FSD v12.x (end to end AI)

[sleepydoc]

Odd that a CEO would hype their product like an overactive cheerleader:

I mean, technically every new version is, by definition, the best they've ever made - why would you make a new version that's worse by design?

No, every new version is the newest they’ve ever made. There are plenty of examples of new versions of software that are worse than those that they replaced.

 

[sleepydoc]

Yes, when you car is doing something that nobody else is complaining about.

Like Alan’s car?

 

[FSDtester#1]

Like Alan’s car?

According to him his old Subaru is closer to L3 than his Tesla....

 

[lzolman]

Omar is cringe and I think Sawyer guy is an AI bot

I saw him in person lol, at a Boston-area Tesla club meetup. He brought his CT.

 

[SSDTester#1]

I saw him in person lol, at a Boston-area Tesla club meetup. He brought his CT.

Wonder if he’s using a bot to post then lol

 

[Tronguy]

I would also expect inaccurate distance and positioning in the visualization as a necessary condition of miscalibration. To anyone who has benefited from a re-calibration, were you seeing issues with the visualization?

Just got back from one of my periodic 500-mile round trips from NJ to Boston and back; FSD(S) all the way!

As it happened, the SO's daily driver had a left fender camera defect: Water got into the assembly and killed the camera. So, dinged the app, mobile tech came down in a few days, and replaced the thing under warranty. Interestingly, the guy suggested that a camera cal could be done, but wasn't necessary. And then it came to me: Both the SO's car and my daily driver, an M3, had been punting around with 12.3.6/FSDS. The SO's because it was that freebie trial; mine because I had bought the package a loonnngg time ago. Both cars, on city streets, had been kinda driving right of center on the roads. So I told the guy to go for the "Reset the Cal" option, which he did. In Service mode, even.

Took the car for its 20-odd mile run up a local interstate (lots of painted lines, cal goes faster the manual says). On the way back on the interstate, after the cal was done, the MY was definitely nailed to the center of the lanes. Except when passing trucks, when it shifts away from them.

So, in a fit of whatever, I cleared the cal on the M3 and repeated the process.

At that time, in that place, I thought that made a difference in the centering. Having just finished driving 500+ miles back and forth, a goodly number of which was on city streets, it's not a thought: It really did center the car.

Before the camera cal runs, neither car had been through a cal run; hers since 2021 when we bought the MY, mine since 2023 when I traded in my old 2018 M3 for the newer variant. Both, however, has had all the software updates that have come down the pipe.

 

[lzolman]

J related, I think it's funny when my SO gets an update on her MY, she always asks me about it thinking that I get it in my car before she does.... little does she know, I don't get the updates she gets until several months later lmao

Wouldn't it be funny if your wife was like mine and refused to ever activate FSD?

 

[aronth5]

How many times do we have to hear this only to be fairly disappointed? I am not saying we haven't seen improvements - we have with every release of course. But it's never been transformational, even with v12 (which does drive less robotically, but with a fairly similar intervention rate).

To drive interventions down, Tesla must execute on:
1) Reading signs correctly.
2) Getting speed control correct.
3) Getting stopping behavior correct.
4) Getting turning behavior correct.
5) Getting lane selection correct without decision wobble.
6) Get unprotected lefts working to the first or second 9.

Those are the key items which are likely highest in the Pareto.

Most of your list is not a priority for me since my car drives pretty well. My passengers rarely complain about ride comfort they always complain when the car doesn't do one of the items below. Do agree with #1 on your list. I am much more interested in the following list since they definitely affect if FSD will get me to the destination safely and many are the law. That is my priority and Tesla's I suspect for 8/8. Except for #1 none of the above would normally prevent you from getting to your destination safely.
My priorities in no particular order

• Road closure signs so FSD automatically reroutes
• Emergency vehicles- pull over it's the law
• School zones- slow down it's the law
• School busses- stop it's the law
• Hand gestures- stop/go when directed- it's the law
• Backing up so you can get to your destination
• Avoid Potholes
• Read signs correctly- Example "No left turn" etc.

 

[lzolman]

Depends, could be 15 miles, or 30. Highways are the best place to recalibrate.
Pull over right before getting on one, then start the recalibration.
It first recalibrates standard autopilot, then it reminds you it is still recalibrating for Fsd.

Around a month ago I cleared the calibration on Nameless ('21 MSLR refresh) and it has still never successfully completed the full calibration, despite lots of highway miles. I've actually tried a few times in that period; the first time, it hadn't finished the first phase even after over 100 highway miles so I cleared it on an onramp to I-95 and tried again, and that time it did complete phase 1 after 20 highway miles and then about 8 more on backroads. But it has been over 500 miles since then, and still no FSD.

I do have a mobile service scheduled and they're thinking about changing it to a Service Center visit. I'm wondering if anyone has had this happen?

 

[MP3Mike]

But it has been over 500 miles since then, and still no FSD.

If you go into service mode doesn't it show the percent complete on each camera? Then you would at least know which camera(s) are an issue. Are there any active alerts?

 

[gottagofast]

Around a month ago I cleared the calibration on Nameless ('21 MSLR refresh) and it has still never successfully completed the full calibration, despite lots of highway miles. I've actually tried a few times in that period; the first time, it hadn't finished the first phase even after over 100 highway miles so I cleared it on an onramp to I-95 and tried again, and that time it did complete phase 1 after 20 highway miles and then about 8 more on backroads. But it has been over 500 miles since then, and still no FSD.

