Discover more from Understanding AI
The death of self-driving cars is greatly exaggerated
GM’s Cruise aims to turn self-driving into a billion-dollar business.
I’m a journalist with a computer science master’s degree. In the past I’ve written for the Washington Post, Ars Technica, and other publications.
Understanding AI explores how AI works and how it’s changing the world. If you like this article, please sign up to have future articles sent straight to your inbox.
Seven years ago, hype about self-driving cars was off the charts. It wasn’t just Tesla CEO Elon Musk, who has been making outlandish predictions about self-driving technology since 2015. In 2016, Ford set a goal to start selling cars without steering wheels by 2021. The same year, Lyft predicted that a majority of rides on its network would be autonomous by 2021.
None of that happened. Instead, the last few years have seen brutal consolidation. Uber sold off its self-driving project in 2020 and Lyft shut down its effort in 2021. Then last October Ford and Volkswagen announced they were shutting down their self-driving joint venture called Argo AI.
Today a lot of people view self-driving technology as an expensive failure whose moment has passed. The Wall Street Journal’s Chris Mims argued in 2021 that self-driving cars “could be decades away.” Last year, Bloomberg’s Max Chafkin declared that “self-driving cars are going nowhere.”
But a handful of well-funded projects have continued to plug away at the problem. The leaders are Waymo—formerly the Google self-driving car project—and Cruise, a startup that is majority owned by GM.
These companies don’t believe self-driving technology is “decades away” because they’re already testing it in Phoenix and San Francisco. And they are preparing to launch in additional cities in the coming months. Waymo expects to increase passenger rides 10-fold between now and the summer of 2024. Cruise is aiming for $1 billion dollars of revenue in 2025, which would require something like a 50-fold expansion of its current service.
There is no guarantee they will succeed. Even if they iron out all the technical problems, it will take many years to make these services profitable.
But I think the pendulum of public opinion has now swung too far in the pessimistic direction. Self-driving technology has steadily improved over the last few years, and there’s every reason to expect that progress to continue.
“It’s definitely happening a lot slower than people anticipated back in 2017,” industry analyst Sam Abuelsamid told me. “But that doesn't mean that there isn't progress being made.”
“I would not be surprised if by the end of 2025 each of those companies is operating in 10 to 12 cities across the US to varying degrees of scale,” Abuelsamid added.
“Smoother and more confident”
Alex Roy has had a colorful career. He’s been a rally racer and a journalist. He joined Argo AI in 2019 and stayed until it was shut down last year. Roy now lives in the Phoenix area doing consulting work related to self-driving cars. He also hosts an excellent podcast about the self-driving sector.
In short, Roy knows a lot about cars in general and self-driving cars in particular. And he has nothing but good things to say about Waymo’s driverless taxi service in the Phoenix area.
“I've now taken several Waymo rides and they're exceptionally good,” Roy told me.
When I talked to Roy last Thursday, he had just taken a Waymo ride from the Phoenix airport to his home in Scottsdale. Technically, Waymo cars don’t pick people up at the airport—airport pickup areas are still too chaotic for that—but they do the next best thing, serving two stops along the airport’s fast and free Sky Train.
I also recently talked to Joel Johnson, a Phoenix area college student who has created dozens of YouTube videos of his rides in driverless Waymo vehicles. Johnson told me that Waymo’s service has been steadily improving over the last three years.
A big step came late last year with the debut of Jaguar I-PACE SUVs outfitted with Waymo’s fifth generation hardware. Johnson told me that the new vehicles represented “a huge leap” in performance over Waymo’s previous Chrysler Pacifica minivans and were “measurably better in many respects.”
The new driverless Jaguars were “much smoother and more confident,” Johnson said. An added bonus: whereas the Pacifica’s trunk was filled with computer hardware, the Jaguar’s trunk was empty and available for passenger use.
Waymo and Cruise are preparing for expansion
Earlier this month, Waymo announced it was doubling the size of its Phoenix taxi service to 180 square miles. Waymo also offers driverless rides to a hand-picked group of passengers in San Francisco. And last October, Waymo announced it was preparing to expand to Los Angeles. Overall, the company is aiming to grow the business 10-fold—from 10,000 to 100,000 weekly trips—by the summer of 2024.
