For over a decade, companies from Google to General Motors have poured billions of dollars into the pursuit of what was seen as the Holy Grail of driving technology: the fully self-driving car. Such vehicles would usher in an era of consumer safety and convenience, experts promised, and would be an immensely valuable product for carmakers.
But recently many of the main players in the autonomous vehicle game have been scaling back or outright abandoning their lofty ambitions. Last week Ford and Volkswagen pulled the plug on their self-driving effort, Argo AI, the latest admission from a hype-fueled industry that building a good self-driving car that’s also a profitable business may not happen anytime soon.
When it all began
Argo AI CEO and co-founder Bryan Salesky was part of a famed Carnegie Mellon University team that developed a primitive self-driving vehicle that won a Pentagon race in 2007.
Four students on the team went on to co-found self-driving companies, which have raised billions since: Salesky, who founded Argo; Dave Ferguson at Nuro; and Drew Bagnell and Chris Urmson at Uber-backed Aurora.
“We started out doing this stuff because it was cool and it was a neat idea, but we weren’t quite sure how it was going to be used,” Salesky said in a 2019 interview.
The team’s self-driving Chevrolet Tahoe caught the eye of Google co-founders Larry Page and Sergey Brin, who went on to launch a self-driving car program in 2009, later dubbed Waymo, that ignited an industry. Competitors like Uber jumped in, fearful that being left out of self-driving would destroy their business’s future. (Uber launched its program in 2015 before selling it in 2020 following a costly lawsuit with Alphabet.)
General Motors bought a self-driving company in 2016. Ford followed suit a year later with Argo AI. Uber rushed to scoop up engineers from Salesky’s alma mater, Carnegie Mellon.
More than $10 billion was invested in self-driving cars since 2010, McKinsey estimated.
Proponents of self-driving believed they were onto something big. Driving is the deadliest of the common forms of transportation, and a leading cause of death for many age groups. More Americans have died in traffic crashes than fighting in all of its wars.
Investors swooned at the potential to make billions while saving millions of lives.
This business would be huge, they said. “$7 trillion as autonomous vehicles become mainstream,” claimed Intel.
Big talk, smaller results
Urmson, while leading Google’s self-driving car project before founding Aurora, talked of his preteen son never needing to get a driver’s license. Optimism ran wild.
Why would anyone even own a car? Autonomous taxis would be even cheaper than walking, said one industry analyst.
“Autonomous vehicles are able to acquire new skills faster than humans,” Salesky said in 2021.
But there was one problem. The technology could do impressive things but mastering all the situations we face as human drivers is tough. The billions invested in the technology haven’t yet been able to drive better than a human in some situations.
That’s been a problem for the long list of companies that set aggressive targets for self-driving cars. General Motors in 2017 promised mass production of fully autonomous vehicles in 2019. Lyft said in 2016 that half its rides would be self-driving by 2021. Ford talked up 2021 too, promising full self-driving vehicles deployed broadly.
But there’s a funny saying from the software world, known as the 90-90 Rule. Once 90% of the work is done, you only have 90% to go. The ability of self-driving car software to steer the vehicle within a highway lane is great, but being able to do so, even for thousands of miles at a stretch, isn’t enough. A self-driving car, without a steering wheel or pedals, would have to be able to drive itself in literally every situation possible.
With millions of miles being driven by humans every day, the number of tricky and unusual situations, called edge cases, is vast. Unusual events are individually, by definition, uncommon. But the sheer number of unusual events encountered by millions of drivers all over the world means that “unusual events” are quite common.
“It’s really, really hard,” Waymo’s then-CEO John Krafcik said in 2018 of self-driving technology. “You don’t know what you don’t know until you’re really in there and trying to do things.”
Industry leaders have backed off their big claims. Waymo has said it will be decades before the technology is everywhere in the world.
Tesla CEO Elon Musk now largely stands alone for sticking with aggressive predictions. He’s said every year since 2017 that the software capable of full self-driving will probably arrive in the “next year.” The software hasn’t arrived though. Instead, the company has drawn increasing scrutiny from regulators for its big talk. It’s being investigated for the rear-ending of emergency vehicles stopped on roads, and appeared unable to identify motorcycles at times.
Humbled but not defeated
In recent years many industry leaders have quietly acknowledged and tried to tweak their business plan. Urmson and Bagnell shifted to the development of self-driving trucks that drive mostly on highways, which are much easier to master.
Ferguson left the Google self-driving car project to start a company building fully autonomous vehicles without a person inside for goods delivery. It’s a far simpler challenge as there’s no worry about protecting people inside the vehicles. Even Google itself started working on self-driving trucks.
Companies developing lidar, widely seen as a key component for self-driving vehicles, as well as self-driving companies, have seen their stocks plummet recently.
Lidar companies Velodyne, Quanergy, Luminar, and Ouster have all seen their stocks pummeled this year. Autonomous truck companies Aurora, TuSimple, and Embark have all had huge dives too this year.
Tesla, which has promised “full self-driving” for years, has delivered a less ambitious driver-assist feature that’s enjoyed by some, but it is also a beta product with lots of room for improvement. It’s also facing several government investigations related to the technology.
Only VW and Ford’s Argo AI and GM’s Cruise stayed focused exclusively on true self-driving cars.
Now only one of them remains.
GM’s Cruise offers a nighttime ridehail service in San Francisco, which completed 84 trips in June, 224 in July and 416 in August. (Driving in daytime is much harder as there’s more traffic including pedestrians and cyclists.) Ride fees are similar to riding with Uber or Lyft.
Cruise has lost $1.4 billion this year, and it cannot stand to lose hundreds of thousands of dollars per trip forever. It plans to expand service to offer rides in Austin and Phoenix later this year.
Cruise CEO Kyle Vogt said last week that “next year marks the beginning of our rapid scaling phase.”
There will be plenty of kinks to work out. The local government has raised safety and traffic concerns about Cruise’s expansion plans.
Some longtime industry observers caution that self-driving technology still has a long way to go, a line of thinking Ford CEO Jim Farley echoed Wednesday.
“Things have changed,” Farley said of Ford’s 2017 plans to broadly deploy self-driving cars in 2021.
The company is instead focusing on advanced driver-assist technology.
Automakers, aside from Tesla, largely rely on Mobileye, an Israeli tech company, for driver-assist technology. Intel spun out Mobileye last week, raising $861 million. The successful IPO suggests there’s still a robust market for driver-assist technology, even if it falls short of fully autonomous vehicles.