Waymo Cofounder Says Safety Is ‘Paramount’ to Dominating Robotaxi Race

Sebastian Thrun, a pioneer in the self-driving industry, looks at Waymo like a proud dad.

Nearly 20 years ago, Thrun and his team at Stanford University made a pivotal leap in autonomous driving after their retrofitted Volkswagen completed a 132-mile course in the Nevada desert.

There was no human behind the wheel.

The feat captured the attention of Google cofounder Larry Page. The then-CEO called Thrun and convinced him to head Google’s self-driving project, which would be part of Google X, the company’s research and development arm. There, some of the smartest and brightest pursued “moonshot” projects.

Since then, Thrun has watched autonomous cars grow from an experiment to a real-world business.

The Google project was rebranded Waymo in 2016, two years after Thrun had already left the search engine company to start up a tech school. Thrun also headed the now-defunct Kittyhawk, a flying car project backed by Page.

Thrun spoke to BI about Waymo’s early aspirations, what it will take to dominate in the race to autonomy, and why he still hasn’t let go of his flying car ambitions.

A spokesperson for Waymo declined to provide a comment.

This interview has been lightly edited for length and clarity.

When you jump-started what was once just a “moonshot” project inside Google X, what was the vision you had for the Google self-driving project, and does Waymo live up to or exceed your expectations?

Oh, it lives up to my exact expectations. We wanted to make cars safer with self-driving technology and build a business in which cars are used better than they are today, AKA a ride-sharing business.

On the timeline side, it proved to be a hard nut to crack because the tolerance to errors is extremely low.

For example, compare this with recent advances in large language models and chatbots. When a large language model hallucinates, we scratch our heads and are not happy, but no one dies.

When a self-driving car hallucinates, it might kill somebody or run a red light.

What did you imagine Waymo’s business model would look like at the time?

When I was running the team, we had business aspirations, but we had not worked out practical things like go-to-market strategies. And our aspiration was really to be the number one ride-sharing, ride-hailing taxi company in the world. But 99.9% of my time was dedicated to making the technology work. And remember, this was 10 years ago.

Do you think Waymo made the right decision by going city-by-city to get these highly detailed maps before it deploys a robotaxi service to the public?

I believe that safety is paramount. The data suggests pretty clearly that Waymo is taking the right course here. Not only have the cars been remarkably safe — as you know, a recent study by Swiss Re showed that they’re safer than human drivers — but I also think Waymo has been able to build the trust of the public necessary to operate.

Do you think scaling autonomy is fundamentally a software problem that could be solved with more data, similar to AI models? Or do you think autonomy will always be bound by geographic limitations or infrastructure?

No, I firmly believe that at some point, the technology will be good enough that you can forgo the necessity to build maps.

But I also want to point out that building maps is not an expensive step. It’s a bit of a side product of testing. We look at the number of miles that Waymo has been tested. It’s now in the tens of millions. If you apply this to the United States, that’s like mapping the United States over and over and over again. It’s not an expensive factor.

I’m pretty convinced that the technology is good enough even to drive without maps, but it’s just such a good thing to have — an advanced knowledge of where to expect certain things. And remember, the map changes all the time. There are construction zones and routing changes, so it’s not a constant thing. The vehicles are constantly remapping the environment.

So, do you envision a future where Waymo will increasingly have to rely less on these detailed maps or maybe even fewer sensors?

I can’t really speak for the current team here, because I’ve not been with them for a decade.

One of my observations is that, just looking at the field as a whole, the cost of sensors has gone down dramatically.

Right now, I’d say the cost of lidar relative to IRAD (Internal Research and Development costs) is roughly down by a factor of 50. So, to the extent that the sensing argument is a cost argument, I think that as these cars will be produced in larger numbers, that argument will go away.

Then it comes to safety. My firm belief is that we should have any piece of software or equipment, including sensing equipment, that can be demonstrated to enhance safety. We are very tolerant of the lack of safety in human driving and tolerate more than a million deaths every year worldwide, which is a significant number. But I don’t think we should have the same tolerance for robotic systems. We should have a higher standard, and the higher standard will lead to fewer funerals.

John Krafcik, the former Waymo CEO, told me that when we talk about sensor costs, the cost argument is trivial. He believes that there are “quantifiable” benefits to safety.

You’ve tested it, trusted your life into it, and can see the reception now in San Francisco, which is really widely positive. It’s become an icon in the tourism industry. You come to San Francisco to try it out.

