Eight Years and $1.1 Billion Later, How Did Google's Self-Driving Car End Up Playing Catch-Up? | QDaily Data

Google underestimated how fast autonomous driving would reach the road, underestimated what it takes to build a new car, and underestimated how hard its first business model would be.

Google's self-driving project, which drew so much attention in its early years for that adorably goofy little car, has lost some of its shine.

Talk about autonomous driving today and more people will think of the tens of thousands of Teslas already on the road, cars that can follow traffic on the highway and drive themselves. Tesla CEO Elon Musk has gone further, saying every car built since mid-2016 will be capable of full self-driving next year through a software update.

Even Audi, never known for its computing, has launched a new A8 series that can manage Level 3 autonomous driving.

Google got there first, though.

Google's autonomous driving program started in 2009 and has now run for eight years. It was in fact one of the first companies to make autonomous driving a formal R&D goal, and its signature test cars, lidar units perched on the roof, became a distinctive sight in Mountain View, California, long ago.

These days the bigger news about Google's self-driving work is its lawsuit accusing Uber of stealing its trade secrets.

A court document in that case recently leaked what Google has spent on autonomous driving: from the project's founding in 2009 through 2015, $1.1 billion on self-driving R&D.

What did that $1.1 billion buy?

Google began autonomous driving research in 2009, and in 2012 it put lidar-equipped test cars on public roads — one of the earliest attempts to use lidar for self-driving.

As of 2014, Google was running modified test cars on public roads in California: six Toyota Priuses, 23 Lexus RX450h SUVs and one Audi TT. The cars carried computers, multiple lidar units, GPS antennas and other gear, and by June 2016 they had logged more than 2.77 million kilometers on real roads.

In 2015 Google also unveiled an electric car of its own design, with no steering wheel and no pedals — a fully autonomous machine.

Eight years on, though, Google has fallen behind. In 2016 it announced a partnership with Fiat Chrysler, fitting out more than 100 modified Pacifica minivans as test vehicles for a passenger-carrying self-driving service. Set against building its own driverless car, that was plainly a retreat.

By the standards of a traditional automaker, Google spent roughly enough to build a whole new electric car

Apple once tried to build a car too.

In 2015 Apple bought up large tracts of land near its headquarters, and no shortage of analysts concluded the land would be used to develop self-driving cars. But from 2016 on, Apple's car ambitions shrank sharply. By Tim Cook's account, the goal may have shifted to developing a self-driving system instead.

Google's change of direction looks a lot like Apple's, but took far longer. Apple needed a little over a year to work out that it had taken the wrong road. Google needed seven.

Apple and Google are the most profitable internet companies of the moment, so money isn't the problem. In fact, the $1.1 billion Google spent on cars is a lot even by the standards of a traditional automaker.

Take BMW's all-electric i3. The car reached mass production and went on sale in 2013, and BMW designed and built it from scratch rather than reworking an older model, so what was spent on it is a fair measure of real development costs.

The i3 concept car appeared at the Frankfurt Motor Show in 2011. By industry convention, that means the exterior design was largely settled by then, even if the powertrain was not.

A more finished i3 prototype was shown during the 2012 London Olympics. By that point the technical problems had essentially been solved, and the remaining work was tuning the process for mass production.

According to Wired's report, the i3 program began in May 2008, when the guess was that BMW was building a small electric car. The "i" products announced in 2012 showed that BMW had widened the plan, turning the i series into a full line of electric vehicles.

Forbes reported that the whole i3 project cost an estimated $2 billion, and counting from 2008, the car took BMW five years.

But the i3 has features an ordinary car does not. Its body structure, for instance, is pure carbon fiber, an aggressive choice: BMW built a dedicated plant in Leipzig to make the carbon-fiber bodies, and that plant alone cost $400 million.

The i3 also has no self-driving features at all, so on that count it has nothing in common with Google's work.

But Google underestimated how fast autonomous driving would reach the road

While Google was slowly building a whole new self-driving electric car from scratch, on its own schedule, Tesla's Model S arrived.

Musk announced the Model S plan in 2008, calling it the first important product in Tesla's electric car lineup. In March 2009 the company showed a concept car whose exterior was largely settled; in 2010 Musk showed a Model S prototype, by which point the car's design and development were essentially finished.

The core engineering Tesla did in those years was the battery pack's cooling and central control, plus its patented induction motor.

Tesla bought a new factory, and solving mass production took a long time — the first car wasn't delivered to a customer until June 2012.

Going by Tesla's F-1 filing from its 2012 listing and its 2009 and 2010 financial reports, the Model S burned through roughly $270 million in R&D during its development period, 2007 to 2010.

Next to the BMW i3, Tesla's construction was more "conventional": a metal body built with traditional methods, with small body parts such as the buttons in the cabin supplied by Mercedes-Benz, which must have saved something on design, testing and lining up production.

