Forbes – Tesla Robotaxis Average Just 20 Miles A Day — Meanwhile, Where’s Mobileye?
Editors note: under 20 miles/vehicle/day? Ouch!
See original article by Brad Templeton at Forbes
Tesla’s second quarter earnings call included some tidbits about its robotaxi project and other plans, including that its cars aren’t driving very much at all, and that the level of deception on its quarterly safety statistics has reached disturbing levels.
In other news, today Lyft announced a new partnership for 2026 with Holon autonomous shuttles powered by Mobileye drive. Such announcements don’t deserve much of a mention, but it’s interesting to contrast the progress of the players.
In the call, Tesla stated that its robotaxis have now driven about 7,000 miles in Austin. (The audio is not perfectly clear, most transcriptions read it as 7,000 but it could possibly be 5,000 or 10,000. Tesla does not respond to press requests.) Since there are reportedly a dozen vehicles operating, and it’s been 30 days since the service began operations, that works out to under 20 miles per day per vehicle, which is a surprisingly small number. While Ashok Elluswamy stated that there had been “no notable safety-critical interventions,” that’s at odds with one influencer who reported his Tesla robotaxi hit a Toyota with its wheel in a parking lot, and a number of videos show the cars stopping to release passengers in dangerous locations. Other events have included unsafe and illegal moves which did not result in any impact events, and thus might not count, but they don’t inspire confidence.
Tesla Robotaxi Safety
There are many classes of intervention, and indeed many companies feel that if nothing got hurt, there has not been a problem. In the prior Q1 call, Elon Musk and others stated the cars were up to roughly 10,000 miles per intervention, without classifying them, but with all the incidents seen in the first 7,000 miles that’s not very likely. Indeed, we’re not seeing video of every mile, and the influencers allowed to ride often don’t even focus on the incidents. The challenge is that meet Musk’s goal of “much better than a human being” that he has set for going unsupervised, you need to do a million miles between serious safety incidents of any kind.
The very low usage numbers suggest Tesla is being very, very cautious with deployment of the service, which is the right safety approach with an immature system.
Tesla expanded its service area to look like a giant upside-down Tesla logo (or perhaps something else if right-side up) and they said they will soon expand it again. The large expansion makes the low usage even more surprising. Expansion, though, is not particularly challenging for a supervised system like Tesla’s, in fact Tesla could easily open up most of the country in that mode if it wants to pay for the staff. Because a supervised system can need interventions it can be operated even if it needs 100x improvement to go unsupervised, as Tesla’s appears to need by Musk’s criteria. So while Musk boasted at the start of the call that Tesla would expand its service area to be much larger than its competitors, that’s not a meaningful comparison. In the call they were optimistic about expanding to serve “half the population of the USA by the end of this year”–that’s actually not a very bold claim for a supervised service, though some states may insist its safety driver move to the driver’s seat rather than the passenger seat if the state’s laws have technical requirements for where the driver sits. (In fact, even today there are Tesla owners driving for Uber who use Tesla FSD and Autopilot while giving rides to passengers, and it appears to be legal, though this is not an official offering.)
On the plus side, reports suggest Tesla has made significant improvements to the “pick-up” function, an important component of a robotaxi. Early videos had riders walk fairly far to be picked up in parking lots, suggesting the vehicles weren’t ready to do curbside, which can involve waiting. (Drop-off is easier as there is minimal waiting.) If confirmed this shows impressive generality in the system.
Misleading Chart
In the call they also made note of its quarterly report on Autopilot safety, which included a post of a graphic to X/Twitter:
This chart is incorrect. In the fine print in Tesla’s report, it discloses that for the blue line, Tesla is defining “Accidents” as “Airbag deployments” and on Autopilot, which is used 90% on freeways, where the crash rate per mile is around 1/2 to 1/3rd of that for all roads. The report for Autopilot+Human “accidents” also is measuring Tesla drivers, who are known to be wealthier and older than the average member of the public, and thus statistically better drivers.
According to State Farm data, regular drivers have an airbag deployment about every 4 million miles, though the rate varies by car OEM, as different vendors have different rules for when an airbag deploys. What this means is that combining all these factors, there is much uncertainty, but Autopilot+driver is very likely no better than, or even slightly worse than driver alone. Tesla depicts it as 10x better than the orange line, which is police reported crashes (airbag or not) for all drivers on all roads in all cars, and this is clearly false. Though the number is getting modestly better, at least. Tesla has the real data but declines to release it, it would be nice to know the truth. I have written more details on this in the past. What’s new today is that in the past Tesla usually revealed the different base of the numbers in the fine print and did not overtly make the apparent false claim of being 10x safer.
Lyft And Holon (With Mobileye) Partner For Shuttles
Tesla continued with many claims of big milestones it is confident about for 2025 and 2026. At the same time Musk admits they sometimes get the dates wrong. Recently we’ve been seeing a number of claims from Mobileye that it will deploy self-driving through its partners in 2026. This includes earlier claims of a robotaxi service with MOIA/VW, and a declaration today from Lyft that it will offer self-driving rides in shuttles from Benteler’s “Holon” unit. Holon is also operating shuttles in Jacksonville, FL.
Holon shuttles currently offer fixed-route shared ride service, but users will soon be able to book one of these shuttles in the Lyft app. Lyft has been eager to partner with self-driving companies to match the fast pace of partnerships that Uber has been announcing.
What’s less clear is whether Mobileye Drive, its unsupervised self-driving product, will be ready. Mobileye has an advanced team, and is the leader in ADAS for traditional OEMs, but we are now at a different phase in self-driving, the Yoda phase where “there is no try, only do.” Reports of MOIA’s trial service in Germany indicate safety drivers are still quite active. If Mobileye and its partners want to convince the world they can drive around unsupervised in 2026, they will need to start showing hard milestones very soon.
See original article by Brad Templeton at Forbes
