Here come the driverless taxis

Some of Silicon Valley’s top engineers have spent the last decade writing code for driverless cars to expertly manoeuvre around the most complex streets in San Francisco.

But, faced with an orange traffic cone deliberately placed on the hood, the cars experience the robotaxi equivalent of a PC screen turning blue: they become paralysed and clog up traffic.

Activists from the anti-car group Safe Street Rebel are not sure who came up with the idea of “coning” — but as soon as they realised how easy it was to bring any of the city’s hundreds of robotaxis to a halt, they filmed the disruption, posted it on the web and encouraged others to do the same. The antics, known as the “week of cone”, are part of a wider backlash against robotaxis in San Francisco, one of the first US cities where such cars are now able to operate around the clock.

Alex, one of the group’s organisers who declined to use his real name, says he wants city officials to place more funding into public transit, not for them to support robotic chauffeurs likely to cause more traffic congestion. The coning campaign, he says, is about making more people aware that autonomous vehicles are overseen by state-level regulatory bodies without any democratic representation.

“A lot of us came together in frustration, a feeling of powerlessness about these robotaxis,” he says. “We saw what you can do with just a traffic cone, and we wanted to tie that into taking some power back, you know? To bring some power to the people.”

It would be easy to dismiss the protest as a fringe social media stunt, but some of their frustrations are so widely shared by top city officials and emergency services that the vote to approve the expansion of robotaxis in San Francisco — seen as a key moment on the journey to national acceptance — was twice delayed this summer. And then it involved nearly seven hours of contentious debate amid demonstrations outside.

Over the past decade, San Francisco has emerged as the unofficial headquarters for driverless cars. GM-owned Cruise and Amazon-owned Zoox are both based there, while Alphabet-owned Waymo has been testing there for years. The chaos of the city is part of its appeal: algorithms learn faster when confronted with the unexpected.

Driverless car executives have touted the potential to bring dramatic safety improvements. Robots do not get sleepy, distracted or enraged. Nor will they drive drunk, talk on the phone or rush to get to work. For passengers who feel vulnerable hailing a ride with a stranger, especially late at night, a driverless car can be a sort of refuge.

But what critics emphasise are the robot-only problems. Their inability to recognise and remove a cone is just one example. All too often, driverless cars get confused and abruptly pull to the side of the road or, worse, sit in the middle of an intersection until a remote employee can extract them. This week, one robotaxi got stuck in wet cement.

These incidents are not widespread, given that large numbers of robotaxis on the road are still in test-mode, often with a safety engineer behind the wheel. But frustrations are likely to grow as more driverless cars hit the roads.

Robotaxis have gummed up the San Francisco’s fire department’s services 55 times in the past year, its chief Jeanine Nicholson told the California Public Utilities Commission. “That’s just unacceptable,” she says. “It is not our job to babysit their vehicles.”

Such reservations made it difficult for the CPUC to decide if Waymo and Cruise could operate commercial services, 24/7, with paying passengers.

Although the permits were eventually granted on August 10, the difficulty in obtaining them in one of the world’s most tech-forward cities could foreshadow deeper challenges as these groups plan to expand their services around the globe.

A survey published last year from Pew Research revealed the scope of the problem. It reported that more Americans thought the “widespread use of driverless cars” would be “a bad idea” than those who felt it would be “a good idea”. More than half of men, and 72 per cent of women, said they would not even enter a driverless passenger vehicle if they had the opportunity.

Industry executives believe such numbers will shift in their favour as more people experience their vehicles and recognise the safety merits.

But this is far from certain. Cruise alone plans to have 1mn robotaxis on the road by 2030. So even if the rate of accidents were to decline dramatically, the growing ubiquity of robotaxis would raise the raw number of driverless car crashes, each generating news and social media that could sway public opinion against them.

“Each accident is like a man biting a dog — it generates so much attention [and] goes to court,” Amnon Shashua, chief executive of Israel-based driver-assist tech group Mobileye, said at a recent Financial Times conference. “This is really the stumbling block for doing a wider deployment.”