For its part, Cruise operates a driverless commercial taxi service in a portion of San Francisco during the overnight hours. Cruise has begun testing a driverless taxi service that operates 24/7 across the city of San Francisco and is now awaiting regulatory approval to open it to paying customers. In the Phoenix area, a small number of driverless Cruise vehicles are providing both taxi rides and Walmart grocery deliveries. Cruise also runs a small driverless taxi service for paying customers in Austin.
Last week Cruise announced it had completed two million driverless miles just three months after reaching one million miles in March. Earlier this month, Cruise announced plans to expand to Dallas and Houston.
This isn’t the first time self-driving companies have announced optimistic growth forecasts. Back in 2018, I wrote about Waymo ordering 62,000 Chrysler Pacificas, a sign that the company thought it was ready for large-scale deployment back in 2018.
Instead, Waymo spent several more years testing its service in a small corner of the Phoenix metro area and didn’t start offering driverless rides to paying customers until 2020. Today its commercial fleet numbers in the hundreds of vehicles—far fewer than 62,000.
Cruise, too, has rolled out its service more slowly than expected. Back in 2018, Cruise was planning to launch a driverless commercial service in 2019. In reality, Cruise didn’t begin driverless operations until 2021 and didn’t start charging for the service until last year.
So are gullible journalists like me about to be disappointed again? It’s possible, but this time feels different. Waymo and Cruise are already running driverless commercial services, so they have a much better idea of what’s required than they did in 2018. I expect to see significant expansion over the next year or two—though it’s certainly possible they won’t meet their aggressive growth targets.
The challenges they’ll face in the next phase of the expansion will be different than those they faced in the past. Until now, Waymo and Cruise have been almost exclusively focusing on safety. Now they need to figure out how to turn a profit—without compromising safety in the process. That won’t be easy, but it seems doable.
When in doubt, hit the brakes
On March 18, 2018, a prototype Uber self-driving car slammed into Elaine Herzberg as she walked her bike across a road in Tempe, Arizona. First and foremost, Herzberg’s death was a tragedy for her family. But it was also a catastrophe for Uber’s self-driving project and a turning point for the broader self-driving industry.
Uber’s self-driving division never really recovered from the crash, and Uber sold it off in 2020. The rest of the industry vowed not to repeat Uber’s mistake. This focus on safety explains a lot about the evolution of the industry over the last five years.
During the industry’s early years, every self-driving vehicle had a safety driver behind the wheel. If a car encountered a situation it wasn’t sure it could negotiate safely, it would signal to the driver to take over.
Once cars went fully driverless, this was no longer an option. Instead, when a Waymo or Cruise vehicle encounters a situation it isn’t sure how to handle, it will slow down and stop. Sometimes, the situation will resolve itself and the car can move again on its own. Otherwise the vehicle phones home and asks for remote guidance.
This strategy works well in the suburban, residential Phoenix neighborhoods where Waymo started out. Streets there are wide and well-marked, speed limits are low, and there are few cars or other obstacles.
But it doesn’t always work well in more complex environments. Some urban neighborhoods are so crowded and chaotic that self-driving cars encounter tricky situations all the time. These neighborhoods can leave self-driving cars nearly paralyzed, inching forward at a slow and unpredictable pace. This isn’t necessarily dangerous, but it can be very frustrating for others on the road. Stories about Waymo and Cruise vehicles causing traffic jams are a staple of local media in both Phoenix and San Francisco.
I believe this is why Waymo first launched its service in suburban Phoenix, and why the company’s early tests in San Francisco excluded the high-density neighborhoods in and around downtown. Cruise pursued a similar rollout strategy for its driverless vehicles, initially offering service in residential neighborhoods in the northwest corner of San Francisco.
But as I argued last year, this creates an economic problem for these companies. Airports and urban downtowns are two of the most popular destinations for taxi riders. More generally, more crowded and chaotic places tend to be places where demand for taxis is high because these tend to be precisely the places where parking is scarce.
