FSD Beta 10.69

[S100D]

I am having an issue with FSB Beta 10.69.25.1 on my neighborhood streets. While driving using FSD Beta without my home destination, instead of following the main street, my Tesla 2016 Model S will randomly turn on a right blinker and start turning into a side street. I am also having problems with multiple random phantom breaking with driving down the main street to my house.

If I turn off FSD Beta and use normal Auto Pilot, the car follows the main street. Is anyone else having this problem?

 

[Supcom]

I am having an issue with FSB Beta 10.69.25.1 on my neighborhood streets. While driving using FSD Beta without my home destination, instead of following the main street, my Tesla 2016 Model S will randomly turn on a right blinker and start turning into a side street. I am also having problems with multiple random phantom breaking with driving down the main street to my house.

If I turn off FSD Beta and use normal Auto Pilot, the car follows the main street. Is anyone else having this problem?

Nope. Though my car found itself in a right turn-only lane recently and could not get out of the lane due to traffic, so it turned right when it had to and rerouted to the destination. I was not happy that the car was too timid to merge left, but pleased that the car slowed and took the turn, as it should have under the circumstances.

 

[Twiglett]

I am having an issue with FSB Beta 10.69.25.1 on my neighborhood streets. While driving using FSD Beta without my home destination, instead of following the main street, my Tesla 2016 Model S will randomly turn on a right blinker and start turning into a side street. I am also having problems with multiple random phantom breaking with driving down the main street to my house.

If I turn off FSD Beta and use normal Auto Pilot, the car follows the main street. Is anyone else having this problem?

My car started doing this with this release. I retried the same road with the same nav as last time, it was showing it was going to turn left in 1/4 mile but then signaled right and started slowing down to make the right turn into a neighborhood, all the while with the nav showing the route going straight ahead.
Repeatable and useless

 

[S100D]

My car started doing this with this release. I retried the same road with the same nav as last time, it was showing it was going to turn left in 1/4 mile but then signaled right and started slowing down to make the right turn into a neighborhood, all the while with the nav showing the route going straight ahead.
Repeatable and useless

Hopefully Tesla will fix this in the next release.

One thing I forgot to say is that I was not following a car when it happened.

 

[archae86]

Exits. She seems to be missing exits. Everything is highlighted and sometimes even has a turn signal on but goes straight. Thought it was a fluke but happened 3 times today. Started after I updated to latest build. Sorry about the typo in the original post.

Any chance you don't have FSDb actually enabled for the trip? Specifically is this icon showing looking this way at the bottom of the trip stuff on the right two-thirds of the screen?

The way I had my autopilot section options configured early on, FSDb was only enabled for a trip if I tapped on the non-blue version of this icon to switch it to turned-on.

The puzzling thing was that on surface streets it would exercise complete control--stops, turns, the whole thing even the FSDb was not officially "on". Also on the freeway it would do lane changes, speed control--and the voice would stay I should turn off at an exit--but it would not turn off.

Now if you have your autopilot section options set to enable FSDb at the start of every trip, and your icon has the blue background, then this is not your problem, but otherwise, you can fix it.

 

[Twiglett]

Any chance you don't have FSDb actually enabled for the trip? Specifically is this icon showing looking this way at the bottom of the trip stuff on the right two-thirds of the screen?

View attachment 898616
The way I had my autopilot section options configured early on, FSDb was only enabled for a trip if I tapped on the non-blue version of this icon to switch it to turned-on.

The puzzling thing was that on surface streets it would exercise complete control--stops, turns, the whole thing even the FSDb was not officially "on". Also on the freeway it would do lane changes, speed control--and the voice would stay I should turn off at an exit--but it would not turn off.

Now if you have your autopilot section options set to enable FSDb at the start of every trip, and your icon has the blue background, then this is not your problem, but otherwise, you can fix it.

Thats the Navigate on Autopilot (NoA) icon which is engaged independently from FSDb as you discovered. It is also controlled by the NoA settings which are separate from the FSDb controls.
The behavior you describe is working as it is currently designed to do.

 

[archae86]

The behavior you describe is working as it is currently designed to do.

My point is that failure to engage could give the behavior described by pdubs. Do you disagree?

 

[Pdubs]

Any chance you don't have FSDb actually enabled for the trip? Specifically is this icon showing looking this way at the bottom of the trip stuff on the right two-thirds of the screen?

