it meant to integrate the software and sensors into a car, and then the other was to have the first shot at having a car that was designed to be a self-driving car. And you know, what do you want that vehicle to look like so it can be a good avatar for the technology in the community and what experience do you want people to have in it. What was fascinating was just how much extra space there was and how ...

Right. There’s no wheel, no driving wheel, no pedals, no nothing, right? Really just a screen.

That’s right. You just tell it where you want to go and it takes you there. So you get to think differently about the interior of the car, you have to think differently about what people want to do in the car and that was part of the exercise in developing the vehicle.

Yeah, we’ll get more into that concept at the time, but it was really cool. It was sort of like being in a Disney ride or something like that. That’s what it felt like because it was small and adorable and stuff like that.

Yeah that’s the kind of thing we were shooting for, right? We wanted your experience in the first time you’re in a truly self-driving car to be not scary, to be friendly, to be fun.

Right. But you kept it in a parking lot, correct? Because you were testing it, you didn’t want it out in the wild, essentially.

It wasn’t ready yet for us to let it loose on the roads.

Right, right. All right, we’ll talk about that more going forward, but let’s go to your background. Now, talk about ... we’re going to get to what Aurora is, but how did you get into self-driving cars? You were there super early and now everyone seems to be a self-driving car engineer.

Yeah. I guess I started working on robots in ’98 when I went to Carnegie Mellon to do my PhD.

Did you do that before when you were in high school? Were you a big robotics person?

No. I built little robot things out of Lego and had fun with that and at some point while I was doing my undergraduate degree I saw a poster for this robot crawling out of a volcano and I said, “That looks really cool. That would be fun to go do.”

You wanted to build a robot crawling out of a volcano? Or just the concept.

Just the idea that you could ... there’s something really appealing about technology that you can touch. With robots, everyone’s seen them in the movies — whether it’s R2D2 and C3PO or what have you — but they don’t really exist. And so the chance to go and work on that, and this was a place that was doing it that seemed really exciting. I went there and I spent a number of years working on NASA projects.

Just before the grand challenge, which were these big robot races out in the desert, I was part of a project that was testing a robot down in the Atacama Desert in Chile. So we had this little four-wheeled robot, it was called Hyperion. It was really ... it was a cute robot and it would move around at 15 cm a second to 30 cm a second which is like a slow walk to a slightly less slow walk kind of speed. And we’re out in the middle of the desert testing this thing and there’s a dozen engineers and most of the time it didn’t work. We were experimenting, playing with it and fixing it, and that was when this DARPA challenge got announced. The idea was to build a robot that could drive from Los Angeles to Las Vegas across the desert.

Which isn’t very hard because it’s straight, right? The desert, it’s ...

Well, it was pretty hard.

No, so no, I get that, but they did a challenge that was relative to something that was manageable, correct?

Well, they called it a grand challenge ...

I know it’s hard, you smart people, but you know what I’m talking about.

Well honestly at the time it wasn’t obvious that it was doable, right? Nobody had done it before and in fact, people thought that it was not solvable, at least in the year and a half or year or whatever it was that we had. And it turns out they were right. We had a bunch of graduate students and undergraduates and we took a Humvee and tore it apart and put lasers and radars and stuff on it.

So robot car, not a robot driving a car.

A robot car.

Right, right, okay. Because that’s creepy, a robot driving a car.

It could be pretty cute. Have you seen the little Asimo robots?

No I’m thinking Terminator, but go ahead.

Yeah, that would be creepy, yeah. So we got this thing, we got it out to the desert, we tested it a bunch. Turns out 10 days before the race we rolled over during testing so that was disappointing.

Wow, did you roll it over? Or did it roll itself over?

It rolled itself over. This was ... we wanted to do a 150-mile test and we were going to do it at 30 mph and take five hours. We said, “Well, why don’t we do it at 50 mph because that would take three hours. That seems better than taking five hours." That was a bad decision.

Too fast.

Yeah, a little too fast. It ended up getting off the road and rolling. We got it put back together, we took it to the competition. It was kind of like robot Woodstock, right? There was all kinds of different things there, from something inspired by centipedes to little ATVs to our big Humvee and everything in between. Most of them didn’t work that well, ours ended up working the best. We got it out to the race course, set off in the morning and it was just magnificent, right? We’d been working on this for a year and a half and this was kind of the first time we let it loose in the desert.

Was someone in the car?

No. It was completely by itself. Somebody from the government was chasing it in a pickup truck and they had just one button that could basically kill.

Blow it up? No, no ...

There was some rumor that they had people that were out in case one of them went crazy, but no, they just had a little remote e-stop for it. And so this thing, you know, we’re stood off to the side and off it goes charging into the desert and it had this giant fin on the top for no real good reason. It looked cool and that was all you could see as it went into the sagebrush. So it charges off into the desert and that was all we knew until a little later in the day, seven and a half miles into the 150-mile course, basically it burst into flames.

Oh, all right. That’s an issue.

It got off the side of the road and high-centered. It was trying really hard to move forward, it didn’t realize it was stuck so it spun the wheels harder and harder and the tires melted and big clouds of smoke.

Oh. Did anybody win?

No, not that year. We went the furthest. We drove through, I don’t know, three fence posts on the way. The guys who went behind us had a much smaller vehicle and so it was really good for them that we took the fence posts out because they were riding our tire tracks and would’ve driven right into the fence posts and probably not broken them. So this was the end of the first challenge.

This is how great inventions are made.

Right. A lot failure, right?

So have you been interested in this as a kid? Or were you just going to go into regular computing?

I didn’t know what I wanted to do. When I went to college, up until the last minute I was debating between going into medicine and engineering. At some point, I realized I really didn’t like the sight of blood and that seemed kind of gross. And then I was like, “Hm, maybe I should do engineering.”

And you like the sight of metals, so you’re fine with that.

It’s okay with metal, it’s not ...

