FSD Beta 10.69

[MrTemple]

Well, twitter is rife with fake accounts, so some of those people might not exist. But, it doesn't matter. Cherry-picking anecdotal reports from individuals saying things like, "It's 10x better than the previous version" would seem to by just slightly hyperbole. And, the guy you quoted lauding over 10.69.2.1 could not have tested it very long at all before posting his praise. Not likely to have been a very thorough test. Or, maybe he doesn't really know what software he has?

In any event, a bunch of twitter posts from random people is worthless. Absolutely worthless.

I’m sorry, BS. This is navel gazing garbage to deflect from actual real truth.

The argument is that nobody uses FSD more now that it’s better.

And we can find no shortage of reports by people saying they DO use FSD more now that it’s better, everywhere we look including:

- in this thread
- all over twitter
- facebook enthusiast groups (big ones and local ones)
- youtube

But ALL these reports, ALL these people saying a frankly extremely easy to believe statement (they use it more because it’s better), ALL of them fake?

Fake? Every one?

Discounted as any sort of evidence such a strange person that would use a better thing more often even exists.

Because they could be bots suddenly changing their tune about FSD’s quality and comfort and useability.

And so therefore we have to keep the theory that they DON’T exist alive and on equal footing that they do?

I’m sorry, no. That is a garbage take. Talk about your no true Scotsman.

That is pure delusion.

 

[DarkForest]

After driving with 10.69.2 since Monday, my thoughts:

The Good

• Much smoother drives
• Able to draw conclusions based on partial information

The Bad

• Whimsical hesitations
• Late and abrupt stopping at stop signs

The Ugly

• Roundabouts
• Made me a shill

The worst driving I’ve seen FSD do was at a two-lane roundabout where each lane was pretty wide. I’m sure people thought I was drunk until I disengaged FSD.

 

[aronth5]

A couple more examples of sudden brake-filled stops with FSD Beta 10.69.2.

I would estimate 25% of stops are like this (which means they are jerky with pulsing regen, not necessarily using brakes) on the flat, perhaps 10-20% on uphills, and about 50% (or more?) on downhills.

Uphills definitely seem to be easier for it. It’s not the only reason for it, but it is one of the reasons. Not sure why they can’t handle it.

Downhill:

Flat (note I had released the accelerator well before any slowing was needed, though unfortunately I don’t verbally mark that in the video):

This constant anxiety over regen and braking has gotten bit a much so I'm going to pass on these posts for awhile.

 

[AlanSubie4Life]

This constant anxiety over regen and braking has gotten bit a much so I'm going to pass on these posts for awhile.

But I have new material! Sadly I have none of the wide angle camera equipment, or HDMI plug-in, which is a must for properly capturing the scene.

@jabloomf1230 ‘s suggestion to go at a quiet time is good, but noon on a Saturday was not it! Need to go very early on a Sunday I think.

Anyway this is just a reference point. Now that I have done this myself a few times (I did the turn three times myself; this was the trickiest maneuver that I completed), I think I don’t trust FSD to do it. I will have it do it a few times with zero traffic, as suggested, before having it attempt anything more complicated.

So what happened here is the close lane was clear when I went (this is why I typically like TWO lanes to be clear!), but the BMW changed lanes into my target lane as I went (you can see what happened on the visualization!). I saw this when I looked over my shoulder and made eye contact. Was a bit surprised to see my lane occupied, so I waited for him, then went. It was exciting! Obviously my bailout is to go into the double yellow (which actually my wheel did, I think - would not have happened if the BMW had not changed lanes). Anyway no honking from the BMW, because he had just changed lanes into my target lane, and we saw each other. He just accelerated to help me get in. And check out the regen bar - I didn’t even go into regen to get the timing right! Pulse 1: 17mph, saw BMW, eased to allow timing. Pulse 2: gradual speed adjustment while waiting for BMW to 19mph. Pulse 3: Get up to speed.

Hopefully FSD can thread into traffic similarly, relatively soon.

Here’s a picture of roughly what the car sees when it has advanced to nearly the furthest forward creep limit (I’ll get more pictures of this later to see how it looks at the TRULY very farthest forward position - to me it looks like I could go forward another couple feet, but I will pay closer attention next time - there was a lot going on). Held the camera just in front of the b-pillar, probably an inch or two lower than the camera. Note the driveway is right there too (what I was babbling about in the video).

I sure hope that occlusion network works really well and has some memory!

I wouldn’t worry about the disengagements here. I never planned to let it go, unless there was a huge opening both ways. Need to see behavior in zero traffic as suggested first.