All these things are possible because of the radical focus on safety. Everything is speculation, but I think it’s such an important cornerstone that this technology is accepted and that we feel it defines a completely new level of safety and transportation.

In that case, what do you think of Tesla’s proposition: A generalized AI driver? No high-definition maps. No lidar. End-to-end AI driving.

Look, I cannot comment on Tesla. I don’t know the details of the technology. I can only tell you what my ethos was when I built up the early version of Waymo. Our ethos was that safety is so paramount.

I can tell you, positively, that if you took Waymo and ripped out all the radars and lasers, that would make the car less safe. I can say that with confidence, even though I’m not part of the current team.

The laser and the radar provide a layer of environment understanding that is succinctly different from a camera. They’ll pick up objects just by virtue of being there, even if they’re unknown to the machine learning system. I know from the team that they’re obviously getting better and better with a subset of sensors.

Can I ask like this then? What do you think will determine the “winner” of the robotaxi race? Is it the company that proves to be the safest, or is it one that will scale quicker first?

There are a number of elements. I would say safety is paramount.

The two remaining variables are cost and scale. They are intertwined, but they are not the same.

On the cost side, a self-driving car ultimately has to be cheaper than a human-driven car or comparable. I believe that cost is an important factor in transportation for the vast majority of people in the world.

And then scale is obvious, and scale is hard. It’s an enormous undertaking. If you look at the details of what this really entails, there’s capital, there’s maintenance, and there’s even manufacturing. So I think that now that we have the proof-of-concept, and it’s a working, very accepted and safe taxi system in four — and soon five cities — I’m sure that, and I’m not part of the leadership team at Waymo anymore, but I’m sure that, at Waymo, people are now thinking about how to scale it up.

Is there anything you think Waymo should be betting on or doubling down on right now?

Well, I’m a long-term fan of an even more radical proposal, which is flying cars.

Would I recommend that Waymo drop everything they do and start working on flying cars? No, I would not. But I can tell you the way flying cars are today is exactly the same way self-driving cars were 10 years ago. Ten years ago, when I left, we had prototypes in hand that were able to drive hundreds of thousands of miles without what we call “critical interventions,” where the driver had to take over — although with the caveat that this was mostly highway driving at the time.

But it was nowhere near a level of safety that would allow us to operate a commercial business. And obviously, between that time and today, new laws and regulations came into effect, putting a formal pathway in place that would’ve retroactively prohibited us from building a commercial service 10 years ago when I left.

The flying car situation is more nascent. There are now prototypes of electrically-propelled vehicles that can take off like helicopters. They’re super quiet and can fly for about a hundred miles, give or take.

But we are far from regulatory clarity. Aviation is regulated federally, not statewide, and has a very high bar that you have to meet to even be airworthy. No one has yet met that bar. This is more of a research and development project than a practical project.

Do you envision a future in which people are less inclined to drive?

Look, my prediction is going to be — and I can’t tell you what the timetable is — there will be cities that will realize that this technology is radically safer and radically greener. Every city in the world is full of parked cars, and we don’t think about this, but it’s quite a burden on the city. A friend of mine once calculated that about 60% of the land mass in Los Angeles is dedicated to cars when you count things like garages and driveways. So more than half.

There will be cities that say, “Look, we are ready to make our downtown area a parked-car-free area.”

We’ll still have car-based transportation. But we’re going to make it greener. You want to make it safer and more pleasant for people. And we’re going to do it in a downtown area with lots of young people who don’t want to own a car to begin with. Or maybe university cities will do this.

I think we’re going to see this in the next few years, somewhere in the world, where cities will say that’s the right time. And what this really means is: For this to be true, it has to be the case that the car will become lower in cost than car ownership. At least that’s my belief. Because then you can even make an economical argument to people and say, “Look, you can own a car, but this thing, I mean, it’s cheaper.” And you save money on top of it. It’s greener, it’s safer, and cheaper. How would cities not want to do this at some point?

But look at New York, for example. It has congestion pricing right now. If you switch to self-driving cars, the very first thing you do away with is traffic lights. You don’t need them anymore because cars can communicate.

Your capacity will also increase. If self-driving cars are allowed on a highway, you could easily reduce the spacing between cars by a factor of two, which would double the capacity of the highway.

Anything you would be doing differently if you were behind the wheel at Waymo?

I’m the biggest fan. I open my door every morning and see two or three Waymos zipping by my house, and I couldn’t be prouder.

I’m really very, very proud. The leadership has done an amazing job navigating and has really earned the public’s trust, which I think is so important for technological innovations.