But have you thought about ... robotics was not ... just a regular computing career, correct? Engineering.

Yeah, I went into computer engineering. It seemed like the thing to do. At the time, Nortel Networks was a big thing in Canada and it was hiring and it seemed like that could be fun. And then, like I said, this poster just caught my attention and I was like, “That’s cool.”

So a lot of people do start in robotics. They do start before they’re doing this. What moved you into cars, because you did this car because you did this — it was just that DARPA had this challenge — that entranced you.

It just seemed really cool. Right? The idea that you could actually have a robot do something meaningful. There had been ... robotics had made it into manufacturing and obviously there were space exploration robots, but there wasn’t really robots in people’s everyday life and there wasn’t a way they could have an impact. I wouldn’t claim that at the time could see the future and see what’s transpiring now in the industry, but at the time it was very much, this is a way to help the military, right, and get young men and women out of harm’s way, particularly moving supplies to the front line, which is a big part of their job.

Yeah, so that’s what you were thinking. Had you thought about other robotic uses in the home, of the actual servants or those kind of things, or not? Driving got you because why?

Because it, again, it just poignantly seemed cool and that’s really all the backstory right there. Having spent time walking very slowly behind a robot in the desert in Chile, the idea that you could have a robot instead of moving at walking speed, move faster than I could run and drive through a desert, that just was mesmerizing.

This was a nascent area of doing this, of doing automated cars, essentially. Although in science fiction it’s certainly, that’s all there is. That’s how people have imagined it and things like that. How did you stay in the area? So you started doing it at Carnegie Mellon and then what?

Yeah. So there was that first challenge where no one won. The second challenge happened and a number of vehicles actually finished and this was a year later. There was a third challenge, which this time was at an air base and it was driving on roads and it was much closer to what we’re doing today. No pedestrians, no cyclists, no traffic lights, but the basics of it and that was in 2007. That was the last of the DARPA challenges. It was exciting and fun, that one, the team I was tech for ended up winning.

And what did you do in that one, in the third one?

So the third one, it was 60 miles driving around on roads.

60 mph.

No, 60 miles of distance. It ended up ... it was probably between 30 and 40 mph, that sort of thing. It had to deal with other traffic. So the ones across the desert, they made sure that nothing was moving near the vehicles. This last one, the urban challenge, they had stunt drivers out there bringing traffic into it and following the vehicles around and the vehicles themselves had to interact with one another, so if two of them came to a four-way intersection with stop signs, they had to stop and take turns and if you squinted, you could kind of see the future, right? This was very cool. And then DARPA basically said ...

Any big mishaps during ...

No huge mishaps, there was some entertaining things. One of the vehicles literally drove into a building.

Oh fun.

Yeah. It was a big truck, it was one of these big military trucks and it drove into the building.

And they had given you the trucks they wanted you to ...

No, everybody got to bring their own. There was another one where a car didn’t see one of these foldover gates and drove into it and basically the gate decapitated these Velodyne lasers, you know the spinning Kentucky Fried Chicken things. Right, it just drove into it and off the things flew.

Gone it went, which makes them get around, essentially.

Yeah, that was kind of the end of that for them. Then the cars from Cornell and MIT were in the competition. They both did incredibly well but they had the first robot-robot crash on course. So, that was ...

Did they crash into each other? One wouldn’t give way for the other?

They were both moving at about three miles per hour and it was kind of like a super-slow-speed train wreck.

So they were like, “Nooooooo.”

“Nooooo.” It was very Austin Powers. Yeah. But then, you know, three, five, six vehicles finished that challenge.

So when you were doing these ... These are all fun. I mean, I’ve seen, I’ve been to some of these events: Is it how things get created? Is these challenges and people try to do the contest ... there’s bunches of space contests, all kind of different things. So you were doing this at Carnegie Mellon, how did you get to Google? This was what year you were doing these?

So this wound up in 2007, basically 10 years ago almost, yeah. It was quite early. I spent a couple years still at the university. I worked with Caterpillar and there we were moving, automating these big house-sized dump trucks, which was ... imagine 400 tons with nobody in it moving 40 mph.

Yeah, I would imagine that would be scary.

Cool and scary, but yeah. So spent some time on that and then I ... Sebastian Thrun and I had been talking about doing something together. We had been competing against one another in the challenges and thought it would be great to work together and we had a lot of respect for one another. The teams we had led finished one and two.

He was where?

He was at Stanford at the time. And then he had just sold View Tool to Google, which is what became Street View over time. So he was now mostly at Google and partly at Stanford. So we were talking about starting something, and at some point it came out that the right thing to do was actually just do it at Google. So in the beginning of 2009, I went on leave for my faculty position at Carnegie Mellon and moved my family, we moved out here. My wife was incredibly understanding; we had all our network of friends and relationships in Pittsburgh and things were feeling pretty good. It was this risk to move out to crazy California where everyone wears Birkenstocks and, you know, whatever. In retrospect, best decision we ever made.

Sure. So you came ... and what was the promise? Did you meet with ... who was most interested, Larry or Sergey? Both of them have been long interested in that.

Yeah. I think both of them have been long interested in it. When I came out, I chatted with Sebastian, chatted with some of the early team members and then we decided, “Yeah, let’s do this.”

But where does the impetus come at Google, from the top? They wanted to do ... what was the idea at the beginning?

So the idea at the very beginning was to find out if this could actually be done and it really did come from the top. I credit Larry and Sergey for having the vision to go and try this before anyone else, right? In 2009. When we started talking about this 2010, it was kind of a joke, right? “Google’s doing self-driving cars, that’s such ...” Effectively, “Why are they wasting money on this? That’s never going to work.”

But they saw this back in 2009, right? And they said it really came from a place of they have amazing technology, they have an incredible ability to harness engineering talent, and transportation is such an important problem to work on.

