This is the third installment in an investigative series. Power lights up our city, but it also animates struggles between officials’ promises and self-interest, and between government and private corporations over control of our energy future. We’re supposed to have a more affordable, livable city with less pollution, congestion, and street mayhem. But many obstacles are in the way.
Parts 1 and 2 of the series investigated SF’s relationship with PG&E and a potential city takeover of its own power grid. We’d love your feedback. Email hello@thefrisc.com with the subject line ELECTRIC, or find us on social media. Thanks for reading.
On October 24 last year, the California Department of Motor Vehicles revoked the operating permit of robotaxi firm Cruise after a spate of serious incidents. Among them: One of its self-driving cars dragged a pedestrian 20 feet along a downtown street after she had been struck by a human hit-and-run driver. Another collided with a fire truck, injuring one of the taxi’s passengers.
The edict ended Cruise’s 16-month run in San Francisco and laid down speed bumps on what was previously an open road for this automated, electrified future of urban transportation.
But there was another significant robotaxi development that day, and far fewer people noticed. In a hearing, the San Francisco Board of Supervisors — serving as the SF County Transportation Authority — watched a short video.
In it, a Cruise vehicle accelerates toward two women and two children in a crosswalk on Scott Street at Sacramento. The robocar brakes briefly, then swerves around them. Julia Friedlander, senior manager of automated driving policy for the city’s streets and transit agency, was narrating the video, which had been posted on social media. “This raises serious concerns for us,” she told the supervisors.
Cruise is gone, but its rival Waymo still runs a robotaxi business here. The close call in the video underscored that high-profile crashes aren’t the only impact these vehicles are having in the city. SF now collects egregious examples, many captured by first responders who, out of fear and frustration, began documenting their own experiences last year.
Still, many unusual incidents – call them close calls, call them unsettling events – aren’t recorded, and the robotaxi makers have been reluctant to hand over data to city officials, citing trade secrets and rider privacy. What little information they divulge goes to state and federal regulators.
In fact, San Francisco has no broad jurisdiction over self-driving cars, prompting some local lawmakers to wage “legislative guerilla warfare” where they can.
Friedlander has worked exclusively on autonomous vehicle issues for the SF Municipal Transportation Agency since 2018. She often hears robocar makers say they want the same thing as SFMTA: safer streets. But she echoes other researchers and observers of the field who say self-driving cars are not ready for wide deployment.
One concern is that they could replace safe and efficient ways of moving people around SF: public transit, walking, or rolling. Another concern is that robotaxis will prioritize on-demand customer service over public safety.
Cruise raced to get as many cars onto SF streets as possible, which backfired. The final straw for regulators was its failure to provide full video of the pedestrian-dragging incident.
The Justice Department and Securities and Exchange Commission have opened investigations. General Motors had invested $10 billion in the firm over a decade, but after Cruise lost its permit, GM took over and cleaned house.
Meanwhile, rival Waymo’s more conservative rollout — with only 200 cars on the street on any given day, according to a spokesperson — is paying off. The firm, owned by Google’s parent Alphabet, now has state approval to move into San Mateo and Los Angeles counties.
Other companies are also moving into SF, including Amazon’s Zoox, which has a driverless testing permit for only a few South of Market streets, according to the DMV.
Proponents say robotaxis will make roads safer, and the companies themselves have touted their own studies. A new independent metastudy of nearly two decades of previous work suggests self-driving cars are less prone to crashes. But there are major caveats: they fare five times worse than human drivers at dawn or dusk, and are twice as likely to crash when making turns.
There’s little San Francisco can do about it right now. A state lawmaker’s bill to shift robotaxi oversight to local governments was shelved this week in the legislature.
It’s enough to give city officials a sense of deja vu. When Uber and Lyft launched more than a decade ago, SF had no say, and for years transit officials tried in vain to get information about operations from the companies and state regulators.
Finally, even with incomplete and heavily redacted information, the SF County Transit Authority issued a 2023 report showing how ride-hail operations clogged the city’s streets and fouled its air. SF had 500 times more trips per square mile than any other city in California from September 2019 to August 2020.