‘Bigger than Google’

These stumbling blocks matter in part because they could spell further delays.

More than a decade ago, when engineers were first demonstrating that self-driving cars could confidently merge with traffic up and down the Bay Area, Google co-founder Larry Page surmised: “If this business succeeds, it could be bigger than Google.”

The idea convinced investors to pour stunning sums into autonomous tech over the past decade — more than $50bn, according to McKinsey. Everyone from small start-ups to legacy carmakers to Apple and China’s Baidu were taken with the idea.

In the years since, it has been easy to ridicule the most optimistic projections, such as Lyft once saying “a majority” of rides would be driverless by 2021. But if the dream of driverless cars has been deferred, it has not really been revised.

Mary Barra, CEO of GM, has told investors she expects Cruise to generate $1bn of revenues in 2025 — when it expects to have commercial operations in the US, the United Arab Emirates and Japan — and $50bn a year by 2030.

“We’re on track to outpace those projections,” says Kyle Vogt, co-founder and CEO of Cruise. “We don’t want to just have this massive safety, and convenience, and cost benefit in the US. We want to do that wherever people are travelling, which is everywhere on the planet.”

He argues that people will be surprised just how quickly driverless services will arrive and then expand. “We’re going to get to the point where nearly every mile driven in the US is by an AV,” Vogt adds. “There are 3.7tn miles driven [per year]. So you do the math on that: At even a buck per mile, that is a massive industry. It’s far larger than Google.

“And so the question becomes, how quickly can we do it? . . . At this point, there’s very little technical risk that remains.”

Research from McKinsey suggests these projections are too rosy. It projects that, by 2030, global industry-wide revenue for autonomous driving tech will be $25bn at most — half what Cruise believes it alone will earn.

But, like Cruise, McKinsey assumes that once the technology is truly ready for the mainstream, revenues will accelerate to between $170bn to $230bn within five years.

The speed of this deployment would represent a paradigm shift for cities.

Long road ahead

For that mass rollout to happen, driverless groups will need to radically compress the time it takes to be up-and-running in new environments: obtaining the necessary permits to operate a service in San Francisco required multiple years of testing.

Saswat Panigrahi, Waymo’s chief product officer, says it took “a couple of months” to expand to Los Angeles — where the company is testing autonomously with and without a back-up driver — and only “a couple of weeks” to expand to Scottsdale, Phoenix, where anyone can hail a fully autonomous ride.

“That rate is definitely accelerating,” Panigrahi says. “That’s proving that these drivers are generalisable to cities they have never been in before.”

Alex Kendall, chief executive of London-based autonomy start-up Wayve, says his company is using a machine learning approach that will allow its vehicles to understand cities they have not driven in before. The company aims to deploy fully driverless vehicles after local legislation is expected to allow it in 2025.

“You see far less personal car ownership in Europe and the UK, and greater uptake of public transport and other solutions,” he says. “I do think we’re going to see a market there that’s more open, engaged and ‘early adopter’, if the regulatory conditions are right.”

But even if the tech is capable of such rapid advances, the experience in San Francisco underscores the possibility of a wider backlash in new cities. Robotaxis have been conspicuously testing in San Francisco for years, yet Waymo and Cruise struggled to receive permits to commercialise, so why would other cities readily embrace them?

Earlier this year, polling company YouGov surveyed consumers in 18 countries and found that 42 per cent of respondents were concerned by self-driving technology. In China, where Baidu’s Apollo Go robotaxis have won licences to operate in at least four cities, 37 per cent of respondents were “worried”.

Mobileye’s Shashua argues that such scepticism is one of the big flaws of the robotaxi strategy. He, like Tesla CEO Elon Musk, believes that evolving driver-assistance technology from “hands-off” to “eyes-off” will prove better at winning people over once they can try it for themselves.

“You need many, many thousands or hundreds of thousands of vehicles with ‘eyes off’ capability in order to generate data . . . to prove to the public and regulatory bodies that this system meets [safety] requirements,” he says. “It will not come by having 200 vehicles roaming San Francisco.”