They had been working on a lot of different things, Fiber, all kinds of different schemes and stuff like that. They had some barge in San Francisco Bay that they were working on. This is an area of great interest to them, for some reason, I’m not sure why particularly. But they were the first, they were the first of the companies that set off the interest in it. At Carnegie Mellon, you were doing it mostly theoretically, correct? That the idea that who would ... except for Caterpillar, I’m thinking, or the military.

Yeah, the military, Caterpillar, we had sponsorship from General Motors. They had set up a research lab. Volkswagen had a lab out here with Stanford. So but it was still in the, “Hey, this is 20 to 30 years out” kind of mindset, not the committed, concentrated effort that we saw at Google.

So what were you trying to do there initially at Google when you were starting?

So the goal initially was just to show, prove to ourselves that this could actually become a technology that works. So we had basically two milestones. One was to drive 100,000 miles on public roads, which was ten times more than anyone had done before. And then the other was to drive a 1,000 miles of really interesting roads.

What’s an interesting road?

So driving down the Pacific Coast Highway between San Francisco effectively and LA, or driving all the Bay bridges and dealing with all of the interchanges. And if you remember when 92 and 680 was all dug up, so we’re driving through that at the time, or driving through the Presidio where there’s these windy roads and in fact, this one place where there’s a road that’s only one lane wide but traffic goes both directions on it. We drove Lombard Street.

Oh, the crookedest street in the world. And those goals were because just to show ... to have hype around it or have proof that you could do these things?

The two goals were slightly different. The 100,000 miles goal was really to kind of get statistical data, right. To say, we’re not just kind of driving anything once. This is an interesting data set we can learn something from.

Right, to teach the cars, that’s what you’re trying to do, presumably.

To teach and understand, teach the engineering team, too. It wasn’t just data gathering, it was like, “Oh I never really thought about how retro-reflective signs are or how difficult it is to understand the behavior of an articulated truck.” And then the 1,000 miles ...

But humans do by themselves, for the most part, successfully.

Yeah. It’s very humbling as somebody working in this space, how easy some of these tasks are for people to do and how hard they are to actually get software and technology to solve.

So you were doing that there and you created both cars that were outfitted? Like those Priuses you talked about and then the car itself, the koala car.

Thank you.

No problem, it looks like a clown. But talk about why the different efforts, because one is semi-autonomous, correct?

No, they were all on the path to being fully automated. Yeah.

Why go in those two directions at Google at the time? Which everyone followed, really.

First, it was expedient. So when we started with the Priuses we were trying to understand whether this was even interesting or viable and so we wanted to get on the road as quickly as we could, safely of course but quickly. So, that meant using a vehicle. We then moved to the Lexus and it was the same thing, we were augmenting the Lexuses with our sensors and getting them out on the road, but again it was towards fully self-driving vehicles.

When we started the koala car, that was to now starting thinking more about what this looks like as a product. So we’ve spent a lot of time learning about the technology and we’re getting closer to having it ready to deploy. What’s the first vehicle we’d like where somebody might see it on the road with nobody behind the wheel? How do we, as we start to think about partnerships with car companies, how do we become a better educated partner so that as we work with them and they say something about the flux capacitor, we have some inclination about what a flux capacitor is, right?

Yeah. You’re making a reference from a movie. Anyway when we get back with Chris Urmson, we’re going to talk more about what are the challenges they faced at Google and why he started his startup Aurora.


We’re here with Chris Urmson, who was the first CTO of the Google car effort and now has his own self-driving car startup, Aurora.

So you were at Google doing these things trying essentially to proof of concept, would that be the right way to put it? Or really wanted to make products. You, I know, wanted to make a car on the road, correct?

Yeah. The company really wants to make a product. I believe that deeply. There’s very little value in working on technology if you don’t get it out there and get it helping people.

Right. What was the path to do that? Because everyone suddenly jumped in from Apple, Tesla, everybody else was jumping in, Uber and others. Talk about that environment when suddenly everybody gets excited about something that you were one of the few companies doing.

Yeah, I think on the one hand, you’d like to be able to be the one company doing this and pushing it forward, and on the other, it’s awesome. Because you know a — what is it? — a rising sea floats all boats. And so if you think about the social values of the increased safety on the road, the better access to mobility for people, we want to see this happen, right? I think it’s really important for society, and so as more companies get involved, there’s a broader ecosystem, there’s more likelihood that one of them succeeds. So I think that’s fantastic, right? That’s very desirable.

Right, and so everybody rushed in, including creating companies. Why did you leave Google? What was the ... what happened?

So at the end of the day, I just ... I wasn’t having as much fun, right? And we had a tremendous team and it’s an amazing company. I owed it to the team and I owed it to the company that I was at my best. It just didn’t seem like that was where I was, and you know they had tremendous leaders there, Dmitri Dolgov is a very close friend of mine and he’s just fantastic. He’s stepped up and leading the technology development; with John coming in they have an experienced automotive person there. It seemed like a good time to step aside, and honestly, when I left, I didn’t know what I was going to do. It wasn’t like, “I’m out of here. I’m going to go make a self-driving car company.” It was ...

What did you want to do different that you couldn’t do with the giant sums of money Google throws at people at all times of day?

I just didn’t know what I wanted to do. I talked to a couple companies, actually. This is hard to say, I was talking to a couple of different companies that were working on flying cars.

Flying cars, okay.

Blows my mind that I can say that today.

All owned by Larry Page, anyway, go ahead. I know he has one of them.

They were in fact ... I talked to a number of car companies. I talked to ... I was just genuinely, “Hey, I’ve spent the last decade-plus working on this.”

“I know a thing or two about self-driving cars,” just when everybody is suddenly very interested in it.

Well, that was part of ... you know like certainly people ... it was very flattering, right? When I left there was a lot of interest in talking to me about it, but I was not convinced that this was the next thing I should be doing, right? It was an amazing ride at Google and it was time to like, “Okay, let’s see what else is out there.”

Were you worried they weren’t going to make a car, like actually produce a car, or ...

