PACEMD
Active Member
Now that hurts my delicate little feelings. Ouch! I've got half a mind to disagree..........you know, the thumbs down thingy.......
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
PACEMD
Active Member
No. No, it is not. You read that wrong. Today is oppositional day. And I'm opposed to you saying the opposite......
Pdubs
Ingesting large quantities of LSD
I want to party with you. As long as you keep the sitting around the camp fire bobbing our heads to the Mamas and Papas out of the equation and maybe bring some psilocybin or some LSD while I crank up some Zappa.
I've noticed recently that the car 'sees' in the in-car visualization lanes that can't possibly be seen. For example, I have a road near my house that is parallel to a road offset vertically above it. When on this lower road, the car 'hallucinates' the adjacent road's lane markings, even though it can't see them.
This indicates that the car is using mapped lanes in its inputs for finding lanes (but, apparently, without full knowledge of Z-axis). Although this goes against the mantra they've been parroting about driving anywhere without maps, I don't mind the car using knowledge of lanes outside of its vision, as lanes don't change very often.
I don't remember this happening before the .69 update, but I may not be very observant. Anyone else see this behavior in their car?
This indicates that the car is using mapped lanes in its inputs for finding lanes (but, apparently, without full knowledge of Z-axis). Although this goes against the mantra they've been parroting about driving anywhere without maps, I don't mind the car using knowledge of lanes outside of its vision, as lanes don't change very often.
I don't remember this happening before the .69 update, but I may not be very observant. Anyone else see this behavior in their car?
No, My point is that V11 along with a couple of point releases should at the very least improve highway driving. Many of the V11 release notes speak to improvements for the highway. If the V11 update can't do any of that than it's a major fail.
clydeiii
Member
It’ll be different, so if some things are worse, it won’t be a major fail. They’ll just work to smooth out those issues over time as they have done with every other release.
Definitely will take a bunch of updates before V11 will be where current AP is.
From past performance, when Tesla adds inputs it just makes thing worse until they figure them all out and fix the new interactions.
I get rhe impression that maybe AP is more confident because it oblivious to some of the things FSD can “see”
This has happened since the very first FSD Beta in 2020 with people claiming usage of HD maps. But it's a "feature" of neural networks making reasonable predictions as a lot of the training data includes adjacent parallel lanes, e.g., 2-way roads split by a median. FSD Beta 10.69.3's release notes have a first entry highlighting low visibility scenarios that probably include nighttime and heavily occluded scenarios.
n.one.one
Info w/o Drama
It looks like software versions installed (both beta and non-beta) are now very tightly grouped in the 2022.44.x.x range. I've never seen installed versions this up to date.
TeslaFi has 18,754 contributing vehicles and 89.7% are in the 2022.44.x.x range. Teslascope seems to have similar stats but I didn't run their numbers.
This seems like a good scenario for a unified single stack release for almost all vehicles.
TeslaFi has 18,754 contributing vehicles and 89.7% are in the 2022.44.x.x range. Teslascope seems to have similar stats but I didn't run their numbers.
This seems like a good scenario for a unified single stack release for almost all vehicles.
EVNow
Well-Known Member
Happens once a year ... around holidays, since Tesla wants to update everyone to holiday release.
Didn’t this also happen once before? Last year around the holiday update?
EVNow
Well-Known Member
Yesterday had an hour long trip ... including freeway. FSD+NOA did quite well. Only when getting out of freeway at destination I had to press the accelerator .... since FSD doesn't seem to recognize that when there is a turn only lane, it is basically a protected turn. FSD stops and waits. But seconds after I pressed the accelerator to start moving ... FSD gave up and asked me to take over. Strange.
While coming back the rain was heavy. NOA kept switching between NOA/AP. Once I got out of freeway though, FSD came on and didn't switch to AP. I guess the faster speed and all that water spraying from fast moving vehicles in the front makes NOA give up faster than FSD.
