Tesla’s Full Self-Driving (FSD) is stagnating with no real improvement in miles between disengagement in months just as CEO Elon Musk said it is going exponential.
The stagnation of FSD could be explained by Tesla making a pivot and focusing on its geo-fenced ride-hailing service instead of its long-standing promises.
Since 2016, Tesla has claimed that all it vehicles produced onward have all the hardware capable of self-driving at a level enabling a robotaxi service and that a software update would eventually enable it.
For the past six years, CEO Elon Musk has claimed that Tesla would achieve that goal by the end of the year, and he has been wrong every time.
Lately, Musk has been focused on hyping Tesla’s latest Supervised Full Self-Driving (FSD) updates.
FSD is technically a level 2 advanced driver assist system (ADAS). It requires driver supervision at all times, and Tesla takes no responsibility in the case of a crash.
Miles between critical disengagement is the primary metric used to track progress with each FSD update. Tesla and Musk have both used the metric in the past.
In January, we reported on Musk sharing a crowdsourced FSD dataset, claiming that it showed Tesla had now reached “exponential improvement” with FSD v13. We noted that this was both false since exponential improvement would require an extra data point that he didn’t have and misleading since he focused only on highway miles between disengagement and Tesla had just introduced its long-used city driving neural network stack to highway driving.
FSD v13 has now been out for 3 months, and it received several point updates, but the same data praised by Musk a few months ago shows that it is stagnating – not going exponential:
Musk had previously claimed that v13 would enable “a 5 to 6x increase in miles between disengagement compared to v12.5.”
The data now shows that v13 barely brought a 2x improvement, going from ~200 miles to ~400 miles.
After over 33,000 miles reported through all versions of FSD v13, the datasets now point to 495 miles between critical disengagement on average:
Ashok Elluswamy, the head of FSD at Tesla, has previously stated that for Tesla to enable unsupervised self-driving, Tesla needs to achieve the average in miles per critical intervention “equivalent of human miles between collision,” which stands at 700,000 miles, according to NHTSA.
Musk is moving the goal post on Tesla Full Self-Driving
The current stagnation is disappointing for Tesla fans and surprising even to critics. Even those who don’t believe Musk’s ambitious timelines for Tesla to achieve self-driving believed that the system would improve faster.
It needs to improve faster if Tesla wants to go from ~500 miles between critical disengagement to 700,000 miles – the company’s own goal of being safer than humans.
We believe that a possible reason for the current stagnation is that Tesla is focusing on a new strategy for self-driving.
Last month, we posted a report called ‘Elon Musk is about to masterfully move the goalpost on Tesla Full Self-Driving‘. After years of being wrong about Tesla achieving unsupervised self-driving, Musk badly needs a win on that front.
That’s why instead of delivering on its long-stated promise of consumer vehicles achieving unsupervised self-driving, Tesla is shifting to releasing a ride-hailing service in a geo-fenced area around Austin, Texas in June.
Using an internal fleet of vehicles helped by teleoperation in a limited area is a complete change of plan for Tesla self-driving, and it is a service similar to what Waymo has been offering for years. Musk has even thrown colder water on Waymo’s approach, calling it “too difficult to scale.”
We believe that part of the reason why Tesla FSD is stagnating is that the automaker is currently using its engineering power as well as its training compute toward this new program rather than its broader FSD product.
Electrek’s Take
You know my take on FSD. I think that if it was developed in a vacuum without Tesla selling it as “Full Self-Driving” and Musk promising that it would be unsupervised by the end of every year for the last 6 years, I think it would simply be praised as the best level 2 ADAS system out there.
Unfortunately, it’s not the case.
Instead, Musk has tainted the product with lies and false promises and we are not even getting into HW3 in this post.
I think Musk has been really successful at misleading people with FSD, and now he thinks that this pivot to a Waymo-style product will enable Tesla to claim a win on self-driving without most people realizing that it’s actually a loss for millions of Tesla owners.
He might be able to pull it through, but we are going to keep reporting it for what it is on Electrek.
FTC: We use income earning auto affiliate links. More.