As their technology has matured, Waymo and Cruise have both gradually pushed their services into denser and more chaotic areas. Waymo now serves downtown Phoenix. And as we’ve seen, it serves Sky Train stations near the airport, if not the airport itself.
For its part, Cruise has expanded its service territory to cover most of San Francisco and is waiting for regulatory approval to begin taking paying customers across the city.
Waymo and Cruise still face significant challenges
While Waymo and Cruise have made progress toward serving chaotic urban neighborhoods, freeways have remained off limits. I suspect the reason is that the high speeds on freeways make it impractical for cars to come to a stop when they get confused. So a company has to be very, very confident in a vehicle’s driving abilities before putting it on a freeway without a safety driver.
And for some trips that makes these services drastically less useful.
A recent YouTube video showed a head-to-head competition between a Waymo vehicle and a Tesla equipped with Full Self Driving technology. Both cars drove the 21 miles from the Phoenix Art Museum to the Chandler Museum. Tesla took freeways—mainly Interstate 10—and arrived in 27 minutes. Waymo avoided freeways and took twice as long.
Sometimes self-driving cars take slow routes for no obvious reason. Earlier this month the Guardian journalist Johana Bhuiyan took a Cruise ride in the Western portion of San Francisco. The most direct route should have taken about 15 minutes, but the Cruise vehicle chose a circuitous route that was three times as long.
I’m not sure why Cruise chose such an odd route. There don’t seem to be any freeways, airports, or other challenging situations along the more direct route. A Cruise spokeswoman wasn’t able to explain it either, stating only that it “could be attributed to a number of factors” such as “road and weather conditions, traffic patterns and other factors.”
Waymo and Cruise have also struggled to interact properly with police officers and firefighters:
In January, a Cruise vehicle entered an area where San Francisco firefighters were fighting a fire. According to a letter from San Francisco transportation authorities, “firefighters on the scene made efforts to prevent the Cruise AV from driving over their hoses and were not able to do so until they shattered a front window of the Cruise AV.”
In February, police officers tried to stop a Waymo vehicle from running over a firefighter’s hose. Body camera footage shows a cop lighting a flare, placing it in front of the vehicle, and shouting “no! You stay!” like the car was a misbehaving puppy.
In March, Cruise vehicles got confused following a storm that downed some trees and power lines. Police extended yellow caution tape across the road to warn drivers, but a Cruise vehicle ignored the tape and wound up with it wrapped around the sensors on its roof.
It’s appropriate to criticize Waymo and Cruise for incidents like this, but we should also maintain a sense of perspective. While these incidents were undoubtedly frustrating for everyone involved, they don’t seem to have posed a serious danger. Incidents like this help the companies learn and improve their technology. Hopefully, Waymo and Cruise engineers are hard at work improving their ability to recognize yellow police tape and fire hoses, so that over time these mistakes become less and less common.
A race against the clock
While Waymo and Cruise have steadily improved their technology, their commercial rollout has been excruciatingly slow. Now both Waymo and Cruise are coming under pressure to expand more rapidly.
The reason: projects like Waymo and Cruise are fantastically expensive. GM said last year that it expected to spend $2 billion on Cruise in 2022. Waymo hasn’t disclosed its spending, but with 2,500 employees its annual costs are likely north of a billion dollars.
With interest rates rising, companies everywhere are looking for ways to trim costs. Last year, Ford decided that its own self-driving subsidiary, Argo, wasn’t worth the cost. If I were in charge of Waymo or Cruise, I’d be worried about my corporate parent making the same decision.
So these companies need a credible path to profitability. And with an overhead exceeding $1 billion, that’s going to require a lot of taxi rides.
Sam Abuelsamid, the industry analyst, told me that he “did some math” on Cruise’s plan to generate a billion dollars in revenue in 2025. He concluded that the goal was “actually probably achievable.” He estimates that the goal corresponds to a fleet of around 6,000 driverless vehicles, which he characterizes as “not very many” relative to the hundreds of thousands of taxis and ride-hail vehicles in the US, to say nothing of the world.
