This is classic Elon. He just claims that Tesla is "close". We don't know what "close" means, we don't know what "so rare" means. His disclaimer that "No point in doing so until passing that threshold" gives him a way out if progress slows. Elon will just tweet that "FSD is harder than we thought" and promise "2 weeks" and "next update will blow your mind".
Yes that is what he is saying. But from our experiences, it seems unlikely that city interventions are close to being rare enough. In fact, even the Tesla faithful on Twitter are telling Elon that they are still seeing a lot of interventions in their city driving.
To be fair, I don't think Elon is claiming city interventions are close to being able to remove supervision. He is just claiming that they are close to where the team can shift focus to the highway stack. He is just talking about when FSD Beta is "good enough" that the team can start working on something else. For all we know, maybe Tesla's internal metric for shifting to highway stack is when city interventions are 1 in 100 miles. In which case, Elon might be telling the truth that city interventions are "close". Since Elon won't release disengagement data, we really have no idea.
It does raise some questions though. When Tesla does shift focus to the highway stack, will that mean that FSD Beta in the city is done? Will Tesla not do any more work on FSD Beta at that point? I hope Tesla will continue working on city driving after they shift to the highway stack. But it sounds to me like Elon thinks FSD beta is close to being "good enough".
There's no way they are anywhere near only 1 intervention per 100 miles on city streets. I have a disengagement every 1.25 miles or so on 10.10.2, plus a go-pedal tap intervention every few miles.
That said, roughly 80% of my disengagements are things that don't happen on the highway:
- mailbox-related swerving
- too far left on un-marked roads
- can't turn right without straying out of the lane on sharp or somewhat sharp right turns
- signaling right approaching roundabouts, even if not planning to take the first exit
- trying to pass queued cars at stop sign / traffic light by driving in oncoming travel lane
The only categories of my most common disengagement reasons that could happen on the highway are:
- following a lead car too closely, especially in cold or poor weather
- poor lane change choices
So, if they would increase the follow distance or allow the follow distance setting to work on FSD, and respect the setting to require lane change confirmation on the highway, I would feel pretty good about how FSD should perform on the highway. An added bonus would be if they can finally figure out how to properly visualize the position of a semi truck beside me.