Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Wiki MASTER THREAD: Actual FSD Beta downloads and experiences

This site may earn commission on affiliate links.
Has anyone else noticed the windows and doors doing something extra on CLOSE… I know the windows come down on open, and go back up on close, this has been the case since inception really.. But, it seems that when I close the drivers door, either ALSO the passenger door and/or the other doors ALSO seem to do a window top up even though none of them had been opened prior. I can hear it in the car, and it is definitely new.
Well, it sounds to me like you are noticing the cold weather thing they did with the windows but this isn't new so I am not sure. In case you weren't aware, in sub freezing temps the car will keep the windows down just enough to clear the trim (window is still against the seal). The car will do this on doors that have been used recently (not sure the time frame for this) the car will then put them back up fully when you get back in. This is to help with the problem people were having of the windows being frozen in place and damaging the trim when people tried to open the door. If you already knew about this and you are talking about something else then I have no idea as this is the only thing my car does with different windows.
 
Anybody know why FSD Beta wants to do this? (Changing lanes to stay out of rightmost lane)

rightmost lane.jpg


This has happened several times on this highway with 2 lanes in each direction and a center turn lane, and more often than not, it decides to do so when someone is trying to pass on the left. Like in this picture, usually there's nobody ahead, so it's not switching because it sees slow traffic ahead.
 
Well, it sounds to me like you are noticing the cold weather thing they did with the windows but this isn't new so I am not sure. In case you weren't aware, in sub freezing temps the car will keep the windows down just enough to clear the trim (window is still against the seal). The car will do this on doors that have been used recently (not sure the time frame for this) the car will then put them back up fully when you get back in. This is to help with the problem people were having of the windows being frozen in place and damaging the trim when people tried to open the door. If you already knew about this and you are talking about something else then I have no idea as this is the only thing my car does with different windows.
I hadn’t really investigated the sub zero cold weather updates, but that could be it.. although it really only gets down to about 45 at night in the garage and higher than that for sure during the day around here.
 
Has anyone else noticed the windows and doors doing something extra on CLOSE… I know the windows come down on open, and go back up on close, this has been the case since inception really.. But, it seems that when I close the drivers door, either ALSO the passenger door and/or the other doors ALSO seem to do a window top up even though none of them had been opened prior. I can hear it in the car, and it is definitely new.
Yes
Noticed this recently and thought it was different. That it used to be that only the open door window went down.
 
Well, it sounds to me like you are noticing the cold weather thing they did with the windows but this isn't new so I am not sure. In case you weren't aware, in sub freezing temps the car will keep the windows down just enough to clear the trim (window is still against the seal). The car will do this on doors that have been used recently (not sure the time frame for this) the car will then put them back up fully when you get back in. This is to help with the problem people were having of the windows being frozen in place and damaging the trim when people tried to open the door. If you already knew about this and you are talking about something else then I have no idea as this is the only thing my car does with different windows.
Either not recently or recent must stretch back. All mine are doing it and the passenger rear has not been used since right after Christmas when the kids were here. The window is where I first noticed it recently.
 
Yes
Noticed this recently and thought it was different. That it used to be that only the open door window went down.
Well, I guess that officially makes TWO of us. What I had noticed months/years ago was that sometimes the windows were a TEENY tiny bit down.. so I got in the habit sometimes when starting a longer than average drive, of doing a little window button TOP UP.. and I could tell they all pretty much would move a small amount. It wasn’t just the motor activating but not moving.. maybe Tesla has just built this in to the cars door close process SOMETIMES. I’ll see if I can figure out what the pattern is.
 
  • Informative
Reactions: FSDtester#1
Here is an intersection that Beta can never cross without changing lave or veering off to the right. Not sure why it is so perplexing but here it is illegal changing lanes in the middle and behind a slower truck, plus it needs to be in the right lane.


Here it is getting back into the right lane a block farther. No signal light and just a last second jerked wheel over and stops blocking lane. Had to take over and try my best to clear lane. Got honked at too. Beta I hope you were embarrassed by that behavior.😰

 
Not sure why it is so perplexing
Not only is that intersection pretty oddly shaped with an angled approach and destination lanes shifted across the intersection, it's also slightly uphill enough that the cameras mounted higher than the driver can't even see the destination lanes/curbs. This means the neural networks need to infer the shape of the road from what it can see, e.g., buildings, trees, signs, other vehicles. Theoretically the neural networks can learn to predict this correctly and confidently, but this is not quite a typical intersection.

peachtree.jpg
 
Tell me there isn't something seriously wrong with 10.9. This was this morning and the 6th time in 8 drives. Hell a ten year old would know not to drive like that.

Sorry says HD processing. I'm not much of a YouTube star.:eek::cool:
This is exactly the kind of example I'd use to argue that HD maps are absolutely necessary.

Lots of drivers would make the same mistake (or close to the same mistake) the first time they encountered the intersection. On subsequent experiences they'd be perfect.

An autonomous system has to get it right the first time. and there is no way to a guarantee that it will get each intersection right unless it has HD Maps.

I don't expect Tesla to ever interpret every single intersection properly because that isn't even something humans can do.

Now I don't mind perception as a backup in case the HD Maps are incorrect or something on the road got damaged.
 
This is exactly the kind of example I'd use to argue that HD maps are absolutely necessary.