I do have a mobile service scheduled and they're thinking about changing it to a Service Center visit. I'm wondering if anyone has had this happen?

It took mine this long too when I calibrated

 

[AlanSubie4Life]

Like Alan’s car?

What’s weird is how my car was behaving like everyone else’s on the prior release after each release. It’s very odd!

 

[animorph]

Our 3 started hugging the left lane line after a recent windshield replacement. I started a cal. SO was pissed that her TACC wasn't available. But a couple of short drives later (20 miles total?), TACC and FSD were back and the car was centering again as usual. A previous windshield replacement did not require a cal as far as we could tell.

As long as you don't get stuck with an endless cal it is pretty painless. I assume a cal problem indicates that a camera needs to be aimed correctly by Service. That might be good to know.

 

[gsmith123]

A previous windshield replacement did not require a cal as far as we could tell.

I'd think recalibrating after a windshield replacement would always be recommended. The main cameras are just glued (?) to the windshield so positioning is prone to move when they're removed and reattached.

 

[LSDTester#1]

It took mine this long too when I calibrated

He said his car won’t calibrate.

 

[SSDTester#1]

Wouldn't it be funny if your wife was like mine and refused to ever activate FSD?

She doesn’t like FSD or AP. Her car doesn’t have FSD either so she gets all the other updates before me.

 

[AlanSubie4Life]

What is limiting progress if training isn't to blame? Design? Execution?

It's just a really hard problem that no one knows how to solve, and it's being attempted on extremely compute-limited inference hardware, with quite limited camera resolution, which makes it even harder.

I think they'll figure out ways to get better results, but it's going to take a while. More training resources will probably help them iterate faster.

I am much more interested in the following list since they definitely affect if FSD will get me to the destination safely and many are the law. That is my priority and Tesla's I suspect for 8/8. Except for #1 none of the above would normally prevent you from getting to your destination safely.

Your list seems to be from the perspective of having a product than can handle every situation adequately, so that it can be completely hands off. Kind of the "jack of all trades, master of none" approach. I am more interested in lower intervention rates where it masters the most common tasks and rare occurrences can be mostly dumped on the human. I can understand the alternative philosophy, but I just prefer to have lower intervention rates, and it sounds like this is a Tesla metric as well.

I just don't think many of the things you list would show up at the top of the intervention Pareto.

• Road closure signs so FSD automatically reroutes

Yep, read signs.

• Emergency vehicles- pull over it's the law
• School zones- slow down it's the law
• School busses- stop it's the law
• Hand gestures- stop/go when directed- it's the law

These just are really rare scenarios for me so I have no problem just taking over for them. (School zones - need to read signs, falls in that category which I think is important as it was in my list!)

• Backing up so you can get to your destination

Should never need to do this if it's reading signs, and it seems like just a situation where it will nearly always be better to just take over.

• Avoid Potholes

Yeah would be nice but seems really, really hard and I don't want the car waving all over the road dodging imaginary potholes. I'm ok just waiting for the potholes to be repaired and when driving the same route I know where to disengage. Some memory or whatever would be nice.
But just lower priority for lowering intervention rate. Some roads in San Diego are just unusable for FSD, and I can't imagine a pothole-dodging ability would be acceptable. Check this out, for an example - and this is BEFORE two solid years of rain! I think we just have to accept that something that is essentially a gravel road complete with potholes, except worse, will be impossible for any implementation, ever. It'll just have to go really slowly rather than veering wildly all over the road at 25-30mph like I do, or just hand off control to the driver. (It's possible this specific example has now been fixed, I'd need to go check - but there are always others!)

• Read signs correctly- Example "No left turn" etc.

Yes it should read signs as indicated in my list.

That is my priority and Tesla's I suspect for 8/8.

Nothing is happening on 8/8 for our FSD. I'm not sure why people think it will be significant. This will be like Battery Day or AI Day 1/2 or whatever. Nothing happens. What could possibly happen on 8/8 which would be significant for current vehicles?

It's possible they will reveal FSD Beta, the beta version of FSD. But vehicle support may be limited. We'll see.

 

[JulienW]

Well it's morning and TeslaFi shows nota and Elon indicating there is a "hold in the countdown". So even this weekend is starting to look iffy. Hope we see 12.4.2 polished up and sent to influencers by tomorrow (Tuesday) so we can have some hope of getting 12.4.2 this weekend.

I see S5OD has the team working hard and they even have the new eye tracking glasses to calibrate the eye monitoring system.

 

[JulienW]

Just for interest here is a hacked (salvaged) FSDS 12.3.6 in Moldova.

 

[aronth5]

Your list seems to be from the perspective of having a product than can handle every situation adequately, so that it can be completely hands off. Kind of the "jack of all trades, master of none" approach. I am more interested in lower intervention rates where it masters the most common tasks and rare occurrences can be mostly dumped on the human. I can understand the alternative philosophy, but I just prefer to have lower intervention rates, and it sounds like this is a Tesla metric as well.

Exactly. My intervention rate I'm sure is now far lower than yours since I'm fine with the ride in almost all situations. Drives are very comfortable for me and passengers so addressing the very few specific items I have is my priority. So much of your list I don't care about since ride quality has improved over time and will continue to do so but my list hasn't improved much so that is my focus. Over 95% of my drives are disengagement free so getting that last 5% addressed would be awesome. I don't want or need the "rare" occurrences dumped on me especially when they are more than rare. We have a lot of road construction which accounts for a good portion of my list.