But Abuelsamid stressed that $1 billion in revenue would be “not anywhere close to profitability.” To break even, Cruise would need to continue expanding rapidly in 2026 and beyond.
I suspect Waymo can count on a little more patience from its corporate parent. Alphabet generated nearly $60 billion in profit in 2022, so spending a billion or two annually on self-driving cars isn’t a big burden. Alphabet is controlled by its founders, Larry Page and Sergey Brin, so the company doesn’t need to worry about pressure from Wall Street.
But Alphabet’s patience won’t be unlimited. At some point, Larry Page is going to want Waymo to become a viable business, not just a research project. And as with Cruise, that will require many years of rapid growth.
And to be honest, this makes me nervous.
So far, the strategy of avoiding freeways, downtowns, and airports has made Waymo and Cruise safe but unprofitable. Eventually, I expect their software will improve enough that they can operate in these challenging environments safely and confidently. But there’s no guarantee this will happen within the next year or two. And so the leaders of these companies may come under pressure to push into these more challenging environments before they’re ready.
I hope I’m wrong about that. These companies have been careful so far and their leaders certainly realize that even a single death would be immensely damaging. But it’s hard to judge the risk of something that’s never happened. So there’s always a risk that Cruise and Waymo’s leaders will become overconfident the way Uber’s leaders did five years ago.
What about Tesla?
I expect some readers have found this article to be frustrating because I have barely mentioned Tesla’s Full Self Driving software, which Tesla fans view as the industry leader. This is because I view Tesla as operating in a different market from Waymo and Cruise. Tesla is building a driver-assistance system that is designed to be used only with direct human oversight, while Waymo and Cruise are trying to build cars that can drive entirely on their own.
Of course, that’s not how Tesla fans see it. They expect Full Self Driving technology to continue improving to the point where human supervision becomes unnecessary. At this point Tesla will supposedly launch a driverless taxi network that competes directly with Waymo and Cruise.
I’m skeptical of this. I’ve watched a bunch of videos of FSD in action in recent days and it’s still pretty common for drivers to intervene when a car gets into a tricky situation. This suggests that Tesla’s technology is roughly on par with Waymo’s technology circa 2016, a year when Waymo vehicles had a safety-related disengagement once every 5,000 miles.
I think a big reason Tesla fans have a misperception that FSD is ahead of Waymo and Cruise is that Tesla’s FSD operates in more situations, including freeways. But that misunderstands what’s going on. Waymo has been testing its technology on freeways for more than a decade, it has just been doing it exclusively with safety drivers. Tesla only tests its vehicles with a driver behind the wheel, so of course FSD is able to go on the freeway too. But this isn’t evidence of Tesla’s superior performance on freeways, it just reflects Waymo’s more cautious approach and different business model.
It also seems implausible that Tesla will just flip a switch one day and allow its vehicles to operate driverlessly everywhere. More likely they will want to follow in Waymo’s footsteps and test out driverless functionality in lower-risk areas first, then gradually expand to riskier environments. This process has taken Waymo six years so far, and while Tesla might be able to do it faster I doubt it will be that much faster.
One reason is that there are some problems that a company will only encounter once it begins testing fully driverless operations. Waymo and Cruise’s problems with fire hoses and caution tape is a good example. A human driver would disengage FSD long before it got into situations like that, which means Tesla would be unlikely to have the training data necessary to train its cars to handle it properly.
Tesla also isn’t laying the necessary groundwork to operate a driverless taxi service. Taxi companies need to develop relationships with taxi regulators, police and firefighters, and other officials in cities where they operate. They also need teams of drivers and mechanics in each city to respond when a driverless taxi gets stranded.
As far as I can tell, Tesla hasn’t started doing any of this. And that suggests to me that the company isn’t serious about entering the driverless taxi business. Rather, the company is building a driver-assistance system similar to (if perhaps more advanced than) those offered by other automakers. There’s nothing wrong with that. But it means that Tesla isn’t a direct competitor to Waymo and Cruise.