At a junction in downtown Detroit, a self-driving car stops beside a pedestrian waiting to cross the road.
Without touching the car, an electric shuttle operated by start-up May Mobility, its safety driver, Andrew Dykman, confidently waves the man across the road.
"I have a pretty good idea of what the car is going to do at any time. So I'm letting him know that it's waiting for him. I guess it just comes from experience," he explains.
When a car in Uber's self-driving testing fleet struck and killed a woman in Arizona last year, images of the safety driver were beamed around the world. The woman, Rafaela Vasquez, appears to be looking down before gasping in shock and grabbing the wheel as the pedestrian, Elaine Herzberg, is hit. A police report concluded that she had been watching The Voice on her phone.
The incident put the spotlight on a mysterious and new line of work. Sitting in the driver's seat and keeping an eye on a self-driving car in training wasn't anyone's job 10 years ago (apart from perhaps a handful of Google engineers). Now, there are hundreds, potentially thousands of them worldwide.
Companies send their partially developed cars out in public to collect data and to examine how they react to real-life situations. Since none of them are yet fully autonomous, they need a human babysitter.
But what is that job like? Put simply, the role involves sitting in the driver's seat of a semi-autonomous car and making sure it doesn't veer off the road, hit anything, or otherwise go rogue. Sounds simple. But it can also be exhausting, lonely, and potentially very dangerous.
"We all know that it's a difficult thing to [do] monotonous driving for eight hours, even when you're doing the driving. So imagine that, now, your task is to maintain complete awareness, while someone else is driving. And that is a very difficult thing for any human being," says Eric Paul Dennis, of independent non-profit the Centre for Automotive Research.
"There is a higher cognitive load doing this than driving all the time because you're thinking about what the car [will] do," says Dykman, who is May Mobility's technical programme manager.
A car that can safely deal with all situations doesn't yet exist, and its human supervisor is still responsible for everything it does. This means the safety driver has to pay close attention to what is happening around them, while not actually doing anything for long periods of time.
Of course, piloting a cutting-edge semi-autonomous car for hours each day on public roads was always going to be a risky job. But Uber's accident brought the risks into sharper relief.
Uber spokeswoman Sarah Abboud says the company has focused on reducing driver fatigue, telling drivers that they can come back to base at any time if they feel tired or if something is wrong with the car, and always having two drivers in the car (the driver who crashed was alone at the time).
It also keeps a closer eye on its safety drivers. Companies which detect distracted driving by looking at body position and eye gaze are increasingly selling their products to self-driving car companies. "We have a driver monitoring system, so we're actually able in real-time to know if, say, for example, a safety driver behind the wheel is looking down for an extended period of time or not paying attention to the road," says Abboud.
Public transport company Transdev, which is developing self-driving shuttles, has advertised for a self-driving vehicle operator in Phoenix, Arizona, to be paid US$20 ($31) an hour, who can "operate a vehicle six to eight hours a day, alone, five days a week; able to sit still for long periods of time".
The company said it uses cameras to detect driver attention levels and encourages them to ask for someone else to take over if they are tired, without fear of repercussion.
"It is important to note that fatigue and distracted driving are not new problems. As a transport operator, Transdev has always needed to do all we can to make sure our drivers are focused and alert," the company said.
As the public begins to take rides in self-driving cars, safety drivers are also one of the few human faces of this new technology, so they're also taught how to answer curious questions from the public, something almost all drivers have dealt with at some point. "I've seen people in Audi R8s drive by us and look at our cars, and I'm like, 'You're looking at me?'" says Dykman.
According to Uber, people with military backgrounds make particularly great safety drivers. Veterans, Abboud says, "have proven to be some of our really robust, well-adjusted, mission specialists" (the company's term for safety drivers), because they are used to following structured plans and detailed instructions.
Smaller start-ups have teams with engineering or technical knowledge, who supervise the cars as well as detecting or fixing problems. Ned Boykov is an experienced mechanical engineer, originally from Bulgaria, who has worked for Silicon Valley self-driving start-up AiMotive for just over a year as a test and integration engineer.
"I kind of feel like the car is something I'm connected to. I already know the system and know the limitations," he says. "You have to be a very responsible person to be able to do this job. Not only a good driver, because you can be a good driver, but not paying much attention to everything surrounding you for a quick part of a second, and something can go wrong."
Many of May Mobility's "fleet attendants" are local students with an interest in the technology, and it also likes to hire veterans and retired engineers. "Those are the ideal people we [have] found, people who are intelligent and motivated, and excited about what we're doing," says Dykman.
An expired job advert for Waymo, which began life as Google's self-driving car project, stipulates that candidates must be 21 years old, have no more than one point on their licence, have no drunk or drug-driving convictions, or ever been at fault for an accident that caused injury or death.
Joseph Rooney, chief executive of Denver-based firm Elevation Proving Grounds, which sources safety drivers for car and truck companies, says asking for a clean driving record is common. "They need to be safety-oriented people because it's a very critical position for autonomous vehicles. If things go wrong it takes a dent on the company [and] on the industry."
Many of the people Rooney recruits to test self-driving trucks are experienced long-distance heavy goods vehicle (HGV) drivers excited at the chance to go home every night, instead of spending weeks on the road, while applicants to work in cars are often students or recent graduates who are looking for a way into the industry, or desk-bound career-changers tired of the "daily grind".
"Some people, they do have a college education, and maybe their college education isn't relevant to AVs (autonomous vehicles), so they use the driver position as a way to break into the industry," he says.
Jesse Clifton worked in retail before becoming a vehicle operator for Uber. He has now spent two and a half years behind the wheel of self-driving cars, originally at the taxi app firm in Pittsburgh, and now at Silicon Valley start-up Voyage, where he works as the company's dispatch lead, organising the cars and the testing process.
"I don't even think I get that much of an adrenaline rush anymore," he says. "The first few times, like in Pittsburgh, if something happened your heart would really start racing, you'd pull over for a few minutes and calm down. But after experiencing it so much - you take over if you need to, and you move through those scenarios safely, and then you just continue on."
His worst fear? Other (human) drivers. "I was honestly more worried about somebody rear-ending me, versus somebody appearing in front of me, or a car running a red light. I was more afraid of the distracted people behind me."
How motorists would spend their time during a commute in an autonomous vehicle
• 28% would spend their time reading
• 27% would listen to podcasts
• 24% would spend their commute watching TV shows
• 20% would catch up on sleep
• 18% would spend their time watching films
• 17% would play video games
• 16% would spend the time learning a new language
• 10% would learn a new hobby (such as knitting or playing an instrument)
• 10% would spend their time having sex
SOURCE: WeBuyAnyCar, 2018