View attachment 898616
The way I had my autopilot section options configured early on, FSDb was only enabled for a trip if I tapped on the non-blue version of this icon to switch it to turned-on.

The puzzling thing was that on surface streets it would exercise complete control--stops, turns, the whole thing even the FSDb was not officially "on". Also on the freeway it would do lane changes, speed control--and the voice would stay I should turn off at an exit--but it would not turn off.

Now if you have your autopilot section options set to enable FSDb at the start of every trip, and your icon has the blue background, then this is not your problem, but otherwise, you can fix it.

Options are set for start of every trip and one push on the scroll wheel kicks it on. I’ve got 25,000 miles on this and it’s the first time it’s behaved like this. Appreciate the response.

 

[Goose66]

I am having an issue with FSB Beta 10.69.25.1 on my neighborhood streets. While driving using FSD Beta without my home destination, instead of following the main street, my Tesla 2016 Model S will randomly turn on a right blinker and start turning into a side street. I am also having problems with multiple random phantom breaking with driving down the main street to my house.

If I turn off FSD Beta and use normal Auto Pilot, the car follows the main street. Is anyone else having this problem?

I know phantom braking has been a long running theme here, but the new lane selection neural networks they put in 3 or 4 versions ago has really had a substantial impact on the usability of FSDb for me. I remember how irate I was during 2022 AI Day when they were so proud the new NNs and talking about how they were based on natural language processing NN patterns and reduced reliance on mapping data - had they ever actually driven a car with FSDb and seen how terrible the new NNs performed I wondered? I think these guys see everything in the abstract and through simulation but have no practical, real-world experience. Everyone doesn't live in Silicon Valley with pristinely marked streets and three-mile commutes. Tesla FSDb and AutoPilot both performed for me significantly better on the map data before these new NNs were introduced. Now my car consistently makes wrong (and often non-sensical) lane selections and then moves aggressively to execute them despite solid-white lines (which evidently FSDb engineers don't understand you aren't supposed to cross) and turning lane only indicators. This has been perhaps the largest regression in FSDb I've seen in my 16 months of testing.

 

[nvx1977]

I know phantom braking has been a long running theme here, but the new lane selection neural networks they put in 3 or 4 versions ago has really had a substantial impact on the usability of FSDb for me. I remember how irate I was during 2022 AI Day when they were so proud the new NNs and talking about how they were based on natural language processing NN patterns and reduced reliance on mapping data - had they ever actually driven a car with FSDb and seen how terrible the new NNs performed I wondered? I think these guys see everything in the abstract and through simulation but have no practical, real-world experience. Everyone doesn't live in Silicon Valley with pristinely marked streets and three-mile commutes. Tesla FSDb and AutoPilot both performed for me significantly better on the map data before these new NNs were introduced. Now my car consistently makes wrong (and often non-sensical) lane selections and then moves aggressively to execute them despite solid-white lines (which evidently FSDb engineers don't understand you aren't supposed to cross) and turning lane only indicators. This has been perhaps the largest regression in FSDb I've seen in my 16 months of testing.

I would concur that the lane-choice idiocy is the most frustrating thing about FSDb atm. Introduced during 10.10 as part of vector lane lines, I had high hopes it would have been polished up by now. Instead, zero improvement at all. They must be aware... there's something fundamentally wrong with the approach.

 

[AlanSubie4Life]

I would concur that the lane-choice idiocy is the most frustrating thing about FSDb atm. Introduced during 10.10 as part of vector lane lines, I had high hopes it would have been polished up by now. Instead, zero improvement at all. They must be aware... there's something fundamentally wrong with the approach.

It’s ok, Elon will review V11 and then we will be good to go.

He has been laser focused on it for the last several months so the polish should be high.

 

[kabin]

I know phantom braking has been a long running theme here, but the new lane selection neural networks they put in 3 or 4 versions ago has really had a substantial impact on the usability of FSDb for me. I remember how irate I was during 2022 AI Day when they were so proud the new NNs and talking about how they were based on natural language processing NN patterns and reduced reliance on mapping data - had they ever actually driven a car with FSDb and seen how terrible the new NNs performed I wondered? I think these guys see everything in the abstract and through simulation but have no practical, real-world experience. Everyone doesn't live in Silicon Valley with pristinely marked streets and three-mile commutes. Tesla FSDb and AutoPilot both performed for me significantly better on the map data before these new NNs were introduced. Now my car consistently makes wrong (and often non-sensical) lane selections and then moves aggressively to execute them despite solid-white lines (which evidently FSDb engineers don't understand you aren't supposed to cross) and turning lane only indicators. This has been perhaps the largest regression in FSDb I've seen in my 16 months of testing.