No, I don’t think that’s necessarily the right path, for Google to make a car. I think Google is very good at the technology side of this, the self-driving technology ...

So as they did in phones. They didn’t make the phones, they made ....

Right, well, that was recently. But no, I think they were very good at the self-driving technology, and there’s people who are very good at making cars and it seems clear that the right path is to marry those up or marry a few of them together. From what I can tell from the outside at this point, it seems like they’re making progress for that. Seems like Fiat Chrysler and them are working well together. I read something in the press about Honda potentially or not, it’s hard to say.

Let’s talk a little — quickly, before we talk about your startup — about the big car companies in this, very slow to it initially but now GM bought up Cruise, everybody’s trying to get a piece of it. Talk about their roles, because they’re the obvious ...

At one point I had one car person saying, “Oh it’s trivial to build a car.” I was like, “No, it’s not. No, it’s not.” Yeah, you’re shaking your head. And I was sort of ... that’s so arrogant of Silicon Valley. I mean, obviously the software is incredibly difficult, but car building is incredibly hard at the same time. And yet they’ve been pretty slow. Why is that, from your perspective?

So I think it’s really important to understand the context, right? And I think the biggest thing that’s lacking — honestly, on both sides — is mutual respect. Because I think it’s very easy for Silicon Valley to look at the car companies and say, “Oh my goodness, they’re so slow.”

“We’re going to disrupt them.”

They’re going to disrupt them, right? And then it’s very easy for the car companies to say, “Oh my goodness, look at those Silicon Valley guys, they’re so seat-of-their-pants,” right? “How can they actually ever do anything big and complicated?” Obviously both of those statements are completely false.

What you have when you look at the car companies, you have to realize they’ve been at this for 100 years, they work in this incredibly regulated industry and they kind of make a miracle happen like every two minutes. They basically put a car on the road that’s going to work for the next 15 years, that’s got explosions going off inside it, right, and it just works. You’re safe, right? If you get in a car crash, more likely than not ...

Most cars you’re safe in, yeah. No, they’re very safe comparatively.

Comparatively. Certainly in the U.S. right? You’re going to do pretty well. So there’s so much constraints on their ecosystem that they’ve developed management processes and they’ve developed business processes that allow them to execute within those constraints. It takes two billion dollars to make a new car. If you could imagine, if to launch any web app you had to spend two billion dollars, the process that we put in place and the way we’d invest in them would be very different.

And so my view is that, the key is to get ... find a way to have a trusting and respectful relationship on both sides, where because of all of those great processes they have in place to allow them to ship cars, they’re not particularly good at kind of innovative software development, right? Those processes are great for the engineering cycles they need, but they’re not really compatible with the top-flight Silicon Valley software engineers.

So this is where we thought that Aurora could maybe fit, is that we do have people who understand — you know, not nearly with the depth that the folks in the automotive space have — but respect how complicated what they do is and at the same time we actually pretty deeply understand the complexity of developing the self-driving car systems and how to do that and how to motivate the team and how to build the team that needs to do that. So that’s the goal for Aurora, is to get self-driving cars on the road as quickly as possible and do that safely and thoughtfully and do it through partnership with the folks that can help us.

Right, which is the carmakers.

It’s carmakers, it’s a broader ecosystem than that.

What meaning?

Meaning that, transportation as a service — you know, whether it be Uber, Lyft or Didi, right? That’s going to be an important part of how this technology comes to market. The car company ... you have to have a car to have a self-driving car. There’s folks who actually ... so I think I understand a little bit about lasers, a little bit about radar, can probably contribute to the design of those, but I don’t really know how to make it so that it can submerge in a salt bath and drive on your car for the next 15 years. So there’s a whole collection of tier-one automotive suppliers who know how to do that part really well. And it would be silly for me to try and do that or for even our team even to do that.

Right, but it requires the coordination of a lot of people and that I think is why it’s so difficult.


What’s interesting is that carmakers themselves, though seen as slow, it’s sort of eating into their current business. I mean, every time they can ship a Ford 150 or whatever, the trucks and things like that, there’s that. Like who’s thinking of it at these car companies?

They’re looking at companies like Uber, Didi and Lyft and they’re seeing that something is happening.

Meaning nobody wants to drive.

Well, not just nobody wants to drive, but there’s an immense amount of value — at least on paper being credited. And if you look at, say, the market cap of Ford, General Motors and Fiat Chrysler, the U.S. big three, and you look at the market cap of Uber, Didi and Lyft, they’re basically the same. I think it may actually be ...

And Tesla is all of them together or something like that.

Well, I don’t know about all of them together, but it’s bigger than one or two of the U.S. guys. I think that the market cap of the ride-sharing companies is probably going to double in the next few years.

Why is that, given that ... they sell millions of cars and make lots of money, this group doesn’t?

I think that part of it is the opportunity to capture kind of the usage of the car in a way that they can’t do as a car company. So when you buy a car, you know you probably use it ... I don’t know if you own a car or not.

Yeah, I do.

You probably use it an hour, maybe two hours a day at most. If you’re in the ride-sharing business and you operate it effectively, you might be operating that car 16, 18 hours a day. And so you get much higher utilization so there’s much higher value you create out of that vehicle.

So the carmakers, the danger they had ... they have been worried about is becoming the dumb pipes, essentially, the way the cable operators were.

Yeah and often the metaphor they use is they don’t want to be the handset provider, actually. And part of that is because what they do is complicated and valuable and so they don’t want that to be devalued.

So how do they stop being that way?

Well, they’re trying. You see almost all of them at this point running a different set of experiments, whether it be Ford with the ride-sharing bicycles around town and Chariot, or General Motors with the Lyft investment and purchasing Cruise, or VW with the experiments they’re running as well. So they are trying to make the leap and they’re trying to be innovative, and I hope some of them succeed. I just think it’s going to be difficult for them given how they’ve operated historically.