With federal and state regulators holding nearly all jurisdiction, San Francisco seemed in a similar bind with robotaxis. Then the fire department got angry.
Hot under the collar
In the summer of 2023, the California Public Utilities Commission had a big vote looming: whether to give Waymo and Cruise unlimited access to SF streets for commercial taxi service.
For a year, their limited service had turned heads, surprising locals and visitors alike with the sight of a car motoring along with no one at the wheel. But their misadventures were the stuff of headlines and social shares: they would “brick” – stall in the middle of a street and block traffic. They would also drive into active emergency operations, or fail to pull over to let a fire truck pass.
First responders began keeping tabs. Their reports include an incident in the Mission where a Waymo had stopped on narrow Julian Avenue and blocked a fire department vehicle responding to a medical emergency. The truck had to back up and go around the block, causing a four-minute delay, according to SFFD records.
Some incidents put first responders in danger and damaged property. A Cruise car entered the scene of a non-injury car crash near Golden Gate Park; a police officer was helping a fire truck back into the intersection. The officer jumped out of the way of the Cruise, but the distraction caused the fire truck to strike a parked car.
In an incident earlier this year captured by a visitor and provided to The Frisc, two fire engines blocked a street during an emergency on Nob Hill. The visitor’s Uber tried to back up but was blocked by three Waymos as firefighters tried to untangle the mess.
In addition to the safety problems, these incidents require a lot of extra paperwork. First responders are under orders to provide copious details. (One SFFD report came with an extra note: “Writer admonished for sparse details and is reminded that these reports are to help the chain of command understand what issues are arising from driverless cars.”)
When the CPUC met to consider Cruise and Waymo’s expansion, SFFD Chief Jeanine Nicholson testified that self-driving cars weren’t “ready for primetime” and cited 34 incidents. Her warnings – and many others – went unheeded. The regulators gave a green light to full driverless service on Aug. 10.
Tracking strange behavior
While first responders were keeping their own file, SFMTA had already begun collecting mentions of “unusual occurrences,” according to Friedlander.
The agency has relied on many second-hand reports via social media, as well as city employees and some regular citizens who notify the agency.
Over the course of 2023, SFMTA counted 877 individual incidents. After the CPUC gave Waymo and Cruise carte blanche in early August, incidents reached a peak of 200 that month.
However, the count dropped all the way down to 10 in November, soon after Cruise’s permit was revoked. There were, of course, fewer robotaxis out and about. But Friedlander acknowledges the reduction might be about quality as well as quantity: Considering Waymo’s time in service, “it’s reasonable to think they were doing a better job last year.”
She also notes, “Neither the city nor state or federal regulators receive data that enable us to make a clear judgment on this.”
People have been seriously hurt by driverless vehicles. In Arizona in 2018, a woman on a bike was killed by one. (The woman dragged by a Cruise in SF last October received more than $8 million in a settlement earlier this year.)
These incidents are tragic. But the companies, when comparing their records to human drivers, stress that humans cause plenty of harm. Pedestrian deaths are at an all-time high nationwide. SF pledged in 2014 to eliminate traffic deaths by 2024 but hasn’t come close. As of June 10th, there have been 14 people killed in traffic crashes this year, 10 of them pedestrians – all killed by humans behind the wheel.
There are a lot of little novelties that can be slightly unexpected for people using the street who are not inside a car. These uncanny events are mundane, until they aren’t.
Melissa Cefkin, a social and behavioral researcher who worked on autonomous vehicles at Nissan and Waymo
Until the robotaxis reduce or eliminate close calls and other spooky encounters, it will be hard to convince many people of their safety. “There are a lot of little novelties that can be slightly unexpected for people using the street who are not inside a car,” says Melissa Cefkin, an independent consultant who previously worked on autonomous vehicle behavior at Nissan and Waymo.
Let’s say a cyclist is at a stop sign across a four-way intersection from a Waymo car. The cyclist wants to turn left but isn’t sure the robotaxi will recognize a left-turn hand signal – or “see” them at all. The question for anyone not in a car is, without eye contact or a wave of the hand, how to know what the driverless car will do.
