Judging Tesla Robotaxi On What It Does, Rather Than What Musk Says

Forbes

3 days ago

Tesla has managed to generate huge buzz around thier robotaxi efforts, even though it does not yet have a working robotaxi. It has announced many highly ambitious plans, but today, hopes don't matter, and the main thing we can look at is what companies do, rather than say.
Recently Tesla has been aggressive with actions, including offering supervised robotaxi services (to friendly influencers) in two cities, and expanding their Austin service area, and a car delivery demonstration. Their Bay Area deployment may flout permit rules. It's been far grander in promises, including deployment to half the U.S. population, and unsupervised FSD for customers in 2025. It is hiring safety drivers for more cities.
Actions speak more loudly than words because actions reveal what the company's own internal data are saying. The most dramatic action is the release of a general vehicle for use with no human supervising it, because to do so indicates the company believes that's an acceptably low risk. If such a vehicle causes a serious incident on the roads, it can kill a project, or even a company, and this happened to both Uber ATG and Cruise. Deploying vehicles in humans supervising (in the vehicle or remotely) is one thousand times less risky--the number is indeed that large. Several companies have safely deployed vehicles with human supervision that couldn't go 10 miles without getting into a crash without the safety driver, and had to get 100,000 times better before they could remove that human.
An interesting question is why Tesla is doing these large-territory supervised robotaxi deployments. The deployments aren't themselves large, as they are only open to employees and selected Tesla-friendly social media influencers. Tesla reported only 7,000 miles of operation in Austin (though has not clarified if that was in one month or one week.) That's very little. Other companies have done supervised robotaxi test deployments, but usually over small territories.
For all companies, the first priority of testing is to improve the quality and safety of the driving. Tesla is well positioned for that, because they have hundreds of thousands of customers driving with their FSD package, though Tesla is only able to make use of a small fraction of that huge trove of data, and now only those with newer cars can run the latest software that they want to test. Tesla has always seemed to do much less testing with employees driving the car, thanks to those customers who pay for the privilege of doing the work.
Taking passengers tests a different segment of a robotaxi system, namely the taxi part--letting riders summon cars, doing pick-up and drop-off, and providing an in-car experience and rider support. It also tests the app and fleet management aspects. People have been critical of Tesla when it said they would soon deploy a working robotaxi before it had ever done any testing of these aspects, but now they have started. No business plan surivives first contact with the customer, and so this testing is essential.
There's no business reason to have a supervised robotaxi service with employees--it's more expensive than running a regular limo service. There is no reason to scale it or make it a long term plan. Because passengers are charged fees, however, it does provide some subsidy to the testing of robotaxi features. Taking passengers does come with risks, because they will report on, or even publish videos of mistakes, and there is a hopefully small risk of hurting them in an injury crash.
Indeed, in general, a company would not normally run a supervised robotaxi service at any scale if they didn't feel they weren't very far from launching a real robotaxi service. That would normaly be a bullish sign for Tesla's internal impressions, if not for one thing.
Tesla's Poor History
That one thing is that Elon Musk has repeatedly declared that he was 'certain' or 'extremely confident' that Tesla was on the verge of releasing an unsupervised self-drive system. Again and again, and over the course of over eight years he has done this, and he's still doing it. He simply has no ability to judge the readiness of the project, or is not truthful with what he knows. If, some day, he turns out to be right, it will be in the way that a stopped clock is right.
In addition, Tesla has an even darker history of not just failing to predict their own timelines, but of hiding information or even being misleading. None of the companies are entirely innocent here, they all keep things closer to the vest than they should, but Tesla's reputation isn't good. Except in one area--they let their customers use their FSD system directly, and see and publish its virtues and warts.
Best would be for Tesla, and other companies, to release their statistics on actual safety performance over millions of miles. They have that data, but won't reveal it. Waymo has released some, and Cruise also released smaller amounts, though all release only what they feel they want to. All companies (except Tesla) release what the government rules require, but the rules are vague enough that these reports are often massaged in a way that makes them less useful. Tesla simply astonishingly declares they are not testing self-driving vehicles in California and so they don't have to report.
This is particularly the case when operating or testing with a safety driver/supervisor. A supervisor prevents mistakes by the system from causing a problem, and so from the outside, you can't tell apart systems that have incidents every trip, ever 10 trips, every 100 trips or even every 10,000 trips (which is a level you must exceed to consider removing the supervision.) Statistics to reveal where on that very long journey a system is are what the outside world needs.
This leaves the public in a quandry. Many fear Tesla makes statements and does demonstrations primarily for publicity. Since Tesla's CEO can't reliably evaluate the state of his own projects, it does us no good to say that their actions reveal their internal thinking. Their internal thinking has never been right!
Most recently, Musk stated several times in earnings calls that they were highly confident they would release an unsupervised robotaxi with nobody in the vehicle back in June. He was wrong, and they have yet to release one. He correctly predicted an unmanned vehicle delivery, but such cherry-picked single drives are a much lower bar that many companies have done going back a decade.
Overall, it seems that these projects do have some value, but a number of them seem to mostly be about image rather than R&D or commercial value. The ultimate action that speaks the loudest will be an unsupervised deployment, should it become able to do that. In that case we'll want to look at the complexity of the service area, and whether it's open to the public.
Another milestone we may see is the removal of the safety driver but doing full time remote monitoring and driving. Already some companies are doing remote driving, so if that is present it is not unsupervised, though it does demonstrate capabilities similar to those of the remote driving companies.