I saw a lot of human drivers giving up on the left turn. Didn’t like what they saw! (And obviously it is pretty easy to go around the block or do a u-turn; there is really very little reason to turn here if it is busy.)

Hopefully I can show a 10.69.2 success soon. Maybe 10.69.3 can solve this one!

I also need to remember to make the visualization larger!

 

[Supcom]

I’m sorry, BS. This is navel gazing garbage to deflect from actual real truth.

The argument is that nobody uses FSD more now that it’s better.

And we can find no shortage of reports by people saying they DO use FSD more now that it’s better, everywhere we look including:

- in this thread
- all over twitter
- facebook enthusiast groups (big ones and local ones)
- youtube

But ALL these reports, ALL these people saying a frankly extremely easy to believe statement (they use it more because it’s better), ALL of them fake?

Fake? Every one?

Discounted as any sort of evidence such a strange person that would use a better thing more often even exists.

Because they could be bots suddenly changing their tune about FSD’s quality and comfort and useability.

And so therefore we have to keep the theory that they DON’T exist alive and on equal footing that they do?

I’m sorry, no. That is a garbage take. Talk about your no true Scotsman.

That is pure delusion.

You apparently do not know the difference between anecdote and data. I'll just leave it there.

Good luck to you.

 

[Supcom]

But I have new material! Sadly I have none of the wide angle camera equipment, or HDMI plug-in, which is a must for properly capturing the scene.

@jabloomf1230 ‘s suggestion to go at a quiet time is good, but noon on a Saturday was not it! Need to go very early on a Sunday I think.

Anyway this is just a reference point. Now that I have done this myself a few times (I did the turn three times myself; this was the trickiest maneuver that I completed), I think I don’t trust FSD to do it. I will have it do it a few times with zero traffic, as suggested, before having it attempt anything more complicated.

So what happened here is the close lane was clear when I went (this is why I typically like TWO lanes to be clear!), but the BMW changed lanes into my target lane as I went (you can see what happened on the visualization!). I saw this when I looked over my shoulder and made eye contact. Was a bit surprised to see my lane occupied, so I waited for him, then went. It was exciting! Obviously my bailout is to go into the double yellow (which actually my wheel did, I think - would not have happened if the BMW had not changed lanes). Anyway no honking from the BMW, because he had just changed lanes into my target lane, and we saw each other. He just accelerated to help me get in. And check out the regen bar - I didn’t even go into regen to get the timing right! Pulse 1: 17mph, saw BMW, eased to allow timing. Pulse 2: gradual speed adjustment while waiting for BMW to 19mph. Pulse 3: Get up to speed.

Hopefully FSD can thread into traffic similarly, relatively soon.

Here’s a picture of roughly what the car sees when it has advanced to nearly the furthest forward creep limit (I’ll get more pictures of this later to see how it looks at the TRULY very farthest forward position - to me it looks like I could go forward another couple feet, but I will pay closer attention next time - there was a lot going on). Held the camera just in front of the b-pillar, probably an inch or two lower than the camera. Note the driveway is right there too (what I was babbling about in the video).
View attachment 853834

I sure hope that occlusion network works really well and has some memory!

I wouldn’t worry about the disengagements here. I never planned to let it go, unless there was a huge opening both ways. Need to see behavior in zero traffic as suggested first.

I saw a lot of human drivers giving up on the left turn. Didn’t like what they saw! (And obviously it is pretty easy to go around the block or do a u-turn; there is really very little reason to turn here if it is busy.)

Hopefully I can show a 10.69.2 success soon. Maybe 10.69.3 can solve this one!

I also need to remember to make the visualization larger!

The blockage caused by the dealer sign probably helps generate walk up sales.

 

[momo3605]

Anyone wanna speculate what happened to 10.69.2.1? Seems it was rolled to employees 2 days ago. Usually the update would’ve gone out to YT crew 1/2 day later.

 

[Supcom]

Took 10.69.2 out for some urban driving Sat afternoon. Picked points at random for it to drive to, keeping away from interstates so that the drive would be all FSD beta. The car did well with only one disconnect over 20 miles. Nothing very difficult, basically simulating utility driving. Intersections were generally all controlled and driving was on six-lane artery roads for the most part.

I did need to use the accelerator a few times when the car would not proceed on a right turn on red despite the way being quite clear. Beta is inconsistent here and sometimes makes the turn correctly and other times needs to be kick in the rear to make the turn.

My one disconnect was a forward UPL at a flashing red. The car nosed out into the intersection and got caught there when the light turned solid red. I expected it to clear the intersection, but after a few seconds, it was apparent that it had no intention of doing anything but block cross traffic, so, I put the car in reverse and backed out of the cross lane. Another inconsistency where sometimes beta does the right thing, other times it wont.

