As a robocar team progresses towards its eventual goal of making a fully unsupervised vehicle that can drive without the full-time involvement of a human, they will go through many steps. It is often depicted that the “big milestone” is the removal of a human from the vehicle, there is now “no driver.” That’s important, but it’s not the big milestone that comes later. However, a few companies, in an attempt to show progress have created a PR milestone by moving the in-car supervisor from the traditional driver’s seat to another position in the car. This is flim-flammery and should not be done, which will be explained below.

The actual big milestone is the ceasing of full-time human supervision. This means the car is, for a substantial part of the time, eventually the vast majority of it, watched by no human supervisor. It can call home for advice if it encounters an unusual situation, but that’s usually done while stopped, and only from time to time.

This is the big milestone because you finally have a car that can scale, without requiring an hour of human labor (and actually more) for every hour of driving. All teams, when they first remove a supervisor from inside the car, still have an employee watch the vehicle operations remotely over the data networks. That employee can command various forms of “emergency stop,” typically commanding a move to what’s called a “minimum risk condition,” such as pulling off the road. It’s also possible, though less common, for them to have a remote steering wheel and pedals and be able to do remote driving, just like an employee in the driver’s seat, though not quite as effectively due to computer network limitations. (Some people think this isn’t possible, but other teams how shown that it is.)

After telling the vehicle to abort, they can usually perform the same “remote assist” functions which occur when a vehicle ask for help, such as giving advice on what to do next, or even a series of waypoints to drive onward.

Everybody does this because it would be reckless not to do so. You’re not going to send your car out for the first time with nobody in it and not watch what it does, just hoping it comes home! You will only have a small number of cars, so the cost of having a full time human watching is low, on part with the humans who used to be sitting in the cars during the “safety driver” phase. The change from safety driver to remote supervision looks big, because we now see vacant cars, so press is often made about it. Much less press is made when the remote supervision stops, but that’s the actual big event.

Waymo and Zoox have both declared they stopped full time remote supervision: Waymo several years ago, Zoox last year. Chinese robotaxi operators have also declared this. Tesla has yet to declare it.

Some might argue it’s also a big event when you finally get the number of remote assist staff down so that you only need one person for a decent number of vehicles, like 1:20. That’s important because you are finally getting the amount of human labor per hour of driving down to a minimal amount, and your service has shown workable economics. There’s actually workable economics even with 1:1 humans and remote driving, because there you don’t need human labor while the car is sitting stopped, charging or waiting for a ride request. Because that’s about half the time, sometimes more, that is still a worthwhile win, or would be if the big win of 20:1 or 50:1 were not possible.

Most testing is done with a human in the vehicle, in a job that Waymo/Google called the “safety driver.” This person is not there to drive, that only happens if something goes wrong, but they are the legal driver-in-command, watching the vehicle and ready to intervene. They are there for safety, because the systems has not yet been rated safe enough to be on its own.

Several years ago, AVRide, a robotaxi company that was formerly Yandex but was transferred out of Russia, announced a stunt where they would move the safety driver to the passenger seat. “Nobody behind the wheel,” they declared, and so it appears. However, any of the billion people who have done driving school will know that the driving instructor sits in the passenger seat. They often have their own brake pedal for emergency stop, and they also can, and do, grab the wheel with their left arm if they need to.

This system of safety driving actually works. Student drivers in this configuration have crash rates similar to adult, licensed drivers. It’s worked a billion times. It’s necessary, obviously, because the student has to sit behind the wheel.

In a robocar, though this shift can be looked at two ways, neither of which is good:

  1. Like the driving instructor, it’s still just as safe. If so, the move is not demonstrating any milestone of performance by the software, or
  2. The right-seat safety driver can’t resolve problems and so having them there is more dangerous. We’re showing that the software is ready for that, as a halfway step to taking them out altogether.

It’s either just as safe, and so means nothing and should not be done, or it’s less safe but serves no function except PR, so it definitely should not be done. Neither of those is a suitable choice. There’s no money saved moving the safety driver, the way there is with remote supervision. There’s no extra seat for passengers. The job is slightly harder, since if they have to switch to manually driving they have to crawl around or get out. There’s downside, but not benefit except being able to say “nobody behind the wheel.”

AVRide did this first, but others have done it as well. Cruise did it for their first “driverless ride.” And Tesla made a big deal about doing it when they were unable to deploy unsupervised robotaxis back in June of 2025, attempting to call the safety driver a “safety monitor” to make it sound like the role had been removed. Tesla’s recent video of the first Cybercab models with no controls also calls the person a “safety monitor” but it would be shocking if they have no means to do an emergency stop; it makes no sense for them to be there if they don’t. It’s true that can’t rescue-drive the car in manual mode after an intervention, unless they pull out a game controller or call on the remote driving system Tesla has stated they have built.

A few of the nascent robotaxi projects in Europe have also taken to trying to call their vehicles “driverless” or “nobody behind the wheel.” In some cases, this is partly forgivable because the law in their region requires a human be in the car. But only slightly--there is no reason to deliberately make things less safe just to look good.