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

FSD Beta 10.69

This site may earn commission on affiliate links.
Since V11 will likely be the wild step child for a stretch, TSLA might need to update the 10.69 path before a wide V11 release. The big question is whether TSLA was aware of these issues, if they can isolate them, and ultimately train them appropriately without adding more regressions given some of the current NNs are at or near max capacity (saturation).
TACC already properly Slows down when there is a lower speed limit, while FSD does not.
should be an easy fix

 
Assuming it's still single-stack, anyone care to speculate as the winners/losers?

It's probably likely the highway side of things can likely maintain speed/distance/lane changes but degradation on decisions like when/where to change lanes. Plus I'd expect some Nav issues like routing, wrong exits, etc.

In the city I'm expecting baby steps at this point. Just incremental progress on lane choice and other bugs. But I would expect the turn signal thing mostly to be fixed. We might be complaining about mostly the same issues. 🙂

FWIW, which 🚫 isn't much.
 
On the subject of existing FSDb problems:
Michigan Left-Turn Confusion
We now have a few major intersections with a so-called "Michigan left turn" setup. This means there is no direct left turn lane; to execute a left turn you must proceed straight through the light, crossing the intersection completely, and then get in a special signal-controlled U-turn lan. From there you can get back to the intersection and then execute a standard right turn. I think the main purpose of this "Michigan left" is to eliminate the the left-turn-arrow portion of the light timing cycle, and therefore speed up traffic flow overall. It does seem to help in that regard

Here's the problem FSDb is having: at the minor U-turn-only traffic signal, only a few hundred feet past the main intersection, the traffic signal is a median-placed green arrow / red arrow light that is only there to control the U-turn lane. The main (through) lanes have no reason to stop there and so there is no traffic light control.

I've been going through one such intersection every day since I bought the car. There was never a problem at that place on FSD-Autopilot with Traffic Light Control. However, FSD beta thinks there's a light there (possibly from map data? Or possibly because it sees the red arrow light at the median?), and it wants to slow to a stop if left alone.

It's no problem to override this with the accelerator, I just have to keep it down until the car gets about 100 ft past the trouble spot - but of course it's annoying and potentially dangerous unexpected behavior; cars behind me have no reason to think I might stop there.

I should note that while FSDb definitely introduced this particular problem, it also solved several other consistent AP phantom slowdowns and crossing-intersection "zig-zag" glitches along the same road.

It's been suggested in the California tunnel slowdown accident, and in other phantom slowing reports, that there might be map confusion relating to a traffic signal on an upper or lower road. If this is anything similar, it might be map confusion due to a traffic signal on the current road, but not pertaining to through lanes.

I'd be interested to hear if anyone else knows of a similar situation where there are turn arrow lights but no stoplight for the through lanes? How does FSDb behave at that location?
 
Assuming it's still single-stack, anyone care to speculate as the winners/losers?

It's probably likely the highway side of things can likely maintain speed/distance/lane changes but degradation on decisions like when/where to change lanes. Plus I'd expect some Nav issues like routing, wrong exits, etc.

In the city I'm expecting baby steps at this point. Just incremental progress on lane choice and other bugs. But I would expect the turn signal thing mostly to be fixed. We might be complaining about mostly the same issues. 🙂

FWIW, which 🚫 isn't much.
I'm not sure, but perhaps this belongs in the v11 anticipation thread. In any case, I think it would be right to expect some regressions but hopefully also some improvements on Highway NoA. Though not a perfect feature, that is well-established and familiar, many users here say it's the most reliably functional part of FSD now.

Considering that, my suggestion would be that Tesla should make it a toggle selection in the v11 beta. Almost any deviation from prior NoA behavior is going to annoy some people, and then you'll see a flood of messages where people say Elon broke the last working part of FSD.

