In Tesla’s 4Q24 Earnings call, which reported sales and profits below expectations, Elon Musk claimed that Tesla would launch an “unsupervised” robotaxi service in Austin, Texas by June of this year, and in every market by next year. Musk says 2025 may be the biggest year in Tesla history.
Musk’s claim was met with extremely strong skepticism in online forums for self-driving. He has made predictions of the imminent arrival of such a capability every year for over 8 years, and not delivered. This 5-month prediction is, however, one of this most short-term predictions.
The service will be offered, he states, through Tesla fleet-owned vehicles. Customers will receive FSD software the the same capability reasonably soon after. Cars will pick up riders with nobody in the vehicle in June. He admits it is a bold goal, and they will put their “toe in the water gently at first” but because the system “does not require mapping” it can spread quickly, and will be in “many regions in the US by the end of the year.” He believes the only thing that has held them back is an “excess of caution” and that the media are focusing unfairly on Tesla crashes.
Next year, it will be “in almost every market, limited by regulatory approval” and that customers will be able to hire their own vehicles out via the Tesla fleet. Tesla said they will prove to regulators that Tesla FSD is “unequivocally safer ” “than regular driving mode, one crash for every 5.9 million miles driven, 8.5 times safer.” (He may be referring to data from NHTSA which states human drivers have a police reported crash about every 600,000 miles. When Tesla reports crash numbers for Autopilot/FSD, it reports numbers for the much less frequent airbag deployments mostly on freeways–not police reported crashes on mixed streets–and for the supervised system.)
Analysis
Musk’s claim is, to all available evidence, extreme to the point of being ridiculous, and it is indeed receiving much ridicule. While Tesla has not released signficant data on their supervised FSD system, their Q4 2024 data on Autopilot (which now uses the FSD software when on freeways) did claim a 5.9 million mile interval between “accidents.” This is for crashes which trigger an airbag, which is a small fraction of total crashes, and mostly on freeways, where crashes are much less frequent (though more severe.) This is for the supervised system, where the driver is nagged to constantly watch the road, and takes over if the system does anything to make the driver uncomfortable.
On the other hand, independent Tesla owners have been collecting data on how often they feel they are doing a “critical disengagement,” namely taking over when the feel the vehicle was about to do something serious, about every 400 miles with the new 13.2 release of Tesla FSD. Because they intervene, it is not determined whether an actual crash would have happened, this is based on fear. Constrast that with a recent study from Swiss Insurance underwriter SwissRe which examined the data on Waymo Robotaxis operating with nobody behind the wheel. They found those cars were engaged in a crash event which would have resulted in an insurance liability claim every 2.3 million miles. While the two numbers calculate something moderately different, the gulf is so immense as to make Musk’s claims entirely incredible. While Waymo is now beginning to scale their service, including in Austin, this is over 6 years after they first were able to put a car out on the roads with no safety driver.
That’s a crucially important fact. As hard as it is to reach that level of safety, even arguably the best team in the business took 6 years to get from that to serious deployment. In spite of Musk’s claim of the importance of their reduced use of maps (they do use them in some places) the Waymo cars are not blocked in their expansion by mapping issues. Their cars drive off-map every day when the road changes; they just don’t wish to do that regularly as they feel it reduces safety.
Tesla recently showed off their Cybercabs giving rides on a closed movie set, and their cars driving themselves off the factory line to the loading area. This is a wise effort, but it is entirely unimpressive in 2025. Teams have been doing private-course driving since the 2000s.
As the new Director of Government Efficiency, Musk may be able to remove regulatory hurdles from Tesla’s deployments. Texas already has an extremely light regulatory regime, which is why they are starting there. Tesla has not yet even applied for a test permit in California for an uncrewed vehicle, and the process there is very unlikely to be managed all the way to permits to provide service to the public in just a year. That’s true in a number of other states, which don’t even have laws or experience with it yet.