Crossing thick solid white lines is frustrating, unsafe, and a potential ticket. And in my experience it crosses them blatantly close to intersections with an F-U-like disregard.

And turns signals are right up there. Like everything else FSDb, path processing is probably pruned in an attempt to improve decision making response time. The other day I was approaching a slower moving vehicle in the right lane but FSDb initiated the right turn signal as if turn signal logic was disconnected with the location of vehicles sharing the roadway. Of course there was plenty of time to initiate the turn signal and complete the lane change after passing the slower vehicle but FSDb isn't designed to play well with others - it's all about FSDb. Some roadways I frequent I know exactly when the turn signal will be initiated and the lane change is attempted. Lane change logic is woefully simplistic and does not represent safe driving.

 

[FSDtester#1]

Most of the above still same for me.
Seems like less phantom blinkers... anyone else?

 

[aronth5]

Crossing thick solid white lines is frustrating, unsafe, and a potential ticket. And in my experience it crosses them blatantly close to intersections with an F-U-like disregard.

And turns signals are right up there. Like everything else FSDb, path processing is probably pruned in an attempt to improve decision making response time. The other day I was approaching a slower moving vehicle in the right lane but FSDb initiated the right turn signal as if turn signal logic was disconnected with the location of vehicles sharing the roadway. Of course there was plenty of time to initiate the turn signal and complete the lane change after passing the slower vehicle but FSDb isn't designed to play well with others - it's all about FSDb. Some roadways I frequent I know exactly when the turn signal will be initiated and the lane change is attempted. Lane change logic is woefully simplistic and does not represent safe driving.

I have the opposite problem when exiting highways when a thick solid line is ignored by everyone else (entering and exiting) except FSD. FSD waits until the thick line is gone before moving to the exit lane which other drivers get very annoyed at. This also happens when entering highways. FSD waits but most other drivers just cross over the thick line. So much for following traffic rules!

 

[MP3Mike]

Crossing thick solid white lines is frustrating, unsafe, and a potential ticket.

I think that varies by state. For example, in Oregon it is permissible to cross wide solid white lines:

 

[Twiglett]

My point is that failure to engage could give the behavior described by pdubs. Do you disagree?

not really. My point was that the FSDb settings have no effect on NoA settings.
So starting a drive then enabling FSDb will not automatically enable NoA when on highways. That is controlled by the "Enable at start of every trip" in the NoA settings and is independent of FSDb.
Currently FSDb only applies to "city streets" so when you get onto the highway the car switches to AP and optionally NoA
Sorry, it's a common misunderstanding that even folks like MKBHD get confused with, so what chance do we have
Hopefully the unified stack on V11 might fix it (or not)

 

[sleepydoc]

Most of the above still same for me.
Seems like less phantom blinkers... anyone else?

I've seen a big increase in phantom blinkers - to the point that I frequently don't use FSD because it's so confusing to other drivers around me. Currently (v2022.44.30.5) FSDb turns on a blinker every time there's a turn lane, whether it takes the turn or not.

 

[EVNow]

I have the opposite problem when exiting highways when a thick solid line is ignored by everyone else (entering and exiting) except FSD.

Apart from speed limit, this has to be rule most often ignored. Infact I've hardly seen anyone strictly follow it.

 

[YNoVA]

My subscription ended yesterday but my car kept the updated FSD visualization. I assume this is because I'm still on the FSD software.

This makes me believe that when the merge happens with V11, the 'FSD visualization preview' should be updated to the latest visualization. If you've ever used the current, non-FSD visualization, you'd know how bad it is compared to the current FSD one.

 

[Supcom]

Crossing thick solid white lines is frustrating, unsafe, and a potential ticket. And in my experience it crosses them blatantly close to intersections with an F-U-like disregard.

Crossing a single white line is discouraged, but legal. Crossing a double white line is not legal. Changing lanes close to, and even within an intersection is also legal.