Self-driving car startups have made a rosy promise of developing truly self-driving cars and have raised tens of billions of dollars as a result. But industry executives and experts say automakers may always need remote human supervisors to help troubled robot drivers.
Remote operators for self-driving cars
For self-driving car development, its core premise is that computers and artificial intelligence can dramatically reduce accidents caused by human error, a vision that has driven significant research and investment in the industry.
Fatal shortcomings
But here’s the rub: It’s very difficult to build automated cars that are safer than human drivers, because self-driving software systems simply don’t have the human ability to quickly predict and assess risk, especially in the event of an accident or “edge case” (extreme situation).
When asked if the elimination of remote human supervisors could one day be considered, Kyle Vogt, CEO of GM’s self-driving subsidiary Cruise, said, “Well, I would ask, ‘Why (do that)? . I can give my customers peace of mind and let them know that there’s always someone there to help if they need it. I don’t know why I’d want to get rid of him.”
GM’s Cruise self-driving test vehicle
This is the first time Cruise has acknowledged that there is a long-term need for remote human operators in self-driving cars. In June, Cruise’s self-driving car crashed in San Francisco, injuring two people. This month, General Motors recalled and updated the software in 80 Cruise self-driving cars. Cruise says this unusual situation will not happen again after the update.
For some, the idea of keeping human supervisors raises more questions about self-driving technology. True self-driving cars are far behind the industry’s optimistically predicted rollout plans from a few years ago.
In 2018, GM sought U.S. government approval for a fully autonomous car without a steering wheel, brake or gas pedal and promised to include it in its commercial shared mobility fleet in 2019. But now Vogt says production of the car, called the Cruise Origin, won’t begin until spring 2023.
Elon Musk draws the big picture
In 2019, Tesla CEO Elon Musk promised a million self-driving cabs “for sure next year,” despite criticism of Tesla’s “Fully Self-Driving” (FSD) system. Tesla’s FSD has been criticized for its inability to drive itself without safe driver supervision, requiring humans to take manual control in an emergency.
In June, Elon Musk said in a YouTube interview that developing self-driving cars was “much more difficult than I initially thought. But when asked about the timeline for the launch, he said Tesla could make it happen “this year.
However, Tesla’s FSD is still being tested. Elon Musk also tweeted today to remind users of the FSD beta to be safe. “FSD Beta 10.69.2 is now available. Please note that safety is a top priority, so please be sure to exercise caution, especially around pedestrians.” He said.
Elon Musk warns FSD testers to be safe
For self-driving startups, the promise of truly autonomous driving has yet to materialize, increasing their risk. “If these companies don’t succeed in the next two years, they will cease to exist,” said Mike Wagner, CEO of Edge Case Research, a firm that helps self-driving car companies assess, manage and underwrite risk. “It’s the action or shut up on this issue.”
Extreme situations still require humans
Today, many self-driving startups use humans as remote supervisors and have human safety drivers behind the wheel at the same time.
These remote supervisors come with additional expense but can help self-driving cars deal with edge cases, which can include basic things like unfamiliar lane closures during road construction or erratic, unpredictable behavior by pedestrians or human drivers.
Koosha Kaveh, CEO of Imperium Drive, notes that when a robot driver encounters an edge case, it gets overwhelmed and says, “I don’t know what’s going on. The remote operators. Over time, these people will play the role of “air traffic controllers,” overseeing an increasing number of self-driving cars.
Cruise CEO Vogt said its self-driving cars currently rely on humans less than 1 percent of the time when driving on San Francisco roads. But for hundreds, thousands or even millions of self-driving cars, that time adds up to a lot of time spent on the road waiting for human supervisors to guide them.
Truly autonomous driving is a long way off
Imperium Drive’s Carvey says the number of edge cases will drop as more and more self-driving cars that are more predictable than humans hit the road, “but you’ll never get down to zero edge cases.”
“Even decades from now, you’re not going to get to 100 percent truly self-driving cars,” Carvey added.
Competition, however, is intensifying. Some Chinese cities are pushing faster to allow self-driving cars to launch aggressive testing. As investor investment in self-driving cars has plummeted, startups are under pressure to solve edge cases faster and cut costs on everything from sensors to headcount to get to market.
Skepticism is emerging as investors are confused about how long it will take for self-driving companies to become profitable. Simpler or slower self-driving cars may turn a profit first, such as trucks operating on low-speed routes on roads or film sets or last-mile delivery vehicles, but it will also take years to do so.
Self-driving investments have already plummeted this year
Overall investor investment in the future of mobile mobility startups has slowed, with self-driving-focused companies hit particularly hard, accounting for less than 10 percent of venture capital in the second quarter of this year, according to investment site PitchBook.
According to PitchBook, investor investment in self-driving startups fell to $958 million in the second quarter. And just two years ago, self-driving car investment was still in its infancy. At the time, Google parent Alphabet’s Waymo raised $3 billion, DDT’s self-driving car unit raised $500 million and Amazon.com Inc. acquired self-driving car startup Zoox for $1.3 billion.
The risk of rushing to the bottom
Chris Borroni-Bird is an independent consultant who has led advanced vehicle programs at General Motors and Waymo. He says autonomous driving systems are not as capable as humans because their “perception and prediction algorithms are not as good as the human brain’s processing and decision mechanisms.
Byrd said, for example, that when a human sees a ball rolling down the road, he or she thinks a child might be following it and brakes faster than a self-driving car.
“I worry that self-driving car companies will rush them to market without proving that they are safer than human drivers,” Byrd said.
The problem is that self-driving cars could encounter “tens of billions of potential edge cases,” said James Routh, CEO of AB Dynamics. The company tests and simulates road conditions in cars, covering advanced driver assistance systems (ADAS), which are the basis for self-driving capabilities.
Sarah Larner, executive vice president of strategy and innovation at automotive data startup Wejo Group, said they receive 18 billion data points a day from millions of connected cars and help self-driving cars simulate road conditions.
“But there are just so many variables, like weather, that you can come up with an edge case and then have to consider all the different variables,” she said, “and the output is literally in the millions.”
