Cern Basher and Ark Invest provides some analysis and tracking of the rapidly improving quality of Tesla FSD and robotaxi.
Tesla is will soon pass the 6 billion miles of driving data which Elon said in 2016 would be needed for robotaxi that is both safe enough and provably safe for regulators.
Tesla is on track for the June 2025 Austin unsupervised (no human driver) paid rides.
This will be expanded to other Texas and California cities through the remainder of 2025.
By October, 2025, Ark Invest expects Tesla FSD to surpass the safety of human drivers. The unsupervised human driving in June 2025 will be safe but Tesla will use more cautious driving in the locations with the best data to get unsupervised 6 months before globally better robodriving.
This also appears on track to achieve driving in 2026 that will be ten times safer than human driving. This will be what is broadly rolled out to millions of Tesla cars around the world.
The billions of miles driving data and the growing AI clusters are ensuring the rapid complete solving of global fleet of millions f robotaxi over 10 times after than human drivers in 2026.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
I have not seen any videos of driving on snow covered roads. We get them in Canada.
I posted an article with FSD youtube driving in canada for winter snowy roads. Tesla FSD 13.2.X or better seem to be doing good.
Apparently stop signs are not recognized by FSD if covered with stickers. With Musk creating so much ill-will with his Nazi salutes and layoffs from the government one wonders if people might try to sabotage the roll out of FSD.
A good real world “acid test” would be to deliver the Tesla from Las Vegas to the owners door via Robo Taxi only, charging included.
It knows where stop signs are based on past data.
Interesting point. Most people know what a stop sign “is”, even if covered in graffiti. Why and how? Most people (usually) understand the “shape, color” of a stop sign. Just for the record did people know the English word “STOP” is used on “stop signs” in places where English is not even spoken? I’m not making this up, look it up… Curious. “Stop signs” are so universal. Musk is going out of his way, NOT to be. So sad. So much potential, and such arrogance…
Accumulated miles before Tesla went to cameras only can probably be thrown out, as can miles before AI came into the equation. Obviously, any comparison with human-assisted driving can be discarded when the robo-taxi won’t even have a driver. The robo-taxi is supposedly a new model of car, never seen before outside of a closed circuit very limited demo a few months ago.
It may be that NONE of the previous track record means anything on the new driverless platform.
It would both simplify and validate the safety claims of Tesla regarding its robo-taxis if an independent qualified body evaluated the safety of Tesla’s robo-taxis. And since Musk and to a lesser extent, Trump, have compromised every agency they’ve taken charge of, including the National Highway Traffic Safety Administration (now under the unqualified federal Transportation secretary Sean Duffy, whose dept. is being cut arbitrarily by Musk/Trump too), there’s only private industry to turn to. The best place to look for independent analysis then, because they have skin-in-the-game, is the insurance industry (not Tesla’s in-house insurer, which purportedly provides bad, expensive insurance even for “normal” Teslas: https://www.reddit.com/r/TeslaLounge/comments/1cmifkf/finally_had_enough_tesla_insurance/ and https://quotepurple.com/tesla-insurance-how-does-it-compare-to-normal-insurance/).
When independent insurers cover Tesla robo-taxis at competitive rates to “drivered” cars, and do it through a year or so of intensive, real world (including urban) driver then maybe I’ll feel safe getting into a robo-taxi, at least in areas where it’s been tested to that degree (NYC, where I live, will probably take a decade to clear this hurdle, if ever).
Ignoring Tesla claims; open source tracking of FSD disengagements suggests that there is no significant improvement happening https://teslafsdtracker.com/ with (accident averting) critical disengagements every few hundred miles, rather than the hundreds of thousands of miles needed for commercialization
There’s been no significant improvement happening in last 6 months – in fact they’ve gone backwards in last three months from current leading version released at xmas ’24, and clearly they have no idea how to set up a development process that will create the massive monotonic improvement needed to eliminate the necessary 999 out of 1000 current deisngagements.
At this point it is clear that a huge technology change is needed to have any hope of success – likely much more powerful computers at minimum.
You cant use that as a useful metric, you are in move 37 territory, that’s where in the famous game between the A.I Alpha zero and Le Sedole every expert watch watching was explaining why it was a mistake, everyone would have if allowed HELPED Alpha zero by taking back the move and they would all have been wrong, so how many interventions are necessary, you don`t know do you… Tesla believes it is on track well aware of this monitoring problem, its ultimately not data they can fake as if cybercabs start getting in lots of crashes they will be in very big problems, won`t they .. and they know it
10X safer then a human? Hope so. Unless of course when something “bizarre and unanticipated” happens. Isn’t that how accidents “happen”? Algorithm’s can predict cause and effect, very quickly, actually But not (IMO) those “intangible” factors. Not just an event you never experienced before. But something you never imagined before. People are really good at to use a football expression, “punting”. My computer has a 1or0 out-of-place, it’s NEVER pretty…
As a former police officer .. no .. most 999.9999 and on %) accidents are not the unimagined like a moose jumping over a hedge in front of you, they are everyday poor human judgement, watch videos on crashes, its nearly always clear who is driving like a idiot, too fast or not braking etc
Teslas under FSD crash for all kinds of reasons that would be easily avoided by humans. There are too many to summarize here, but a quick search on Youtube bring up endless shorts showing actual footage:
https://www.youtube.com/results?search_query=tesla+crashes+under+fsd
Musk and Tesla keep resetting the bar with new versions promising to be safer and better than ever! But this is a double-edged sword, since it also means discounting all previous trials, not just lumping them into a meaningless billion miles plus dataset under radically different car, software, driving, and testing conditions. It’s like saying all airplane safety should be evaluated by accident rates going back to the Wright Brothers’ Kitty Hawk test flight. Just about all of airplane flight results can be thrown out when evaluating today’s commercial jetliners. It’s the same for >90% of Tesla FSD miles driven.
Here’s one video from the Wall Street Journal just 2 months ago:
https://youtu.be/mPUGh0qAqWA?si=Q8DOkOcqZDCv2Ne1
To reiterate my earlier post, third party evaluators with real skin the game (i.e. money, reputation, legal responsibility) are needed to fairly evaluate Tesla’s FSD at this point. It is safe enough to move “out of the Tesla lab” and into the real world for others to evaluate. This means independent, not in-house, insurance companies who will actually insure a truly Full Service Driving Tesla (also, presumably the not-yet-released robotaxi, designed from the ground up not to have a driver. Again, this will obsolete all previous testing on this new platform; it needs to be evaluated on its own merits).