Right, so what do they have to do? Buy things or just ... move or what?

So I believe they need to work with partners, right? I get that that’s kind of my thesis.

So let’s move then to the tech companies that are working on it. Apple which seems to have pulled out somewhat, Google, Tesla, Uber, all of them. What’s their problem? What challenges do they face?


They don’t make cars, one.

Well, except Tesla, Tesla makes cars, and so on paper I think Tesla is positioned pretty strongly in this space. I think the biggest challenge with Tesla is that the approach they’re taking is not maybe connected with reality on what it takes to actually make the car drive itself.

Meaning what?

Meaning, I think they’re underestimating the complexity of actually getting to a fully self-driving car.

Right. Theirs is semi-autonomous.

Yeah. And I think it’s potentially a great feature, what they’re developing, and Elon has a very complicated space that he’s fighting in and he needs to have feature parity with the luxury cars. So some of what he’s doing really makes a ton of sense, but it’s hard for me to believe that with that just cameras that they’re going to, in any time in the near future, get to the level of reliability you really need to launch a car where you can kind of go to sleep in it and let it take you where you want to go.

Right, we’ll talk about it in the next section. And then, the regulatory environment, you know that’s another thing and then humanity is one of the biggest issues. But talk about the regulatory environment first.

And this is what actually makes this space so exciting, right? It’s the technology is really interesting and cool, the business cases are pretty exciting and then there’s this incredible opportunity for social good, but you have to communicate that. You have to work with the regulatory bodies, whether it be at the state, city or federal level. Right now, I think in general the tone is relatively optimistic.

The federal government set guidance last year that you know was basically as attractive as they could make it, I think, reasonably be expected to make it. A number of states are trying to jump in to support and to kind of encourage the technology coming to their state. The challenge with that is that we don’t really need them to do a whole lot, and the worst thing would be for there to be 50 different sets of rules because then you’re in your self-driving car and you get to the border, and now it’s not legal in a new state.

How does the government do that? Is this federal government equipped to do that?

They’re working hard to be. What’s ... and maybe this is nuanced, but the way regulation works is at the federal level they regulate the safety of the car and at the state level they regulate the safe operation of the car. So the driver ... and so what’s fascinating about self-driving cars is you’re now making a virtual driver and so there’s a little bit of state-federal power grab for who gets to make the rules around that.

Right, right. Where do you think it ends up?

In the interest of my sanity and other people’s work in the space, we wanted the federal so there’s kind of one regulation nationally.

Right, which is something, everything is regulated by the state, DMV, almost everything that touches the user.

Yeah. Anything that’s operating the vehicle, like which roads you can run it on, which speed limits you can drive at, who’s appropriate to drive, that’s all kind of state level.

Are there any states that are really fast forward? I’m guessing California.

So California, through Senator Bhatia’s leadership, ended up putting out regulation ... Nevada actually had regulation out before them. Florida had some guidelines out early. They were really well intentioned, but what we’re seeing is in some cases a little bit of the regulation was there before we were really ready for it. And for the most part it’s not been an issue, but it’s ... you look at some of it now and you say, “Oh, hm. Maybe that wasn’t quite what ...”

What’s the biggest problem?

Oh I don’t have a great example off the top of my head.

But there’s ...

There’s just like little things. There’s no huge problems.

Cart before horse.

Yeah a little bit.

Or car, so to speak. No more horses.

Yeah. So to speak.

And what about internationally?

So internationally, it’s interesting. So in Europe they’re generally more conservative and the law there ... in the U.S. the law is generally permissive: You know, if we don’t say you can’t do it then do it and if somebody gets offended they’re going to sue you. In Europe, they’re much more strict at delineating what it is you’re allowed to do and what you’re not. So I think the law in Europe historically has been relatively protectionist. So as the European auto manufacturers get to the point where technology is ready and the law obviously moves along with it.

Along with the car manufacturers.


And there’s no big tech firm there that really is out in front of this.

Well again, they would argue that a Volkswagen or a Daimler is a big ... yeah, but you’re right, there isn’t outside of that. But those two companies are investing heavily in this space.

Yes, they are. I’ve been in some other cars. So last thing, humans. The idea of humans doing ... I mean obviously Tesla got a lot of press when that man died. That was just in a semi-autonomous car, but the idea of sleeping or doing something else or watching a movie or doing anything else. When I was in the car I was like, “Where’s the bar?” kind of thing, “I think I’ll text,” and things like that, but that was because I was in a parking lot at Google and I felt totally comfortable doing that. How do you switch humanity into that mode? I think they’ll do it really fast.

I do too. And we’ve seen this.

Because I felt comfortable pretty quickly.

Yeah. It’s amazing how quickly people go from, “Oh my goodness, how could it possible work?” to “Oh, all it does it drive?” Right? And they do that in a matter of minutes to hours. And I think this is one of the biggest challenges for some of the semi-autonomous features, is that ...

The grab for the thing.

Well, it’s not even that they grab for it, it’s that they experience it for a while and it works, right? And maybe it works perfectly every day for a month. The next day it may not work, but their experience now is, “Oh this works,” and so they’re not prepared to take over and so their ability to kind of save it and monitor it decays with time.

But when we think about the rate at which bad things happen, they’re very low. So you know in America, somebody dies in a car accident about 1.15 times per 100 million miles. That’s like 10,000 years of an average person’s driving. So, let’s say the technology is pretty good but not that good. You know, someone dies once every 50 million miles. We’re going to have twice as many accidents and fatalities on the roads on average, but for any one individual they could go a lifetime, many lifetimes before they ever see that. So that experience with the technology and kind of becoming falsely comfortable with the safety of it is one of the challenges they face.

Also, liking to drive, how do you beat that? What’s your argument?

So, people like to drive some of the time, right? There’s an awful lot of driving that people do that they really hate. Road rage would not be a thing if people truly liked driving.

So enjoyable this traffic jam, we’re looking right now at the Bay Bridge.

