This SHOULD be way it is but for some strange (aka: Tesla) reason it is actually FSD(S). Supervised sounds clumsy at the rear and why the parentheses????....I can see using SFSD over time... SFSD V12....
You can install our site as a web app on your iOS device by utilizing the Add to Home Screen feature in Safari. Please see this thread for more details on this.
Note: This feature may not be available in some browsers.
This SHOULD be way it is but for some strange (aka: Tesla) reason it is actually FSD(S). Supervised sounds clumsy at the rear and why the parentheses????....I can see using SFSD over time... SFSD V12....
What does simulated data have to do with the way the system handles input? Every situation needs handled.You've left out the fact that Tesla also uses simulations to train FSD.
DudeThis has aged very well! Plenty of complaints, as I expected. I do not have to concede I was wrong! Whew - I was starting to think I had overstepped last night. (I'll gladly concede when I am wrong, of course.)
Yes, it does seem that it goes very slightly faster. Tonight, I saw 47mph in a 45mph (manual target was 52mph). And I think I saw 52mph in a 50mph (manual target 57mph). I think before I might have seen 43mph-44mph and 48mph-49mph on these same roads? But I'd have to review my footage (which I'm not going to).
My initial sense that maybe it was a couple mph faster seemed to be correct, even though I doubted it, and I wholeheartedly concede that is an improvement, which is why people say it is "better" - but it is far from resolved.
Of course, on freeways, the nice familiar strong behavior comes back with v11 - it just goes the speed you ask for! Not looking forward to v12 on freeways until they fix this problem.
To me this seems like the same FSD as 12.3 though. I don't think 12.3.3 is fundamentally retrained. It seems likely that they somehow just tweaked some parameters to achieve this ASSO change, rather than doing retraining. The behavior is just so similar in all the areas where I have problems - I cannot detect any obvious differences which you'd expect to see if it were actually retrained. It just drives exactly the same way in exactly the same places on the street.
For the unprotected lefts there definitely seems to be a change in a "caution" parameter. It just sits in the middle of traffic lanes now most of the time on my unprotected left. Creeps, then commits at 3-5mph into traffic lanes, then just stops (1-2mph). No relevant traffic present. This did not happen on 12.3. However, in every other way it approaches this turn the same way (down to the exact same incorrect lateral position on the unmarked street, blocking right-turning traffic), so it seems like 12.3 and 12.3.3 are the same underlying neural net. 3/4 times I have tried this turn, this is what it has done. I've seen others report the same today. I had no such failures on 12.3.
So I conclude that 12.3 and 12.3.3 are actually the same, with parameter adjustments.
I think at this point everyone knows how it all works:
1) In ASSO mode, speed offset is +50% (it doesn't seem to matter if you're in chill, moderate, assertive, @arnolddeleon - I checked this tonight as best I could). So in a 50mph it's set to 75mph, etc. Presumably it maxes at 85mph but I haven't checked. This is adjustable (invisibly), and it acts as a cap on speed.
2) In manual mode, with % offset of your choosing, it seems to work as before, with one major difference: it won't (in general) go at your requested set speed (this is the big problem that we hear many complaints about, for which there is no workaround, and it is a clear regression from v11). It is adjustable on the fly as normal, and it will honor it as a cap, of course.
3) When switching on the fly from ASSO mode to manual mode, the ASSO limit will be preserved (that's how you determine what ASSO limit is set to empirically). If you disengage and reengage the manual limit will come into force as normal.
No. If you have a set speed offset (which can only be a % now) that will be honored as a cap. In ASSO mode that % is set to 50%. In manual mode it's whatever you set.
ASSO mode has a terrible interface, but since this is a cap, it's quite unlikely to go 40%+ over the limit.
In my particular area, I've only seen it going about 5% over the limit, max. And it's often so slow to get to that point that on average it is under the limit.
It certainly could go higher than that. It's probably calibrated to limits on California roads, not Oregon roads, which I know from experience are much lower limits than California roads, all else being equal. That may be why we see higher excursions in Oregon. It should take the limit into account of course but it does not seem to (as usual you can use manual mode to just apply a cap, though).
But overall I don't think this is a major source of concern for releasing to all Tesla drivers. I think it's fairly well calibrated to the average driver, and some will think it accelerates too fast, but that's not related to speeding.
You stated that the number of situations encountered via video clips from the cars was not sufficient to cover all possible real word scenarios. My point is that video information is augmented by simulations to increase the possible scenarios that are encountered.What does simulated data have to do with the way the system handles input? Every situation needs handled.
If used to train NN, the system is NN. If the system is hand coded, the simulations represent more test cases to write for or train upstream NN to properly classify.
I don't think I did though. If I did, it was unintentional as I agree simulations can add variation to the data set. And collecting the imput data is distinct from handling the data.You stated that the number of situations encountered via video clips from the cars was not sufficient to cover all possible real word scenarios. My point is that video information is augmented by simulations to increase the possible scenarios that are encountered.
There isn't enough storage nor compute in the universe to code the proper response to every possible combinations of inputs.
The only reason any approach could be feasible is that things can be generalized.
Apply that across all values and it's beyond human ability in a useful amount of time, but a big block of training compute can tweak all the parameters simultaneously while checking for regression.
