FSD v12.x (end to end AI)

[kpanda17]

Well since today is THURSDAY and no evidence of 12.4, much less 12.4.1 that sounds as likely as all of us pooling a couple of dollars and winning the $1.2B lotto.

the week is not over

 

[drtimhill]

1. Clearly an issue that they need to fix, but I’ve noticed something interesting with 12.3.3 - it accelerates quite well up to about 5-7 MPH under the target speed then sits there totally content. If I don’t focus on the speed and there’s no one behind me it’s not at all unpleasant to let FSD do its thing. Not what I would do but not dangerous in any way and not illegal.

I've heard others say this, but I rarely find that myself. In fact in my case it seems to push the speed limit a bit much, hitting 35 in a 25 zone for example (which, around here, is edging into getting a ticket territory). using Average setting with the smart speed on. The only times I see the car drive below the limit is when I would also (traffic and/or road conditions). I wonder why the different experiences?

 

[Electroman]

Seeing more and more of this...

 

[sleepydoc]

I've heard others say this, but I rarely find that myself. In fact in my case it seems to push the speed limit a bit much, hitting 35 in a 25 zone for example (which, around here, is edging into getting a ticket territory). using Average setting with the smart speed on. The only times I see the car drive below the limit is when I would also (traffic and/or road conditions). I wonder why the different experiences?

Do you use the ASSO setting? I have it in the ‘old’ setting that lets me set the max speed.

 

[Dewg]

So far I'm finding it (12.3.3) still hanging out at the speed limit when I'm in light traffic or no traffic. I've been waiting for it to do 35 in a 25, but so far it hasn't. Note: I'm in Chill mode (at all times).

However, one thing ASSO did for me this morning, which was not a preferred behavior, was this:

Two (2) right lanes were set to enter the freeway a block ahead. For some reason the right lane was pretty empty, but the left lane had a lot of traffic. FSD moved into the right lane, but maintained the speed of the cars in the lane next to it (left). They were all going about 35MPH, with the speed limit for that street being 50MPH. In this case, it was following traffic around it, but it wasn't the correct option, as there was no one in front of me, so it could have accelerated a bit. I know that it's generally safe practice to not drive significantly faster than the lane next to you, in order to avoid an accident if someone cuts out of the slower lane.

 

[JulienW]

the week is not over

I'll remind you Monday since it is sorta like the Dec. proclamation you made that v12 was coming in two weeks.

 

[JulienW]

Do you use the ASSO setting? I have it in the ‘old’ setting that lets me set the max speed.

I keep switching back and forth trying to decide which is the LESSER of the two since neither work well. ASSO works OK on some streets but can vary so much on other streets, especially on the low speed end. Max seems to work better on other streets that ASSO fails but it also can be inconsistent. I sure hope Tesla trains hard on getting some consistency in 11.4. Would also love to see the 3 modes reflected in the % of MAX speeds it maintains.

 

[Therealm80]

Had this happen after an update forever ago. I just cleared the mirror fold location and then set it again in my garage. Has been solid since. I think there is some uncommon bug which clears or alters driver save data afterban update. Every couple of updates I have some user setting revert. Usually it's benign however. Like changing the location of the blinker cameras.

how do you clear the mirror fold location?

 

[rlsd]

Another point I want to add is that I wish Tesla implement FSD Supervised more towards ADAS centric.

I was watching Chuck Cooks video this morning on protected left turn testing. I question the attention the driver need to monitor the car and surroundings during the turn. That is no way to make the user more "relaxed". For me personally, I would just drive to the stop line, one look at the left traffic, if it's very busy, I just relax and do my thing and come back to check the conditions after I think the traffic had passed. But not with FSD, I need to monitor conditions all the time, worry about FSD not making the right move, etc. That's not relaxed driving at all! Not ADAS at all. A true ADAS in this situation would tell me to relax and FSD is not going to do anything until it signals to me that its actively thinking of going ahead again, and I need to monitor its execution.

I asked for this before.
I think FSD is very weak in user experience aspect. Tesla is very good in engineering technologies; however I think they need to hire a good UX chief designer.

 

[enemji]

Another point I want to add is that I wish Tesla implement FSD Supervised more towards ADAS centric.

It should plan the route so that such edge cases are minimized if not avoided completely

 

[OxBrew]

Another point to consider for those asking “how in the world could they let this version out for wide release???”

From what I’ve seen in the comments, the variability in experiences between users is much greater with this version than it has been with previous versions. Even if we leave @AlanSubie4Life out of it I don’t think we can chalk it up to personal preferences. I don’t know if it’s related to location, hardware configuration or something else but it’s definitely real.

Whatever the reason, if the testers (and Elon) have my experience then it would make perfect sense to do a wide release. If they have others’ experience it wouldn’t.

It's on Tesla to test in every situation before release. Standard engineering practice: research and understand the entire set of environmental conditions your design is subject to, and relentlessly test against that. Then check again, and have unbiased peers double check your work, your assumptions about everything, your research, try to poke holes in everything.

Just because it works well sometimes means almost nothing if someone else can find flaws as easily as many of us do.

FSD cheerleaders: you have lost the plot. There is no place for glass half full optimism. You need to deeply interrogate the failures until you understand their true root cause, and can make an infallible argument for how they have been fixed, that even the highest level subject matter expert engineers cannot find fault with.

