Autonomous Vehicles startups have raised tens of billions of dollars on promises to develop truly self-driving cars, but executives and experts say the need for human supervisors of self-driving cars may be ever-present to help troubled driving robots. Reuters (Reuters).

The basic premise of self-driving vehicles, that computers and artificial intelligence will drastically reduce accidents caused by human error, has prompted many companies to research and invest.

But the problem is that robots being able to drive cars more safely than people is very difficult because self-driving software systems simply lack the ability for humans to predict and quickly assess risks, especially when faced with unexpected accidents or “extreme situations.”

"Well, my question would be: Why?" says Kyle Vogt, CEO of Cruise, a unit of General Motors (GM), when asked by Reuters if he could see at what point the human supervisors should be removed from operations.

"I can give my clients peace of mind because I know there is always someone to help if needed," he explains, "I don't know why I want to get rid of that."

It is the first time that he has recognized the need for the human element to remain in the process of operating self-driving cars.

Like air traffic controllers, these human moderators can sit tens of hundreds of kilometers away and watch video footage from several vehicles, sometimes with a steering wheel, ready to step in and move the stuck robot drivers back in. Automated vehicles always stop when they can't figure out what to do.

Alphabet's Waymo and Argo, backed by Ford and Volkswagen, declined to comment when asked the same question from Reuters.

General Motors recalled 80 self-driving Cruzes this month to update its software after a June accident in San Francisco that left two people injured.

US safety regulators said the software, which was recalled due to an automobile technical glitch, could "incorrectly predict" the path of an oncoming vehicle, and Cruz said the unusual scenario would not recur after the update.

For some, the idea that human supervisors might be necessary in this type of vehicle raises even more doubts about the technology.

General Motors recalled 80 self-driving Cruze cars this month to update its software after an accident (Reuters)

Promises and aspirations gone awry

And it seems that truly self-driving vehicles are far behind the optimistic roll-out schedules that were predicted just a few years ago.

In 2018, General Motors sought US government approval for a fully autonomous vehicle, without the steering wheel, brake pedals or accelerator pedals, that it predicted would enter its combined commercial fleet in 2019.

That car is the Cruise Origin, which isn't currently scheduled to start production until the spring of 2023, according to the company's CEO.

In 2019, Tesla CEO Elon Musk promised 1 million taxis "next year for sure" — though his company's offering at the time for "Full Self Driving" was criticized for not being able to drive themselves without Having a human behind the wheel of the car is ready to take manual control in an emergency.

In a June interview on YouTube, Musk said developing self-driving cars is "a lot more difficult than I thought, by far".

But when asked about a timeline, he said Tesla could achieve it "this year".

Tesla did not respond to a Reuters request for comment for this story.

The unfulfilled promise of true autonomy has raised the stakes for the self-driving car industry.

“If these companies don't succeed over the next two years, they won't exist anymore,” said Mike Wagner, CEO of Edge Case Research, which helps self-driving car companies assess, manage and insure risks. ".

Watching people from afar

Many emerging self-driving car companies today are using humans as remote supervisors, along with safety drivers sitting behind the wheel.

These remote humans are an additional cost to these companies, but they help self-driving cars deal with difficult cases.

These situations can be unusual actions such as closing lanes during road construction, or irregular and unexpected behavior by pedestrians or drivers.

"Over time, these people will act as watchdogs overseeing an increasing number of self-driving cars," said Kosha Caffee, CEO of Imperium Drive, which uses humans as remote car operators in Milton Keynes.

Cruise's Vogt says its vehicles on San Francisco's roads currently rely on humans less than 1% of the time.

But with hundreds, thousands, or even millions of additional vehicles, that would add quite a bit of the time that cars park on the road waiting for human guidance.

With more self-driving cars — which are more predictable than humans — the number of emergencies will go down, says Caffee, “but you never get to zero human intervention.”

"Decades from now will not get you 100% of truly autonomous vehicles," he added.

However, competition is rising, and some Chinese cities are pressing to allow active self-driving testing more quickly.

The need to address critical situations and reduce costs for everything from sensors to the number of people in the chain of command to reach the market has intensified, due to lower investor funding for self-driving cars.

rush to market

Autonomous systems are not as capable as people because “their perception and prediction algorithms are not as good as how the human brain processes and makes its decisions,” said Chris Buroni Baird, an independent consultant who previously led advanced vehicle programs at GM and Waymo.

"For example, when a person sees a ball rolling in the road that is harmless in and of itself, he assumes that a child can follow and hit the brakes much faster than a self-driving car."

"I am concerned that autonomous vehicle companies will rush into the market without demonstrating that their safety standards are better than human-driven vehicles," he added.

James Roth, CEO of AB Dynamics, said the problem lies in the "tens of billions of possible situations" that autonomous vehicles could encounter, which he is running simulation tests on, including advanced driver assistance systems that are the foundation of driving features. subjectivity.

Sarah Larner, executive vice president of strategy and innovation, says robotic data operator Wejo receives 18 billion data points per day from millions of connected cars and helps with vehicle simulations.

"But there are many variables like weather, you can take the worst case scenario and then have to make scenarios for all the different variables. It's really millions of outputs."

Driverless delivery services

In track tests of cars, AB Dynamics uses a robotic arm that it plans to modify on slow-moving mining and farming trucks to make them largely autonomous.

Roth envisions a remote team of humans overseeing fleets of such cars, for example, self-driving mining trucks that operate in closed environments.

But he doesn't see that this scenario could work for vehicles in faster and more open environments, because it could be difficult for remote human supervisors to respond quickly enough to dangers.

Within the next month, British online food delivery company Ocado will roll out a small fleet of driverless delivery vehicles with Oxbotica to launch its autonomous vehicle program - supported by remote human supervisors - that will operate only one distance.

Meaning there are few streets on specific roads in a small UK city and you never drive more than 30 miles (48 kilometers) per hour.

"At 48 km/h, if the vehicle's systems are jammed, it can apply the emergency brakes and call for help," said Alex Harvey, head of advanced technology at Ocado. "This seems like a very viable strategy at low speed."

"But you can't play this game on a highway" because hard stops in acute situations can pose a safety hazard, he added.

He said it should take about 5 years for Ocado to develop an efficient and profitable driverless delivery system.

More than half of UK customers can be reached with self-driving vehicles traveling at speeds of no more than 48 kilometers per hour.