I don't even live in CA (any more), but SFSD sounds like it's only good on trips from San Fran to San Diego.This SHOULD be way it is but for some strange (aka: Tesla) reason it is actually FSD(S). Supervised sounds clumsy at the rear and why the parentheses????
Is it possible there was a car in front of you?Interesting, mine consistently stayed above the speed limit (5 - 10 mph) in limited testing (an hour or so). Not sure if it makes a difference but I have mine set to "aggressive", slow was never an issue.
What is your speed setting? Manual/Auto? Offset?Tried it a second time this morning. Still had to disengage about once per minute. Pre-translated into American for you guys:
1) Incorrect lane positioning. Tried to go into the bus lane / right turn only lane a couple of times, or it would drift into the parking lane which is full of rocks and other crap. This is a piss-off-other drivers sort of thing, so is a safety issue in my view.
2) Doesn't smoothly choose a speed. Accelerates too hard, overshoots, then immediately starts slowing down. On certain roads, will still travel too fast if you use auto max. This morning drive was on roads where on most segments, going at or over 25mph is being an a-hole, and only a few segments can you go up to 30 - 35mph safely. There are schools on many of the different possible routes. Also a safety issue in my view.
3) Can't seem to find an appropriate speed to brake to stop to a stop sign. This is mostly not a safety issue, but is annoying.
* Google's preferred route has me passing through 17 stop signs and 3 lights over 2 miles. All of Google's routes pass by busy schools. Imagine doing 17 annoying stops over 2 miles and also passing by busy schools with lots of kids, crossing guards and school buses. Also, part of the route goes over the surface of the moon. Not a good idea.
* My own custom route has 7 stops and 10 lights and doesn't pass by any schools, but the car never shows it as an option. I tried doing FSD without navigation but it stubbornly turned off my turn signals each time I tried to turn, so I used the navigation with an extra pin dropped.
What it should be doing for stops:
a) Quickly start decelerating to the rate which will lead to the car reaching the stop line at a constant rate of deceleration.
b) Hold that rate of deceleration until close to the stop line, then slow down the acceleration when about 20 feet from the stop line.
c) Continue decreasing the deceleration so that you reach the stop line with zero jerk -- i.e., you can't even really perceive the point at which the car stops moving. To make it really smooth, don't actually lock the wheels but always stay moving at around 1-2mph ( I know that FSD is not allowed to do this because "reasons", but this is the only way to be smooth).
I find manually driving a Tesla is actually a lot easier to do this than in automatic gas cars which are always fighting you. My stops are usually butter-smooth. However, FSD drives more like those automatic gas cars with the rough stops.
This sort of calculus is the sort of thing a computer can calculate faster than a human, so it seems to be some sort of NN control issue.
What it should be doing for speed is:
a) If the road is narrow with parked cars and is clearly residential, the car should not be travelling above 20mph. Around here at least, most roads like this are already posted at that limit.
b) If the road is wide and in a more open area with lots of visibility, then it's safe to go faster. Under no circumstances should it pass 30mph on its own (due to possible police presence), or ever touch 25mph near a school.
The car doesn't do this right now. I guess auto max is intended to solve this problem but it goes way too fast on the residential streets. There's no way it would be able to stop for someone coming out from between parked cars or opening a car door, or even someone shooting through their stop from a side street. This sort of thing happens ALL THE TIME so the car has to drive at an appropriate speed. Sadly, some human drivers don't do this either, but FSD is supposed to be better.
I haven't yet been able to test how it will behave if a school bus sticks out its stop sign. Usually, but not always, you get about 10 seconds warning with flashing yellow lights. That's a nasty ticket around here, so you want to be extra vigilant for that even though a lot of people aren't.
One last note: It was mildly foggy this morning and I was getting dry wipes, but when turning off the wipers, FSD would complain with a "beep beep beep: FSD is degraded" every 20 seconds. You could see about 1000 feet down the road at least so this was quite annoying.
Will continue testing and reporting issues to Tesla.
Currently using auto max speed setting.What is your speed setting? Manual/Auto? Offset?
Third drive this morning: well, it was like night and day. The thing actually drove quite well and even chose appropriate speeds, and didn’t take off from lights like it was in a street race.
I also notice a lot of variation from one drive to the next, but even with no other cars around.I think the difference is what other people have noted: having other traffic around helps a lot.
Currently using auto max speed setting.
I’m not sad anymore about being left behind.v12.3.3 drive yesterday:
2 trips in the morning worked fine.
The trip in the evening was not good:
1. Car moved to the lane to go freeway 15 south instead of continue to go straight to go 15 north 0.1 mile ahead.
2. Car made unnecessary lane changing even when Minimal Lane Change was set.
This is a regression.
I always use Chill mode and don't bother to try the other modes because it is useful to me.I’m not sad anymore about being left behind.
On another note, 12.3 was weird yesterday and I think I figured out why it was hugging the left lines, (white or yellow), and taking left turns short. Yesterday I g had it in Chill Mode, today I put it back to assertive and it was just fine. Until this morning, I noticed no discernible difference. I’m sure it was probably other factors influencing it, but I’ll know later today.