That a lowly hardware engineer like me can find glaring errors in literally 30 seconds means there is a deep problem with the dev process.

Reliability testing is not "find a few scenarios where it works and call it good"

It is "find every place it can fail, and deal with all of them, one way or another"

So far, 0 for 2 in the intersection below.

Orange is my route home almost every day.

v11 stopped at the blue creep line and instantly tried to pull out in front of 2 cars doing about 45 down the hill (purple arrows). I slammed on the brakes.

v12: had the path planned great, but wavered on the lane choice, finally disengaging with no warning, left me at the green box position, rolling slowly toward the median.

I took over, got to the creep line, tried to re engage FSD, but no dice, not allowed. Better than blindly driving into traffic, but still bad.

view from car

 

[frankvb]

Is there a way to prevent the car from going into the HOV lane? On FSD, my Model 3 quickly got into the fast lane and then abruptly tried to change to carpool lane. I don't have HOV stickers and I was the only one in the vehicle. I had to disengage quickly and get back into fast lane.

There's an 'avoid/use HOV lanes' option in the navigation settings. Did you set that to 'avoid'?

BTW: I usually have the opposite problem, car likes to exit the HOV lane 10 miles before an upcoming exit navigation plans to take. Though possible map data is outdated, this is going north on I-5 past Solana Beach. Speed limit detected is still 55mph there, even though construction ended a year ago or so.

 

[bhakan]

I think we need to admit that no one understands how ASSO works and it’s likely a random AI-generated speed target that happens to coincide with what we do often enough to make us think we’ve done something. In reality it’s just the FSD engineers watching cabin cam videos and laughing hysterically.

There is no method to madness

 

[FSDtester#1]

I keep switching back and forth trying to decide which is the LESSER of the two since neither work well. ASSO works OK on some streets but can vary so much on other streets, especially on the low speed end. Max seems to work better on other streets that ASSO fails but it also can be inconsistent. I sure hope Tesla trains hard on getting some consistency in 11.4. Would also love to see the 3 modes reflected in the % of MAX speeds it maintains.

I noticed today on a road I use all the time, someone started tailgating me, 40 mph speed limit, car was going 42, then sped up to 49 to put distance between us.
It never has gone 49 on V12 before today.

 

[sleepydoc]

It's on Tesla to test in every situation before release.

It’s almost like they should release a beta version the. have a bunch of drivers test it widely to find problems. Oh, wait…

 

[KLam]

It’s almost like they should release a beta version the. have a bunch of drivers test it widely to find problems. Oh, wait…

I am thankful that Tesla releases such great toy (useful when you are careful) for me to play with.

 

[Porkalletta]

Anyone got Park Assist with version .3.3 on their X or S? Haven't had it on my 2022 refresh X forever. Is Tesla ignoring us?

 

[kabin]

It’s almost like they should release a beta version the. have a bunch of drivers test it widely to find problems. Oh, wait…

Yep and it continues to be a swing and a miss. The team and their NNs have something in common - they can't learn.

Waymo had a much more thorough approach to testing and priority for safety even early in the days when Waymo drivers were behind the wheel. I'd love to know how much each company has spent to date including lawyers and settlements.

 

[KLam]

Anyone got Park Assist with version .3.3 on their X or S? Haven't had it on my 2022 refresh X forever. Is Tesla ignoring us?

Yesterday, I drove through a parking area at slow speed ( approx. 8 mph) and saw the P select marker displayed on my front center small screen. Didn't try to park though.

 

[AlanSubie4Life]

Likewise, I’ve never had it stop in the middle of a maneuver.

I kind of feel like people have doubts sometimes but I have literally never made anything up here.

Try an unprotected left across multiple lanes of traffic with some traffic from the right, about 20-30 times. I’ve posted the location I am testing before but I do not think it is limited to that location. (I only have had this happen in 12.3.3, about 20-30% of time, but I may not have had enough trials in 12.3 to see it.)

Exhibit A&B. In my case it was more severe; I was sitting completely in the first lane, maybe even crossing to the second. Going 1mph. But it might have stopped; did not wait long enough to find out.

(Note Chuck has disengaged here, but irrelevant; a different screen capture immediately prior shows FSD engaged at 0mph, same position.)

I wonder why the different experiences?

For identical streets, speed limits in Oregon and Washington tend to be 10-15mph posted speed limits. I think that makes speeding more likely, even though ASSO & Manual mode take speed limits into account somehow.

Do you use the ASSO setting? I have it in the ‘old’ setting that lets me set the max speed.

I maintain the two modes have no differences other than the target speed.

ASSO works OK on some streets but can vary so much on other streets, especially on the low speed end. Max [Manual] seems to work better on other streets that ASSO fails but it also can be inconsistent.

I maintain that the two modes work identically (with the exception of the limit). I’ve never observed any difference in their behavior. Are there any videos of side-by-side comparisons on the same streets? I just switch back and forth on the fly and the speed does not change suddenly, or even slowly in most cases. Maybe it does make a difference sometimes? Never seen it though and since the modes seem the same I don’t know why it would.

Chuck is 4/7 or 5/7 on UPLs with 12.3.3. It was about 0/3 with traffic from the right. I don’t know what that means for the bet, @Daniel in SD , given minimum number of trials requirement.

We’re not going to get more trials I don’t think. But this seems unrecoverable. First 9 not unlocked. Maybe 12.4, after they incorporate the training (they don’t seem to have retrained in a month or so).