Lots of drivers would make the same mistake (or close to the same mistake) the first time they encountered the intersection. On subsequent experiences they'd be perfect.

An autonomous system has to get it right the first time. and there is no way to a guarantee that it will get each intersection right unless it has HD Maps.

I don't expect Tesla to ever interpret every single intersection properly because that isn't even something humans can do.

Now I don't mind perception as a backup in case the HD Maps are incorrect or something on the road got damaged.
Whenever someone here says lots of drivers would make the same mistake I really want to go stand at that intersection and see how many drivers make the mistake. haha.
I suspect that humans deal with that one fine because all you have to do is go straight. The road curves before the intersection. I do agree that the perception NN is nowhere close to human ability and who knows if and when it will ever get there. HD maps are a great crutch! :p
1643238514570.png
 
Whenever someone here says lots of drivers would make the same mistake I really want to go stand at that intersection and see how many drivers make the mistake. haha.
I suspect that humans deal with that one fine because all you have to do is go straight. The road curves before the intersection. I do agree that the perception NN is nowhere close to human ability and who knows if and when it will ever get there. HD maps are a great crutch! :p
I have been on foot at the intersection MANY times and never witnessed anyone driving across the lanes or being confused as to where to go (do see some illegal turn left, to the right of pic). There is nothing edge, unique or weird about the intersection. What large city in the US doesn't have MANY intersections that are similar? There are lines across that Beta/humans can clearly see and follow. Here is pic from the other side looking back at the way I crossed and you can see the lanes lines easily and it is not that much of an offset Just follow the lines.

Also 2ed pic is coming up to the intersection. I'm on foot when I took pic and didn't die.o_O

IMG_0106.jpeg

IMG_0109.jpeg
 
  • Like
Reactions: FSDtester#1
Whenever someone here says lots of drivers would make the same mistake I really want to go stand at that intersection and see how many drivers make the mistake. haha.

The problem with humans, and with FSD Beta is you can only count the times where there isn't a lead car.

This can make FSD Beta testing really frustrating.

I can't test when there is a lead car because it might skew the results.
I can't test it when there is a car behind me because it might doing something dumb to embarrass me.

Plus certain locations where I wanted to test it have been fixed where lines got painted in areas where humans got confused.

I've done a bit of testing where lane dividers are missing so it thinks 2 lanes are one lane. When there isn't a lead car it will happily recenter into the middle of the two lanes.

I haven't seen how many humans make that mistake, but I have seen a few human drivers either get WAY over or drift into the middle. So some small percentage of human drivers are struggling with having no divider between the two lanes going the same way.
 
Just to add here is an example of a REAL edge case intersection that it fails at and is more forgivable at this point. It hates driving on the railroad tracks (flowing the yellow streetcar) and trying to turn left (between the white streetcar and the silver car). It just can't do it. All the lines, crosswalks and train tracks are just too much for it to figure out and it ends up going straight every time.

IMG_7392.jpeg



EDIT and Fun Fact: At that intersection is something we don't have much of in ATL. A historical building (we learned from Sherman not to keep buildings around for long🤣). It is a the Flat Iron built in 1897. That is 5 years before the famous Flat Iron in NY.
IMG_8269.jpeg
 
Last edited:
I do agree that the perception NN is nowhere close to human ability

This might not be the best example, but I'm actually criticizing human ability.

I don't think humans on average have the perception and recognition capability required to drive safely. That even the best German drivers would sometimes get confused if they encounter a intersection that was not only unique to them, but also a bit confusing.

We can't expect autonomous driving to exceed good human drivers by any significant margin unless we help them out with mapping information, and sensor data that humans don't have access to.

HD maps isn't a crutch in the long run, but its a cheat.

We want Robots to cheat.

We want the entire comparison not to be fair.

Robots will still fail, but they'll fail in ways unique to them.

Having a human fail is rather laughable.
 
This is exactly the kind of example I'd use to argue that HD maps are absolutely necessary.

not necessarily HD maps. Just map data on where the lanes are and their function. But agree there are plenty of locations where due to lack of visibility, approaching the intersection like it's the first time ever will never improve FSD's behavior.
 
There's a section of highway near me where the car somewhat consistently brakes (not AEB). It happens around a particular overpass that was recently (a few years ago) redone. I've checked OSM and TomTom and everything seems up to date.

Last night, the car braked right after I went under the overpass. I noticed that the speed limit dropped from 65 to 50, which was the reason for the braking; the car was just trying to respect my set speed offset of +5mph, so it went from 70 to 55. But where is it getting this speed data? OSM and TomTom doesn't show any incorrect speed limits on the highway segments. Only thing I can think of is the car is pulling the speed limit of the overpass road. I don't actually know what the speed limit is. But over the years, I've heard many people suggest that the car can get confused under intersections regarding which road it's on. Seems like a really easy problem to fix using temporal data....

I don't think this is a mapping issue because sometimes the car doesn't slow at that spot.
 
  • Like
Reactions: FSDtester#1
I'm glad you used "on average" otherwise many of us would have taken offense;)

Ha, but no one should take offence because Robots have to drive much more safely. They won't be given allowances like Humans are allowed.

I don't know on average how often I make a mistake while driving, but its certainly isn't 0.

I'm sure if I suddenly found myself in your neck of the woods I'd make a few mistakes trying to navigate myself around Boston. I have heard its not exactly the most straight forwards of places to drive.