Beta is still quite annoying with how it deals with a lead car turning off the road at mid-block. The car brakes hard nearly to a stop every time. It has a delayed response so it is still slowing even after the turning car has cleared the street. I thought that this was improved with 10.69.2, but after today, I have reversed that and believe it is no better than previous versions. Making matters worse is that 10.69.2 seems to prefer either the left-most or right-most lane when there are three lanes to use. If it preferred the center lane, there would be far fewer cases of slowing for turning cars.

 

[Supcom]

Anyone wanna speculate what happened to 10.69.2.1? Seems it was rolled to employees 2 days ago. Usually the update would’ve gone out to YT crew 1/2 day later.

Maybe we'll see 10.69.2.1.1 soon.

 

[thewishmaster]

I thought surely all the recent work on lane continuity would be worth something and this issue would be fixed since a few months back when I last used FSD beta through this area. Car was in the right lane, nav was correct, all it needed to do was complete the curve; instead it angled for the median that splits the main road from the right turn only lane:

My artwork is not very precise, and the actual angle of the red arrow should be a little bit to the left. Not sure if it was doing its famous “centering itself in the widened lane” behavior or that was just a coincidence.

 

[Reddy Kilowatt]

Ran my typical 120 mile route yesterday, and FSD did great. It did the large traffic circle ok,and the U turn exit with speed adjustment. Most was early AM running in the fog, with lite traffic. Had one disengagement when car tried to turn right to early at a driveway instead of the street. Probably had 5-6 assists.

it really cleaned disengagement issues from previous version. Mainly staying on the road and not jumping into turn lanes. Very impressive run for me.
new issues. Sat through a green left turn light ,taped gas to go.

Where dual rated speed signs are used it is now picking slower speed every time (truck speed). Constantly increasing speed. Still have issue right at my house with simple right turn. Very balky,sometimes stops thinking the is a stop sign.

it did good on my highway jug handle left turn. The car has to make a right get to highway speed the slow down make a wide U turn at cross street then take off. I slowed the car down to 30 going in. It made the U- turn, slightly crossing the white line, then took off down the merge lane then merged into traffic.

still having to change lanes to make exits or upcoming turns. Still having issue with car jumping into left lane just before turning right.

 

[sleepydoc]

Got my first time out last evening driving in the evening. Even tho I was holding the yoke it kept on beeping at me and got my first of the five timeouts. It was not at all my fault as the yoke will get seizure and make all the crazy movements. I reported each time when it made a mistake. It was the same drive I had done numerous times. I disengaged multiple times in my return as it was not worth it. The time out occurred when I drove hardly a mile.

That stinks. I had hoped they fixed the bug(s) that were causing this but apparently not.

 

[KArnold]

Not sure if it was doing its famous “centering itself in the widened lane” behavior

That's what it looks like to me as well. Did you intervene? If not I suspect the car would have abruptly self-corrected back to the correct lane. The "wandering and hunting" in widened lanes due to right turn lanes is a different problem. Seems it would be easy enough to hug the middle lane marker if/when the right lane marker disappears. One of these days.

Perhaps complicated by a lead car taking the right branch.

 

[MrTemple]

You apparently do not know the difference between anecdote and data. I'll just leave it there.

Good luck to you.

Did somebody once scold you about the difference between anecdotes and data, and now you’re knee-jerking your trauma?

Or are you just trying to grasp at any logical fallacy (google the no true Scotsman) because you can’t be wrong?

When the theory is: “People with this opinion don’t exist.”

Tons of people saying, “I have that opinion.” is absolutely evidence that they do indeed exist.

This seems so silly that it has to even be explained to you.

 

[queue]

My feedback on 10.69.2:

1) It is far too timid entering and leaving intersections. I end up intervening in some way, either disengage or press the accelerator probably 50% of the time. The more vehicles in an intersection, the greater the chance of an intervention. It is almost always attempting to move or not, out of order.

2) Because FSD tends to drive on "timid" side, the constant unnecessary lane changes are also imho, unsafe and also provoke either a intervention or disengage. Lane changes it seems to me, should almost always default to a slower lane, unless there is an upcoming left hand turn. At least some mechanism to disable the unnecessary lane changes that simply shorten your trip by a few seconds. In this particular case, I don't want FSD to "drive like a human" weaving in and out of traffic. I don't care at all if my lane is 5 MPH slower than the lane next to me.

I've been driving for 45 years, I've NEVER been in an accident, I always select the lane I'll be using to turn as I enter a road, and I don't speed, and I don't weave in and out of traffic. Improvement on these 2 issues alone would make a dramatic difference in FSD for me.

