Tesla’s quarterly earnings call began with good news, moderately beating analyst estimates for earnings and revenue. There was news about many other fronts, but this article will focus on news regarding Tesla FSD and Tesla’s robotaxi offering.
It’s been clear that Tesla is rather poor at predictions about how these products will do, and if or when they will get deployed. It is necessary with all companies to look at what they actually do rather than what they say, but with Tesla there is another complication–sometimes the company will do things not because they are ready, but because the CEO insists it is time.
Indeed, everything here: FSD Unsupervised, robotaxis and upgrades, is all predicated on Tesla making a safe and working unsupervised self-driving software system with their current HW4 hardware, which is yet to be demonstrated and many think may not happen, or at least not any time soon.
Upgrades for owners of older “HW3” cars
Tesla cars purchased prior to late 2023 have Tesla’s older processor and camera hardware. Later cars have HW4 hardware. Their HW5 chip (more commonly called AI5) has now been taped out and will probably go into production in 2027, but it will be primarily for the Optimus robots and other products. Musk stated that the HW4 chip is sufficient for the cars, but the HW3 chip is not.
However, many customers purchased cars with HW3 and bought the lifetime FSD package when Tesla promised that the car had all the hardware needed to do real (not supervised) self driving. I am such a customer. For those customers, Tesla will offer a path to upgrade. That path will be:
- A discount on a trade in of the HW3 car for a newer HW4 car, or
- An upgrade at a new special service center, dubbed a “microfactory,” replacing the computers, cameras and other hardware needed. These new service centers would need to be built so as to not overload existing service centers.
It was not indicated what the discount would be, or if there would be a cost for the upgrade. Many owners may assert that they were promised their car had all hardware needed, and they should not have to pay anything for this, at least for option #2. For Option #1 it could be that owners would need to pay the difference in book value minus a discount, since they are getting a newer car out of the deal. (This would be fairly profitable for Tesla since on a typical “trade-in” they purchase the old car at a highly discounted trade-in price and sell the replacement car at a retail price already.) However, most owners would like a newer car.
Truth be known, the trade-in method, if not that expensive for Tesla, is much cheaper than having to develop and install all these service center upgrades, since the trade-in requires no R&D or other work. It’s really just some deals.
Purchasers of HW3 cars with have waited a long time, and the product is still not available. My own vehicle has aged 7 years and depreciated by more than half its value, unable to perform the function. There have been threats about class action lawsuits over the promises that were not delivered on, and which can never be delivered on if the cars age-out unless Tesla revises its policy on transfer of the software license to a new vehicle.
Unsupervised FSD
Tesla once again promises that regular customers will get some form of unsupervised FSD by the end of this year in “a dozen or so states.” This will be based on their current 14.3 release which is “essentially” the same software running in the robotaxis. Version 15, which they claim will be much more sophisticated, but not needed for the core functionality, is not expected until next year.
However, for the first time, they have revealed that this customer product will have limitations on its service area, including not being able to go through regions that are deemed too complex or dangerous. This means not everybody will get the product, and it will avoid “unsafe intersections or bad road markings or a lot of weather challenges,” according to Musk. That probably means no winter driving.
Musk also said that unsupervised FSD and Robotaxi revenue will “not be super material this year,” quite at odds with past predictions that it was the future of the company and would reach half the population this year. “We would release unsupervised gradually to the customer fleet as we feel like a particular geography is confirmed to be safe.” This is quite a reversal from Tesla’s previous representations that their approach did not require geofences or maps, though it is as expected by most in the industry.
Robotaxi
Tesla recently started “unsupervised” robotaxi operations in Dallas and Houston. Like all early operations with no employee in a vehicle, these vehicles are probably not actually unsupervised, but have a remote supervisor watching full time. Waymo, Zoox, Cruise and all other companies have started in this fashion. Waymo and Zoox have since declared they no longer have full time remote monitoring, Tesla has not.
This deployment is extremely small. Independent observers have noticed only 2 different vehicles in each city, based on licence plate tracking, and they are only in operation some of the day. The number of vehicles in Austin has increased from 9 to 13 recently, though again, not all are operating at all times. During the call Tesla executives stated they have had “not a single [injury] to date” with the expansion. It is unclear what Musk meant here, because Tesla has reported injury crashes to NHTSA’s federal database of AV crashes previously, but has reported none recently. They may mean to say they have had no injury crashes with the “unsupervised” vehicles, or none in the 2 weeks since the 2 cars per city expansion to Dallas and Houston.
Tesla’s crashes have been happening at a disturbing rate, however. Tesla has reported 18 crashes in about 1.8 million miles, seemingly all in vehicles which have a Tesla employee in the driving-instructor seat. This employee, commonly called a safety driver (Tesla prefers “monitor”) has an emergency stop button and can grab the wheel. Tesla advertises that ordinary Tesla owners using their supervised FSD product have one minor crash every 1.6 million miles. Somehow their professional safety drivers have had roughly 16 crashes in the same distance compared to the amateur ones in regular Teslas. That’s unusual–all teams do extensive testing with supervising drivers and they usually have very good crash records.
Nonetheless, Musk claims that the robotaxi system is already extremely safe, and they mainly are working on problems around road citizenship. They are slowing release because they don’t want to be “stuck blocking intersections, or don’t want to be dropping people off at slightly incorrect locations.” Tesla is “track[ing] basically all the metrics” but declined to answer what their numbers are. Musk enjoyed reporting how a line of a dozen of their cars caused a traffic jam, stuck because “a Waymo had crashed into a bus.” They don’t want to fully release until they can stop that. (Curiously, this would mean it was a dozen of the supervised robotaxis, and the safety “monitors” could have switched sides and driven the vehicles away, or asked their remote drivers to do this. (Tesla recently disclosed they can remotely drive cars up to 10mph, though some riders have reported seeing this happen at much faster speeds.)
This realization may be new to Tesla, but it is not unexpected. Even companies like Waymo, who got their vehicles safe enough to deploy back in 2019, still get somewhat regular reports of vehicles making both safety and non-safety related mistakes on the road 7 years later. There’s a very long tail to solve of both unusual safety situations and unusual traffic situations and it doesn’t seem to solve with other than a lot of time and practice.
Tesla also reported that while CyberCab production has begun, it will be “very slow” to start. Musk denied rumors they are making a new family vehicle, and said the 2-seat Cybercab will be their focus. The Roadster, which he suggested would get shown in a month, will eventually be their only manually driven car.