I’m really enjoying 101 today.

I hate the Bay Bridge.

Yeah. And so, I think what this can do is, so much of getting around is just mundane. Let’s make that pleasurable, right? People are on their phone, people are wanting to do other things in the car. Like instead of kind of continuing prohibition, once this technology is out there we now allow people to do what they want. And while they do that, they’ll still be safe in the vehicle. Then for people who want to enjoy driving, go and drive in your spare time, like go drive on Sunday afternoon when the roads aren’t bad and you enjoy it.

Sure, like a horse.

Yeah. I think over time that’s kind of ... that’s going to be the experience.

Even that whole idea of Americans love their cars ... I don’t think they do.

I think a lot of Americans do. I think they take pride in it, I think they do spend an hour or two a day in it, and the ... you know, people buy a car because what it says about them, right? It’s kind of their avatar on the road.

I have a Ford Fiesta, I don’t know what that says since I don’t care about cars.

You’re like, “I wish I had a self-driving car right now. This is so awful.”

It’s a stick shift. But I only got it because it was the cheapest smallest car that was fast.

You could get a small cheap car with an automatic transmission.

That’s true, but I like my stick shift.

Yeah. Exactly, so there’s part of driving you enjoy.

That’s true. So finishing up this section, so how much money have you guys raised for what you’re doing? What precisely does Aurora do?

Yeah. So we aren’t talking about how much money we’ve raised ...

I’m assuming a buttload because everybody’s ... everyone and their mother is raising money.

We’ve been fortunate in our ability to raise money. So far we’ve just taken a little bit of money from some folks who helped me and my other co-founders think about how to start up a company and just grow.

And your idea is to help ... be the software layer, presumably.

Yeah. We’d like to be the system. So we’re going to work on the software and we’re going to build a reference architecture that we’ll share with automotive partners that’ll include: These are the sensors you should have, these are roughly where you should have them on the vehicle and this is the computation you should have. We’ll be designing our software to work with that reference architecture and then hopefully ...

If it’s a current car that already has a wheel or if it doesn’t, if it moves to the other part.

So we’ll be agnostic. Our intent is to build the self-driving vehicle capability. There’s a lot of companies, all the tier-one automotive suppliers are developing the driver system, so things where you’re still driving but it’s keeping you in lane or doing adaptive cruise control. We want to work on the part that comes after that.

Right, fully self-driving.

Fully self-driving.

All right, we’re with Chris Urmson, we’re talking about self-driving cars and where they’re going. We’re going to talk next about the future and what the future actually looks like when we get back.


We’re here with Chris Urmson of Aurora. Chris used to be the CTO of Google’s self-driving efforts. Talk a little bit about where we’re going. First, I’d be remiss if I didn’t ask, you were at Google during the Anthony Levandowski era, you’ve been sucked up into Waymo. It’s called Waymo now, Google’s self-driving unit. Where is that name from? Don’t tell me, you don’t know either. I don’t know.

No, I don’t know.

What’s going to happen there, this fight with Uber ... I’m guessing you’re being deposed and etc. so you can’t say much.

Yeah, so in fact I was deposed yesterday, so there’s not much I can say. The good news is I don’t know a whole lot, right? I obviously had some experience with Anthony and spent a lot of time with the team, but don’t know what they’ve found and what the true meat of the case is.

Right, but I think at the center of this there’s going to be a lot of fighting over this stuff because everybody wants to sort of have the edge going forward. So let’s talk about how quickly that edge is coming because there’s lots of money going in, every week there’s a different self-driving car unit gets sold or there’s some aspect of it. Let’s talk about what it looks like, where does it come first? Like Levandowski had Otto, which was the trucks. You were talking about working with Caterpillar long ago, military ... where do the first ones really emerge that are useful and not completely accident-proof but pretty close?

Sure. So we’re already seeing this technology come to market today. There are mines where you can go to where they have big haul trucks they call them, big dump trucks driving around with nobody in them. So, in those kind of closed-course environments you can see ...

Sure, where they’re on tracks.

They’re not literally on tracks, they’re on roads and they’re using GPS and figuring out where they are, but it’s mostly a closed site. You can go and see that today. I don’t know exactly where they are, but I’m sure some folks at Caterpillar would be able to tell you.

And you’re starting to see some of the really high-end driver system features, where I think Audi just announced that their 2019 or 2018 model you’ll be able to go on a freeway, on a divided highway at less than 35 mph it’ll stay in the lane and not bump into the cars around you and you can read a book or whatever for however long the traffic jam lasts. So you’re starting to see that today.

I think within the next five years, you’re going to see transportation services, fleets, vehicles where in kind of some limited environments they’ll be on the road and you’ll be able to get a ride in one.

And this is without a driver in them.

I think without a driver, yeah.

You’ll just get in. What’s limited? What do you mean by limited? Like around a university, where there’s no other humans or ...

No it’ll ...

No driving cars. I think the issue is that they’re humans is the problem. If everything was self-driving presumably they could start really talking to each other.

It would be much easier. That is what makes my job hard ... yeah let’s not ....

Me. I’m the problem.

It also undercuts some of the value, right? People move around. It’ll be in environments where it’s relatively benign. So nice wide roads, maybe not as many people using them, I think that’s part of the reason why somewhere like Arizona is appealing, right? It’s a relatively modern city, it’s so darn hot that you don’t have a whole lot of people enjoying the outdoors.

So you can’t run them over. That’s your point.

Yes. You know, and again, the safety arguments here really all come down to statistics. It’s a statistical argument that the rate at which bad things will happen is lower with the automation than it is with people. And so, wherever you can find advantages to push the statistics in the right direction, then you know that’s the way to think about this as an engineer.

So you would, say, have a car come and pick you up, no driver in it, in these places like Palm Springs, for example.