Tesla's most eye-catching self-driving feature didn't arrive until October 2015, delivered over the air. Because the Model X was also in development between 2010 and 2015, there's no way to say how much the autonomous driving part cost on its own.

What is certain is that Tesla's version 1.0 of autonomous driving ran on algorithms from the Israeli company Mobileye and used the cheaper approach of cameras and visible light, which should have cost far less than Google's lidar.

Tesla spent less than Google. But Tesla's cars, with self-driving features, got onto the road.

Then came the departures and the pivots

Google's self-driving program started out with a great deal of "standing." Mostly that meant the people running it were among the best scientists in the field.

At the beginning the team was led by Sebastian Thrun, former director of the Stanford Artificial Intelligence Laboratory and co-inventor of Google Street View. A computer science professor at Stanford, he led the development of the robot car Stanley, and he and his team won the driverless car Grand Challenge run by DARPA, the U.S. Defense Advanced Research Projects Agency.

Inside Google the team began with 15 senior scientists and engineers, among them Chris Urmson, Mike Montemerlo and Anthony Levandowski — on paper the strongest AI research group at any tech company.

Today every senior member of that team has left.

As Tesla, traditional automakers and internet companies all poured into autonomous driving, the market as a whole heated up.

At the same time, the executives on Google's driverless project had less and less reason to stay.

For one thing, the stock they were granted when they joined had vested, which made them financially free. For another, Google's progress wasn't smooth. With a Google pedigree behind them, starting a company or joining another fast-growing one looked more attractive.

So in 2016 alone, more than a dozen executives left the 500-person self-driving team, including the overall head, the head of hardware, the head of strategic partnerships and the head of computer vision engineering.

A wave of core Waymo departures, and where they went

Early 2016

Head of software engineering and founding team member Anthony Levandowski, founded the self-driving startup Otto, later acquired, joined Uber

Early 2016

Fleet manager Matt Williams, joined Uber

Mid-2016

Head of the project and chief technology officer Chris Urmson

Mid-2016

Computer vision engineering manager and founding team member Zhu Jiajun, co-founded Nuro

Mid-2016

Radar designer Russell Smith, went to Nuro to run hardware

Mid-2016

Sensor engineer Nick Vines, joined Nuro

Mid-2016

Software engineer Don Burnette, joined Otto

Mid-2016

Software engineering manager Dave Ferguson, co-founded the self-driving startup Nuro

Fall 2016

Product manager David Lu, joined Google Ventures

2016

User experience engineer Brian Cullinane, joined Uber

2016

Self-driving training and safety operator Brian Torcellini, joined Uber

2016

Hardware engineering manager Chris Stoffel, joined the self-driving company Zoox

Fall 2016

Hardware director Bryan Salesky

Fall 2016

Head of strategic partnerships Cynthia Kwon, joined Uber

Mid-2015

Software engineer Jur van den Berg, founding member of Otto

You can see that in 2016, around the time their boss Chris Urmson left, a great many people walked out of the Waymo team. Nearly all of them joined another self-driving company afterward.

Which says the problem wasn't autonomous driving — that remained the center of the industry. The problem was Waymo.

In 2015 John Krafcik, a professional manager and former Hyundai Motor CEO, joined Waymo as its chief executive. He knew the car business, and after he arrived Waymo changed strategy completely. The Wall Street Journal reported that from the start of his tenure he was in charge of talks with Ford about building electric cars together. Krafcik decided the project cost too much, and it was killed.

Under Waymo's latest plan, the company is a technology supplier: it takes vehicles from Fiat Chrysler, retrofits them, and calls that the collaboration. Compared with building its own car, or building one with Ford, this is a less aggressive and more practical plan.

Google's self-driving program has produced no big commercial result in all these years, and slow commercial progress is probably why the veterans cashed out and left.

Becoming a technology supplier is the choice most tech companies make; it's what Intel and Nvidia both do. For carmakers, it also turns Google from a competitor into a partner. Car companies' profit margins are nowhere near a tech company's, and a tech company building cars would give them a strong sense of crisis.

Today's Waymo is content to supply technology, and it has even started talking about self-driving trucks, a proposition still less threatening to the car industry.

CEO John Krafcik said in a recent speech that car sharing and ride-hailing, along with freight trucking, are the likeliest uses for autonomous driving.

Those, he said, are the two applications Waymo is now mainly working on — and trucking is something Waymo had never mentioned or demonstrated before.

Another pivot.

And Otto, the company founded by Anthony Levandowski, the founding team member who left Waymo, does exactly one thing: self-driving trucks.

Google went first and piled up more real-road data than anyone. It now holds roughly the same position as latecomers such as Intel, with strategic goals that look much like those of the startups its former employees left to found. That is probably not what Sebastian Thrun and Chris Urmson had in mind back in the Google X days.

Graphics: Feng Xiuxia

Cover image: Google