 

[bhakan]

Got my first time out last evening driving in the evening. Even tho I was holding the yoke it kept on beeping at me and got my first of the five timeouts. It was not at all my fault as the yoke will get seizure and make all the crazy movements. I reported each time when it made a mistake. It was the same drive I had done numerous times. I disengaged multiple times in my return as it was not worth it. The time out occurred when I drove hardly a mile.

@AlanSubie4Life I posted right after I got 69.2 was installed and my test drive ( post #2110). I did not have this issue. This behavior i observed and my timeout just happened two days ago and it was during the evening hours and lot if traffic around me. I have used EAP when I first had mt MS 75 in 2017 and then AP on local roads and highways quiet a bit in MS 2018 and know fully well about holding the wheel with pressure. I have extensively used in my current refresh 2022 MS too. However, now it was totally different and there is a bug which they should fix. The car did not register that I was holding the yoke wheel when it had violent seizure and moved the wheel with jerk while I was still holding. I had disengage many times because of that. I don't understand your disagreement with that assessment.

 

[thewishmaster]

That's what it looks like to me as well. Did you intervene? If not I suspect the car would have abruptly self-corrected back to the correct lane. The "wandering and hunting" in widened lanes due to right turn lanes is a different problem. Seems it would be easy enough to hug the middle lane marker if/when the right lane marker disappears. One of these days.

Perhaps complicated by a lead car taking the right branch.

Yeah, full disengagement as I had someone else in my car. 35mph street… not a lot of margin for swerving back and forth

 

[bhakan]

That stinks. I had hoped they fixed the bug(s) that were causing this but apparently not.

I hope so too as this has never happened before and I have been driving with some kind of AP/EAP since 2017 and I have driven in the evenings many time. With this 69.2 and in the evening either it did not notice my eye movement or whatever it does as it kept on telling me to pay attention to the road when I was. They better fix this as I am disengaging so that I won't be in a FSDb jail.

 

[AlanSubie4Life]

When the theory is: “People with this opinion don’t exist.”

That’s not the theory.

The theory is not even: “People are not driving FSD Beta more now that it has improved.”

I would not make either of those claims. I think the amount of driving being done is about the same, probably slightly increased (to make it definite, I would guess 10-20% increase, and as a percentage of overall surface street driving, maybe 5-10% (this is tougher to guess because depends on the typical driving being done in the population)).

But really I don’t have strong priors here. I would just like to see what the data say. Seems roughly stable so far, with the guesstimates we have seen posted so far.

@AlanSubie4Life I posted right after I got 69.2 was installed and my test drive ( post #2110). I did not have this issue. This behavior i observed and my timeout just happened two days ago and it was during the evening hours and lot if traffic around me. I have used EAP when I first had mt MS 75 in 2017 and then AP on local roads and highways quiet a bit in MS 2018 and know fully well about holding the wheel with pressure. I have extensively used in my current refresh 2022 MS too. However, now it was totally different and there is a bug which they should fix. The car did not register that I was holding the yoke wheel when it had violent seizure and moved the wheel with jerk while I was still holding. I had disengage many times because of that. I don't understand your disagreement with that assessment.

I’m sorry to hear you got a strike.

My issue with it was that no information was provided. Was it a wheel torque nag? Was it an attention nag? How exactly did it progress? (The exact sequence was completely unclear from your post.)

If you get nagged for no apparent reason, you just stop using FSD, before getting a strike. That way if you have a hardware issue you don’t get a strike.


I definitely think you should take video going forward, so that if you get another strike, you can post it here, so we can figure out what the heck is going on. It would be unique YouTube content - no one has yet posted a video of a false strike.

 

[sleepydoc]

The argument is that nobody uses FSD more now that it’s better

you're making a logical error here.

The car nosed out into the intersection and got caught there when the light turned solid red. I expected it to clear the intersection, but after a few seconds, it was apparent that it had no intention of doing anything but block cross traffic

Yes - I've had this problem before where it doesn't seem to get the concept of completing the turn once it's stuck in the intersection after the light turns red (something that humans routinely take advantage of all the time)

It has a delayed response so it is still slowing even after the turning car has cleared the street

Agreed. In past versions FSD would do something similar when an approaching car made a left turn in front of it. It wouldn't take the car's movement and trajectory into account and treat it like an obstacle in the road and continue braking until it was completely clear. Kind of like there were 3 states - Car in opposite lane, Car in front of me & Car on side street. In one of the previous updates it finally started to take the car's movement into account and added a 4th state: "car in front of me but moving forward so I don't need to worry." It seems like FSD needs to add a similar analysis to cars in your lane turning right, too.