Somewhere like that, yeah. And I imagine that happening within the next five years. Then, once you kind of are able to let that proverbial genie out of the bottle, at that point it becomes incremental to make it work in slightly more difficult places. The big delta will be when you have a partner that can allow you to scale the technology because making 100 of something or making even 1,000 of something is manageable. But when we want to get to the point where the technology is proven and you want to scale it to the point where it can service a whole city or hopefully many cities, then you need somebody who knows how to make vehicles and you need somebody who knows how to make sensors that’ll last and can do that repeatedly.

Where is the problem? Is it the sensor problem? Or where is ... humans, I’m guessing, is the biggest one. People, other people, when you have that moment of mixing ... probably when you mixed horses with cars was probably one of the more dangerous moments of car introduction.

Hence the person walking in front of the car with the flags. I don’t know if you’ve heard this story.

No, no, tell me.

Yeah, so apparently, I can’t remember which state it was, it was somewhere east. When the first cars came out there was a law in the books that you had to have somebody walk in front of your car with a red flag so that you wouldn’t startle the horses, which somewhat undercuts, again, the value of the car. Obviously we don’t do that today.

Right. So there has to be a transition period. Do you imagine cities getting rid of cars completely and then using only self-driving units? That could be something ...

I think in the distant future, yeah. I think there’s going to be a long period of crossover where there’s going to be people driving cars and self-driving cars together on the roads and that’ll be the most difficult time. And then once the technology starts to prove out, the opportunities again for safety, for low-cost mobility, for mobility for people who can’t see or can’t otherwise drive, reducing congestion and all this stuff, cities will want ... because we’re only seeing ... in the U.S. we’re seeing urbanization. In China, they’re seeing urbanization in a way that I can’t comprehend and so what we need to find is a way to provide the same mobility that you and I take for granted, but that can scale to a much denser population. I think that’s one of the promises this technology has.

What role does .... so people would not own cars, right? Presumably they would hire. This is why Uber and Didi and all these companies have such big valuations because they’re the reservation system essentially for these things that you wouldn’t own a car.

I think you wouldn’t need to.


Right, so for someone like yourself who professes to hate driving, you only own a car because you need the convenience of it. If somebody were offering you the same service but it would show up, you wouldn’t have to worry about driving and you knew it was going to be there and it was going to work and it was as expensive or less than your current thing, then you would take it. And I think you see this in New York today, where a vast majority of people don’t own cars because it’s much more convenient to use ...

Like cabs, public transportation.

Right, which was transportation as a service at the time.

What happens to public transportation during all this?

So I think this ends up becoming public transportation.


You know, buses are fantastic when they’re loaded, right? That’s one of the best ways that you can actually move people.

But on certain paths only, right?

Right. On certain paths, but most of the time they’re not.

Or they don’t go where you want.

Or they don’t go where you want and the whole system of where do we set the routes is a very political process that maybe is not the best that it could be. If instead over time we keep the buses for the core routes and treat them almost like trains, but then we either serve to them or go point to point with much smaller vehicles that really stop at your house, stop at my house, I don’t know where you live but if we live close to each other. That would be a really bad routing. But on the way, and really they ... it looks like public transportation on demand and it’s actually, when you look at some of the numbers on this, you can probably operate that at a lower cost per mile to the transit authority than the bus system.

There wouldn’t even be a transit authority, save money, make money for someone else.

I think there’s some things that should be public good over time and it feels like ... over time this might be something that becomes part of the utility of the city.

The utility of the city. What are some of the weirder transportation modes that you know? Vertical lift and take off, Uber was thinking about that. I don’t know, they’ve got some issues right now to deal with separately, but talk about like where the craziest transportation schemes are coming, from your perspective. You like robots coming out of volcanoes, I’m assuming vertical lift and take off ...

That seems pretty cool.

Yeah. That’s where you say you’re here in San Francisco and you want to get to Berkeley and not go over that horrible Bay bridge, you just get in a ... it’s not a helicopter, it’s something else.

It’s like a giant drone that flies you over there, right? That actually sounds pretty neat to me. One of the neat things that’s happened with drone technology is you solve reliability. The nice thing about being up in the sky is there’s not a whole lot to hit, right?


Yet. But even then, there’s so much more space than there is on the ground. We’re stuck on the surface. Once you get up there you can segment: All the eastbound traffic could be at one elevation, all the westbound at another, north and south, and so you don’t end up with a whole lot.

You don’t have to build roads where we’re going. There’s no need.

There’s no need for roads. Thank you, Doc Brown. Yeah.

Speaking of that, so vertical lift and take off, flying cars ...


Which is an adjunct, a to-the-side adjacency, correct?


When are those coming?

I don’t know.

They’re working on it.

They’re working on it.


Why? I think because there is ... you know, again, I think it’s this density and congestion problem and that there’s just so much resource available up there in terms of space that if we can tap into that maybe it makes it more livable. I worry a little bit about the sound.

They’re noisy.

Maybe this is a little too nimby, but you know there’s going to be a lot of noise.

Well I don’t want someone taking off a car in my neighborhood, you’re right. Like why would you want it? What about jetpacks? C’mon.

I don’t know much about jetpacks. I haven’t seen that.

Okay, you’re not working on those.

Personally I’m not working on flying cars either.

What is the craziest thing that you ... well, a teleporter.

A teleporter would be fantastic. That would solve the commute problem, put me out a job. That would be great. I haven’t seen a whole lot ... the Boring Company stuff, the idea of digging giant tunnels seems on the one hand, I remember looking at ...

This is Elon Musk’s ...

Tunnel boring. Yeah. So on the one hand I remember reading books about the future and it was all going to be vacuum pods moving underground ...


Yeah, they didn’t call it that at the time. I think he coined that later. So on the one hand that would be ... it would make the future happen, at least the 1970-something future happen. On the other hand, it seems like there’s other ways that we can ... digging holes seems hard.

Yeah. I think he just likes to fuck with us all the time. Like he’ll say something crazy. “Next I will be invisible.” “Oh yes. Yay Elon!” Do you know what I mean?