I drive very little on the freeway, so my attention will be on the improvements or regressions of v11 in the City Streets code. Despite the single-stack highway emphasis, there are supposed to be many significant changes in v11 with new NNs etc., so I think it's relevant to discuss the non-highway stuff also. I don't really have a prediction, but my hope is that it will largely fix the following:
  • Unwanted lane changes and those really annoying on again off again lane change blinker threats. Maybe not the most dangerous behavior of lFSDb, but probably the most antisocial and hard to anticipate/override.
  • Right and left turn commitment behavior. The car should not lurch as strongly during creep-for-visibility maneuvers. This happens even at turns with completely an obstructed visibility, and sends confusing signals to me and to oncoming drivers. Then, when the car decides to actually enter the turn, it often moves way too slowly and converts an otherwise safe turn into a braking event for an approaching car. Summary: don't give turn commit indications if you're just creeping, but then commit confidently to the turn if you're entering the lane.
  • Stop sign Behavior: if the stop sign is located well behind the crossing road edge, please don't Stop, Creep, Stop again. Truthfully I'm confused by the NHTSA position on stop sign behavior, so I'm not sure Tesla is completely free to program this properly. Right now it's not very good if anyone is queued behind you.
  • There are of course many other opportunities for improvement, the above are the things that would help me the most.
    • Other than ULT behavior, which is supposedly much better than before but still tends to scare the heck out of me, usually leading to disengagement and manual completion. Some of this is wrapped up with the commitment issue mentioned above. But this is one of the harder things, so I put things in the list above that I think are easier.
 
Last edited:
An EV youtuber asked FSDb owners if they'd accept a refund if offered. After about 900 votes, 69% said yes. It's a good sample size but no way of knowing if any voters really purchased FSDb. As things stand now with HW3 I'd probably vote yes too.
It would be interesting to ask that SAME group, would you keep the FSD that you purchased IF Tesla allowed you to roll it onto a future purchase, or say 50% onto a future purchase… now THAT would be an interesting survey. I think it would be more than %69
 
Today I discovered what happens if you just let fsdb drive with the GPS being off by a fraction of a block. Made a right turn a whole block early because it got confused 😆 not really surprising, but it’s yet another vector of unpredictability. I didn’t notice the GPS being offset until that happened.
Yup, I have a route to a Quick Trip that the GPS is off on location and it gets really confused as it turns early. I have had to really good night drives with it recently, though. A few odd actions but I didn't need to intervene on 8 mile round trip travels. Honestly, if this had been built with the goal of being a travel assist on local roads I think they could have developed something really useful -- it just needs easier ways to correct it's bad behavior without shutting it down completely (maybe). It's really neat on a long road without a turn and lots of stop lights. I

Overall, I've started to appreciate it more after initially thinking it was terrible. A lot of it was just having more trust that it wasn't going to kill me. It's weird how much fun I have on my night FSD drives. It's kind of like going on a thrill ride -- super tense and fun at the same time.

Anyway, it has a long way to go but hopefully the new hardware will help out (I think it needs it).
 
An EV youtuber asked FSDb owners if they'd accept a refund if offered. After about 900 votes, 69% said yes. It's a good sample size but no way of knowing if any voters really purchased FSDb. As things stand now with HW3 I'd probably vote yes too.
Not a good sample since those who respond are self-selected even if all are actual owners.. People who are upset about something are far more likely to voice an opinion than those who are satisfied.

You could say that only 900 out of 360,000+ FSDb owners want a refund. It would be just as valid a conclusion.
 
An EV youtuber asked FSDb owners if they'd accept a refund if offered. After about 900 votes, 69% said yes. It's a good sample size but no way of knowing if any voters really purchased FSDb. As things stand now with HW3 I'd probably vote yes too.
I’ll take the refund and buy FSDb with a new car. One of the main reasons I’ve not upgraded … ofcourse they will have to give me $15k, not what I paid …. Which doesn’t even make sense.

We should be offered rolling to new car instead. That would be very popular.
 
Today I discovered what happens if you just let fsdb drive with the GPS being off by a fraction of a block. Made a right turn a whole block early because it got confused 😆 not really surprising, but it’s yet another vector of unpredictability. I didn’t notice the GPS being offset until that happened.
It's why I've said over and over that people need to check their GPS if they are having problems. It can cause all sorts of issues.
 
So does your gas pedal. Are they recalling cars with those as well?
Can you really not discern the difference?
On the subject of existing FSDb problems:
Michigan Left-Turn Confusion
Well, just because the Michigan traffic engineers designed a cockamamie configuration doesn’t mean Tesla’s are at fault for not understanding it!
 
  • Like
Reactions: FSDtester#1
I’ll take the refund and buy FSDb with a new car. One of the main reasons I’ve not upgraded … ofcourse they will have to give me $15k, not what I paid …. Which doesn’t even make sense.

We should be offered rolling to new car instead. That would be very popular.
I suspect there are many people in that camp. I would absolutely trade in my car if I could bring FSDb with.

I actually think it would be a great way for them to drive sales. As it is now they’re asking people to pay a $10-15k penalty for upgrading