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.
I find it rarely is touching the brakes, never like phantom braking unless it is really hitting the brakes are going heavy into regen. Mostly I find it is just lifting off the accelerator, and regen is kicking in and it feels like braking in a normal car since it won't coast so to speak.

Have had some braking. Still haven't got 10.6 yet...so...yeah. Will see. This morning, 10.5.1 was a mess on the two lane road I sometimes take, the entire way was constantly doing it where I just turned it off. Unnerving and you have to keep your hands on the wheel ready to take over because you have no idea what it is about to do. I do this anyway. But I mean, I have both hands on the wheel like I'm driving...that isn't progress.
 
  • Like
Reactions: ModLTh3
Nobody has it. Check his videos again - he doesn’t have it either.

But, you can cancel a lane change as soon as it turns the signal on. Or disengage and send an error report.
you can try. A couple weeks ago I was sitting in traffic just beyond a split where the road I was on splits off for a bypass. The MY probably could have turned without hopping the curb but was clearly illegal at that point with the paint markings. The split was probably about 50 yards to my rear.

The car kept turning on the turn signal. I kept turning it off and it would go on again. Yeah disengaged as I already looked the fool.

Then as we crept forward beyond where it could not illegally split and I re-engaged, The car decided to signal a left turn. Hopefully at the light that I'm about 8 cars back from. Again I'd turn it off and the car would turn it back on. All the time, nav stayed the same indicating I should go straight for around 2 more miles.
 
Teslafi.com show the following percentages for 10.5 and 10.6. Roughly a quarter of FSD beta cars are at 10.6 but the percentages haven't moved in the past 24 hrs.
1638804401586.png
 
View attachment 741061 To me this is confirmation that FSD is slowing down for the bends
I wouldn't be surprised if this was a general FSD Beta behavior to slow down when the destination is not clear, e.g., if visualization cannot show something clear 3 seconds away, slow down. This second picture at 31mph does have some plants towards the left edge blocking a little bit of what seems to be a slightly downward sloping bend. I would guess sending video snapshots *after* completing the bend will help future neural networks predict further ahead with more confidence.
 
Teslafi.com show the following percentages for 10.5 and 10.6. Roughly a quarter of FSD beta cars are at 10.6 but the percentages haven't moved in the past 24 hrs.
View attachment 741176
There has been complaints from several prominent beta testers who have 2 strikes or actually locked out even though their videos clearly showed they were paying attention. Perhaps Tesla is addressing this and paused the roll out?
 
There has been complaints from several prominent beta testers who have 2 strikes or actually locked out even though their videos clearly showed they were paying attention. Perhaps Tesla is addressing this and paused the roll out?

Ya, but they act like there's nothing they can do to prevent the lockout. Like in Rob's video, he's constantly looking down at the screen and all over place, along with wearing a hat.

The DMS isn't perfect, but they (the drivers) can do better to prevent a lockout.
 
I wouldn't be surprised if this was a general FSD Beta behavior to slow down when the destination is not clear, e.g., if visualization cannot show something clear 3 seconds away, slow down.
I still feel it is a planned move to reduce speed before a bend. Unlike AP, FSD is map-aware and can reduce speed based on mapping information of bends coming up.

For eg. when cresting - so the car can't see much ahead, it won't reduce the speed. Same for going uphill - on the same road, no reduction in speed even though same bends and visibility issues exist.

I'm going to try a few different hills to confirm this - but I think a lot of people who have been reporting "phantom braking" on twisty roads are actually experiencing planning slow-downs. This is actually good news since its obviously much easier to solve for Tesla.
 
Yup. FSDbeta would slow on some turns where it couldn't, nor could I, see around the bend. It looks like a similar condition here.

I retested this with the same 10.5 - but this time in daytime. I got the exact same slowdowns from 40 mph to 30 mph. Two times on 2 bends.

View attachment 741060 View attachment 741061

Then, I returned to the same road minutes later - but this time with just AP using AP profile. As you can see here - no slowdowns at all.

To me this is confirmation that FSD is slowing down for the bends and this is not phantom braking as I had earlier assumed. Those getting such slowdowns on twisty roads should reconsider whether those are phantom braking or just FSD slowing down for the bends.

View attachment 741062 View attachment 741063
 
FSDbeta would slow on some turns where it couldn't, nor could I, see around the bend
To be clear, FSD Beta drives on what it believes (predicts) is there and not actually what is visible. Given enough training data, it can make pretty reasonable predictions of what is the likely road layout even when occluded by vehicles, plants or even buildings:
predict right.jpg

The predicted path here extends well past what one could see, so this should mean other cases like a bending road with elevation change can be learned by the current neural network structure. That's good until it actually gets overfit believing "this is the only way the road could bend in this situation" and should actually slow down instead.
 
  • Like
Reactions: FSDtester#1
Here's the report of the car going into reverse.
Yep, that’s me. It did it unnecessarily, but it was reacting to a car in crossing traffic while creeping. While there was a non-zero possibility that I could have hallucinated but it was unlikely. First, the car was already stopped for a while before creeping backward, and my reference object is the parked big ass truck that’s not moving at all, so not a typical misperception situation. Second, the car was doing it slowly on a slope at the parking exit so I felt the bump, and even have time to take a look that it was in Drive. Still a non-zero possibility, but consider others reported in the similar scenario, it is likely a built-in logic. However the judgement of the car was still off.

My guess is the built-in logic is to do this when the car is way into the lane and there is crossing traffic. In Chuck’s case, car didn’t consider itself in the lane so it never did reverse. But in my case, car was indeed deep in the lane, but due to the big truck parked on the left, it actually didn’t creep enough and did not need to reverse. Very special scenario so rarely reported.
 
Yep, that’s me. It did it unnecessarily, but it was reacting to a car in crossing traffic while creeping. While there was a non-zero possibility that I could have hallucinated but it was unlikely. First, the car was already stopped for a while before creeping backward, and my reference object is the parked big ass truck that’s not moving at all, so not a typical misperception situation. Second, the car was doing it slowly on a slope at the parking exit so I felt the bump, and even have time to take a look that it was in Drive. Still a non-zero possibility, but consider others reported in the similar scenario, it is likely a built-in logic. However the judgement of the car was still off.

My guess is the built-in logic is to do this when the car is way into the lane and there is crossing traffic. In Chuck’s case, car didn’t consider itself in the lane so it never did reverse. But in my case, car was indeed deep in the lane, but due to the big truck parked on the left, it actually didn’t creep enough and did not need to reverse. Very special scenario so rarely reported.
I have not heard of anyone else reporting this. Plus, this was once tweeted as a future feature but was never shown in the release notes. You mention you were on a slope. Sometimes the car will roll back on a slope of the brakes are not held tightly enough. That appears to be the most likely explanation.