So, totally about 2 hours of driving with one disengagement.
While coming back the rain was heavy. NOA kept switching between NOA/AP. Once I got out of freeway though, FSD came on and didn't switch to AP. I guess the faster speed and all that water spraying from fast moving vehicles in the front makes NOA give up faster than FSD.
So, totally about 2 hours of driving with one disengagement.
n.one.one
Info w/o Drama
LOL...I guess I'm a total failure as a historian...thanks for the correction.
Have you noticed 10.69.25's holiday update includes a new navigation display for lane selection? It might help reveal what the navigation / maps data knows or even what's missing. Today it was quite wrong believing 3 straight lanes were triple right turns needing to be in the left two, and FSD Beta wanted to unnecessarily switch out of the left lane. The only thing I can think of causing this confusion is that there's a very small/short/narrow median the same width as double yellow lines, and somebody drew the map data as separated 1-way roads in each direction.
The bottom 2 lanes display was correct with appropriate highlighting of the left and right turns. I would think this is only displayed for an upcoming turn where map data indicates there's more than 1 lane ahead as well as the data including the turning behavior of each lane. I'll need to check again later, but earlier today I was at an intersection making a right turn and the display showed 6 straight lanes and 1 right turn lane when the left-most 3 should have been left turn.
Anybody else notice the new display and if it shows the wrong thing resulting in FSD Beta wanting to make wrong lane selection?
Similarly, if you don't see the display with 10.69.25 when turning at an intersection with multiple lanes, it could mean the map data doesn't have the lane and turn details. This probably means FSD Beta is using a Vision-only detection and heuristics to prepare for a turn at these intersections.
These were exactly my thoughts once I got the holiday update. Sure enough, in every situation where my car screws up lane selection, there is bad map data guiding it, just as I suspected. The question is how does Tesla fix this? Apple Maps has perfect map data in my area. I know Tesla is using some sort of combination of Tom Tom and open street map. Whatever it is they are using is awful. Just plain awful.
If they are going to use any sort of map data it needs to be good map data, not the crap they use now.
Silicon Desert
Well-Known Member
I got that 25.1 on one of the cars Christmas night. I usually get the updates in the first 3 days. I'm a bit embarrassed to say I haven't even driven that car yet. I've gotten so used to being underwhelmed with the minor updates that frankly speaking, I prefer to hear from others on here before I even go out to try it. Yes, I know, you folks that try out the car 8 seconds after getting an update can just slap me silly for not being so excited about some updates. Actually, I wish I could give my updates to you folks first.
PACEMD
Active Member
I may be totally wrong but I'm a fool..........
sleepydoc
Well-Known Member
Phantom braking does that for you now!Ahh can’t wait until my highway driving experience is also ruined by random slow downs, slamming of brakes, unnecessary lane changes, unnecessary turn signals, etc
Hmm... Around here, one of the major fail areas with FSDb has been roads where the visibility of an upcoming intersection has been obscured. For example, there might be a left turn lane with left turn arrows painted all over it and the straight-ahead lane with straight and left-turn arrows painted on it. But when there's a long line of cars sitting on top of both lanes, the car's cameras can't see the arrows and, well, the car tends to get stuck. Or do unwieldy things, like try to move into the left turn lane to pass the cars in the straight-ahead lane.
Another example that's driven the FSDb nutso has been a one lane road that splits into three lanes, just past the brow of a hill. In this case, none of the painted arrow markings can be seen until the car crests the hill, at which point it's far too late to smoothly change into the correct lane.
FWIW, in NJ, the local DMV/state/municipalities are very good at putting up signs that look very much like your graphic, saying which lanes are going where, hills, flat, whatever. But, up until now, FSDb has paid no attention whatsoever to those elevated signs. It would be a major improvement if it did. Is that what's going on? FSDb is doing a better job of lane selection by paying attention to those street signs? If so, it would be a first.
Similar threads
- Article