Cefkin, a social and behavioral researcher, says these “uncanny events” involve things that “are mundane, until they aren’t.” She and others are working on these problems, but no one “has an answer for this yet.”
Too early for a ‘gold medal’
In spring 2023, before the pedestrian dragging and other incidents, both Cruise and Waymo passed the milestone of one million driverless miles without collisions with serious injuries. Cruise compared its driverless record with that of GM’s now-defunct ride-hail service that used human drivers: Cruise said its cars caused 92 percent fewer collisions.
In December, Waymo released two papers using publicly available crash data for both human drivers and Waymo technology. All the study authors were employed by Waymo, but spokeswoman Julia Illina says, “We shared the paper with several experts in the field prior to publication to get feedback and validate our methodology.”
A prominent safety expert says these comparisons can be misleading. When the U.S. National Highway Transportation and Safety Administration looks at fatality rates for vehicles, it bases estimates on at least 300 million miles driven, said Phil Koopman, associate engineering professor at Carnegie Mellon University.
As of spring 2023, Waymo and Cruise each totaled about 5 million miles – including testing with a human behind the wheel. (In its comparison with the GM ride-hail service, Cruise had 5 million miles and the service had 5.6 million miles, both in San Francisco.)
In a video posted to his website, Koopman said company declarations that their cars are safer than humans are akin to running the first mile of a marathon, setting a record pace, and declaring victory: “It’s a little early to give yourself the gold medal, and that’s where we are with this technology.”
Autonomous vehicles rely on cameras and sensors to feed information to in-vehicle computers. Like all software, it needs regular updates and sometimes makes errors.
Waymo has voluntarily recalled its software three times since it began driverless testing in 2015. Two were this year, one because its cars were crashing into poles, and the other after two Waymos hit the same truck being towed. “The Waymo AV incorrectly predicted the future motion of the towed vehicle,” according to a press release. (In May, NHTSA announced an investigation of Waymo’s driving behavior.)
Software isn’t regulated, according to Koopman. The only required national safety standards “have to do with basic functionality like headlights, airbags, and tire pressure warnings, and nothing to do with automated driving features,” he wrote via email. When it comes to software standards, car companies “can do whatever they want.”
The tortoise and the hare
As SFMTA records show, SF streets have gotten quieter (in robotaxi terms) since Cruise was kiboshed last October. It’s starting to look like the old fable of the tortoise and the hare. Cruise was the hare: brash, aggressive, and a failure. Waymo, so far, is the tortoise. It could have put far more cars on city streets – its permit does not have a limit – but it has kept a lower profile.
Waymo’s cars are popular with tourists and locals, and they’ve built up goodwill with state regulators, who approved Waymo’s LA and San Mateo expansions.
During the approval process, the vast majority of public comment was positive. Organizations representing the blind, disabled, and seniors weighed in, along with local nonprofits that aid underserved communities. These groups see in Waymo – and other self-driving companies – a lot of promise.
Former Nissan and Waymo researcher Melissa Cefkin thinks self-driving cars can deliver, noting that paratransit, a common current solution for seniors and others with mobility issues, is “painful to use,” in that it can’t always come when you need it. “I appreciate how significant it is for disabled people to feel independent and get around,” Cefkin says.
Waymo might be expanding in tortoise-like fashion, but it’s also thinking of a big leap forward: autonomous delivery, a potentially far more lucrative market. Imagine all those electric delivery vans, puttering around SF without drivers (and practically everywhere else in the country), making thousands upon thousands of stops in complicated street situations, and needing lots of space to recharge.
As we’ll see in the next part of our series about San Francisco’s electric future, safety might not be the largest concern about robo-delivery. The threat to life and limb is one thing, but the threat of losing jobs is altogether another.
This series is supported by the Fund for Investigative Journalism.
Correction, 6/21/24: We have corrected the year of the death of an Arizona woman killed by a self-driving Uber. The year was 2018, not 2019.