I know what you mean.

You know what I mean? Well good, I’m glad someone’s doing that. I really admire him for putting out.

And what Tesla and SpaceX ...

It’s better than a photo app, let’s just say.

Yeah. You know what Tesla and SpaceX, their mission is just tremendous. Right? I’ve got to believe for the folks that’s incredibly motivating.

So to finish up, give me the timeline. When is this going to be like done?

It’s never going to be done.

No I know, but like right now cars are done. There’s no more horses around.

Oh the transition? I think it’s going to take at least 30 years.

30 years.

At least. Just part of that is it takes about 15 years for a car to cycle through the American, what we call the car park, right? The cars on the road. Part of it is the technology’s not ready yet and there’s going to be continued development on it over time to get it to the point where it can really be deployed. Part of it is that, unlike a web app or a mobile app, there’s actually investment you have to make to adopt it and so it’s going to have a slower roll out.

There certainly is a lot of money going towards it.

There’s a lot of money going towards it, but even a lot of money when you start thinking about buying cars, it starts to get factored down pretty quickly.

I want to finish up on the social concept of it, which we talked to a lot of Silicon Valley people about, is if you’re going to make all these cars, self-driving cars and autonomous vehicles, do you have a responsibility to those who lose their jobs? Do you think about that as an inventor?

Yeah, no, this is actually something I do think about a fair bit. It’s a real challenge and this ends up being a trade-off in social goods. So on the one hand, there’s people who drive for a living today.

Not just at parking lots, insurance companies, like you could iterate and iterate and iterate.

There’s a lot of scale in this business, and a lot of ecosystem. On the other hand, you have the opportunity to make people safer on the road and save tens of thousands of lives.

Save energy.

Save energy, which is good for the environment, good for the planet. You can decrease congestion, give everybody back more time. You can give access to people, probably a commensurate number of people that can’t drive today as those who drive for a living, and then you give them the freedom now to go and contribute to the economy, have a job and work in a way they may not otherwise be able to. So it’s difficult, right? No matter what you do, whether ... you know if I go buy lunch at Subway or Jack in the Box, you know that has some implication in it. I don’t mean to be trite about this, but how you put your finger on the social scale here is a really tough question. I don’t have a good answer.

Well, who’s responsible then? I think one of the things that Silicon Valley is, it’s not responsible, and very similar to a lot of companies that make things and change things. But who? Is it the government? Is it ... because then you could lead to you know we’re right in the middle of a populist political situation, could get worse.

I’m not ... in no way am I trying to abdicate the responsibility of ...

I get that, but wondering who you think ... I’m not blaming you.

No, no, I appreciate that. I think we all are and I think that this is one of the saddest things for me about what’s happening in the political world right now is, we are faced with a variety of challenges, right? It started early with factory automation. It started with globalization. And these are things that have been good, much like self-driving technology. It’s hard to argue, I think, on an Earth-wide perspective that globalization was a bad thing. I think it tied countries together, it raised the economies in parts of the world that wouldn’t otherwise have. Yeah, it made us interdependent in a way that prevents war, and so there was some clear social good from that, but it wasn’t distributed uniformly. Like any change is not distributed uniformly.

It’s part of our community, part of our political leadership, part of our social activism, is finding the right way to bring society, all of society, along with that. I think we’ve fallen short. It’s clear that we’ve fallen short. I don’t think we would be in the political situation we are today if we had not fallen short. I would love to be able to tell you, like, this is my vision, this is the answer. Honestly, it hurts a little bit not to be able to do that because I recognize the implications of what we’re doing. I just, I’m not smart enough and I think we need to attack it together.

Interesting. I had Marc Andreessen onstage talking about this and he was like, “Well, there were blacksmiths.” I’m like, “Yeah, but what about the blacksmith families? What happened to them?” What happened to ... you don’t even know what happened to them. They went away, certainly. They died eventually, but it’s a really interesting question that Silicon Valley abdicates its responsibility for ... and this is a big one. To me, this a really ... like the biggest.

I think this is, yeah ...

You know, they’ve ruined media, but that’s okay we’ll be fine, but you know what I mean, like it’s sort of ...

It’s important and it’s not ... this is a broader challenge that we’re facing that isn’t just self-driving cars. I think the automation is going to kick in more broadly. I think it’s going to be on a ... kind of the net it’s going to be a good thing. I think in general the advance of technology has been a good thing, but it requires us to be thoughtful and it requires us to come together and figure out how do we help those that are displaced. How do we move them into new roles? How do we even think about that? It is something I worry about. Unfortunately it’s in the state of academic worry.

One concrete way to think about this — and we talked earlier about public transit. You could imagine like bus drivers lose their jobs. That’s certainly one way that the pendulum could swing. An alternative is, we put the bus drivers on the core routes that are driving the buses, we don’t hire any more bus drivers, but we phase this technology in over time and we support it with the self-driving cars. The self-driving cars, if we get reasonable occupancy on them and they really are a public transit, then today we subsidize public transit, about 90 percent of it is subsidized by taxes. We instead will no longer have to subsidize it and then we have this extra money that’s sat around in city and transport authority coffers that we can use in different ways.

Yeah, yeah, you have to be. It’s interesting, I was just in Kentucky and talking to some coal miners. They’re like, “They’re going to bring back jobs.” The minute they can automate you, my friend, they’re going to automate your work and they probably already have in lots of ways, which is interesting. It’s definitely an interesting question for all of us, I think. And I’m so glad someone as thoughtful as you is thinking about it. You’re more thoughtful than most people. Most people are like, “Eh, let’s invent it. Who cares?”

No, I think that’s dangerous.

Yup, I agree with you. Chris it was great talking to you. Thanks for coming on the show.

This article originally appeared on

Sign up for the newsletter Sign up for Vox Recommends

Get curated picks of the best Vox journalism to read, watch, and listen to every week, from our editors.