Prescott Watson, general partner at Red Blue Capital, a transportation focused venture fund, says the robotaxi groups are still too wedded to the original dream of a decade ago, in which engineers would build a brain, essentially, that could be dropped into any environment and start driving.

This is the sort of thinking that led Uber founder Travis Kalanick, in 2016, to look into buying 100,000 Mercedes, in the belief he could equip them with self-driving tech and then watch the money roll in. Watson says the industry was prepared to spend $20bn upfront, because they believed scaling up could occur at marginal costs.

“That’s turning out to be not the case,” Watson says. “You make a brain, but you have to train that brain for six months to drive in Phoenix, and train that brain for two years to drive in Columbus, Ohio. And each of these cities’ rollout is another massive project.”

Potential payouts

If driverless groups are able to deploy in multiple cities per year, and eventually dozens per year, convincing politicians and the public of their safety benefits is going to be central.

After Bridget Driscoll, in 1896, became the first person to be struck and killed by a horseless carriage travelling no more than 8mph, the coroner expressed a wish that “such a thing would never happen again”.

Motor car crashes have instead become so common that they are no longer news. Globally, 1.3mn people die in car crashes each year, or about 3,700 deaths per day — more than were killed in the September 11 attacks. In the US alone, fatalities totalled nearly 43,000 in 2021, a 16-year high.

These are “horribly large numbers,” Rodney Brooks, a roboticist at MIT, wrote in a 2017 paper, but over the last 120 years “we, the human race, [have] decided that such high numbers of deaths are acceptable for the usefulness that automobiles provide.”

Brooks argued that society was unprepared for deaths involving driverless cars, even if the numbers point to significant improvements. “Ten deaths per year may be deemed too much,” he wrote. “It won’t be rational. But that is how it is going to unfold.”

Tesla’s Musk has recognised the same dilemma, as the electric vehicle maker tries to evolve its “Autopilot” driver-assist technology into its own robotaxi network. “The people whose lives are saved with Autopilot or autonomy don’t know that their lives were saved,” he said at an FT conference last year.

For Shashua of Mobileye, the relevant analogy is vaccines. A life-saving inoculation for the vast majority of the population could still react poorly with a tiny minority of people, he says. “Someone is going to die, just like with vaccination. How do you handle this?”

To date, driverless car accidents have been rare, and in most reported accidents it is a human driver that is at fault. In a report from Waymo summing up its experience from one million “rider-only” miles earlier this year, there were 18 “contact events” and in more than half of the incidents a stationary Waymo was hit by another car.

Only two of the events met the regulatory definition of “collision,” there were no reported injuries, and no accidents involving pedestrians or cyclists. “Ten of them were somebody colliding with Waymo from behind,” says Panigrahi.

Still, multiple law firms in California have already established a speciality in self-driving car accidents. The appeal, beyond helping victims, is clear: if technology from the likes of Alphabet, Cruise and Tesla is to blame, public attention and potential payouts could be massive.

“As crazy as it sounds, traffic accident victims now need to figure out who was driving the car that hit them: a computer or a human?” explains Martin Gasparian, of Maison Law, on his website. If the fault lies with technology, then “the car company could be fully responsible”.

This outcome is one that some in the industry are already bracing for. Musk told the FT that Tesla’s strategy is to prioritise making the tech safer, combating the lawsuits and carrying on: “We’re just going to take the heat,” he said.

Navigating legal minefields will pile pressure on the companies vying to dominate the robotaxi industry. For Watson, the upshot is that robotaxi groups will need immense resources to scale up in the coming decade, and that makes him particularly concerned about Cruise.

It is arguably the frontrunner today, but while Waymo and Zoox are backed by tech giants equipped with money-printing presses, Cruise is backed by a carmaker already battling with unions as it spends billions of dollars electrifying its portfolio.

“The path towards profitability is going to be very long for these companies,” he says. “They need to have owners that are willing to put up with that timeline.”

Graphic illustration by Ian Bott