Wiki MASTER THREAD: Actual FSD Beta downloads and experiences

[EVNow]

I wonder how Tesla will handle parking lots with a single stack? Whatever they do I suspect it will be phased in. I see lots of questions and controversies in the future with Summon.

• Will single stack include parking lots or will that be a second phase of single stack?
• How will parking lot summon change?
• Will the summon restricted distance be increased?
• How will drivers manage the potential for increased risk when summon functionality is expanded?
• Parking garages? I'll skip this option!
• etc.

I’d start any kind of merging of stacks with summon. Essentially as FSD gets better at driving in parking lots, it becomes the summon.

TomTom has a lot of parking lots mapped, but not as many as OSM. They may have to merge the maps for summon.

 

[Sporty]

That's what it seems like Tesla is aiming for with so many 10.x betas "delaying" 11. For Tesla to release single stack is a milestone in that they believe their new approach for neural networks, planning, etc. is better than the production real-world extensively tested existing Autopilot/NoA capability. (Or at least in some conditions… It'll be interesting to see if a single stack release falls back to old behavior for poor weather.)

I wouldn’t be surprised it if regresses in some ways. It’s not likely they are actively developing both stacks. I feel that the old stack is effectively dead. Eg it’s no longer being supported and updated in meaningful ways. I suspect all the infrastructure and mindshare within T is dedicated to the new stack.

Projects I’ve worked on that was replacing something before it worked this way. The new things goals is to do everything the previous thing did and more. And near the end 90% of the goals are actually met and that’s “good enough” to effect a switch over. (So long as all there are no ship-blocker / critical-issues in the outstanding work). And T has shown they have a pretty low bar quality bar on certain issues. Some of those are on hard problems so ok. But some aren’t. And more confusingly some hard issues have pretty good quality.

 

[old pilot]

Short drive today, overcast, daylight and very light rain. Acceleration set to standard and FSD to average. Some phantom slowing, made left and right turns with some help from me on the accelerator. As the rain intensity increased got the warning "FSD degraded" but the FSD stayed connected, this seemed to occur (the warning) as wiper speed increased (auto wipers) (I wonder what would happen if the wipers were in manual and I did not turn them on?).
Today I tried engaging FSD while I was still in my driveway and it worked well, entering a 2 lane road, but what I found interesting is how it made the turn. As the car left the driveway, making a left turn, instead of a gradual arc it made a very squared off turn and stayed between the white and yellow lines. The reason I find this interesting is it seems unable to make right turns that are squared off. It could be that it was starting from a stop and thus a slower speed on entry. Most of the right turns I've tried are while moving. Edit: To clarify it does made right turns but swings left before making the turn. It did not swing right to make the left turn.

 

[JulienW]

I’ve been saying this for a while - I don’t know what single stack fans are thinking when they ask for single stack. They are going to regret it.

Hopefully Tesla won’t release single stack until FSD beta is as good as NOA in terms of avg disengagements per trip.

I'm so ambivalent. I (foolishly) want the newness but it will probably be a regression based on the amount of phantom braking I have experienced on open roads with Beta.

Did notice a big improvement. Speed Bumps: Was going through a neighborhood with some old speed bumps. The old ones have eroded and are hard to see. Noticed Beta would not see thenm until close and then recognize and display it. Gives me hope that things like potholes and debris will soon be recognized.

 

[Twiglett]

I took a drive today away from the usual neighborhoods I experiment in and instead on the faster, wider roads around here.
Have to say it was a refreshing change.
FSD stopped at lights on the line instead of 10 yds back
It accelerated at normal car speed - in others words, slow for a Tesla but normal speed for regular cars.
It stopped smoothly, slowed down for merging cars rather than panic braking.
Overall pretty good.
It still made stupid mistakes of course - like stopping at a red light in the right turn lane with the right indicator going even though we were going straight on. We would be turning right 40-50 yards after the light, but its really confusing for everyone else.
But still, pretty good

 

[EVNow]

That's what it seems like Tesla is aiming for with so many 10.x betas "delaying" 11. For Tesla to release single stack is a milestone in that they believe their new approach for neural networks, planning, etc. is better than the production real-world extensively tested existing Autopilot/NoA capability. (Or at least in some conditions… It'll be interesting to see if a single stack release falls back to old behavior for poor weather.)

I don’t think single stack would be a priority for them anymore - they are probably looking at disengagement numbers in horror now. Compared to what YouTubers and internal testers did, the 20k+ testers would disengage a lot more often not willing to take the risks. My current disengagement number is probably ~1 per mile. They have to improve on that number before they do anything outside FSD.

They will probably also figure out they need a better planner and better way to manage maps. Earlier the better.

I won’t be surprised if they have stopped working on single stack and declared all hands on deck for FSD.

 

[EVNow]

I took a drive today away from the usual neighborhoods I experiment in and instead on the faster, wider roads around here.
Have to say it was a refreshing change.

I think that is closer to roads in San Jose - the kind of roads the planner has been tested a lot on.

 

[Oil Freedom]

Following a right turn today, FSD Beta confidently accelerated toward the next intersection with the nav's (correct) plan to pass straight through it. Unfortunately, the car remained in the far right lane after having made the turn.

There was no sign that FSD Beta understood it was in a right turn only lane. In fact, it continued to accelerate as if it was in the correct lane. Traffic was light, and there was plenty of room, so I allowed it as much time as possible to self-correct. However, I ended up having to brake hard and steer the car into the correct lane.

Without intervention, that would've ended with FSD Beta doing emergency braking and stopping somewhere in the intersection and blocking traffic (at best) or crashing the car (at worst).

Stay vigilant out there!

 

[EVNow]

@Oil Freedom One more thing we can attribute to likely map problem.

 

[Oil Freedom]

@Oil Freedom One more thing we can attribute to likely map problem.

It's possible the map changed; however, FSD Beta didn't have any problem with the same routing and intersection before 10.8. (Worth mentioning that the intersection hasn't changed in many years.) I'll re-visit that intersection on occasion and will be interested to see how it does.

 

[Twiglett]

just repeated the same drive from this morning, but this time at nighttime.
Still pretty smooth and confident, until it came to lane changes at which point it didn't do very well.
Looks like merging when it only has headlights to spot other cars is less confident than during daytime.
Still not too bad, only a few interventions and a few sudden slowdowns, but no panic braking.

I feel that @EVNow is pretty accurate - now the beta has expanded Tesla is getting much more data from places other than CA. Most of the videos I've seen from the initial beta folks don't show the appalling indecision that we seem to be getting.
Hopefully it just means that the rate of improvement will increase as they figure out how to cope with the mass of data heading their way.

After todays drives I think I might be able to use FSD with the family in the car as long as I'm careful where I engage it.

 

[EVNow]

It's possible the map changed; however, FSD Beta didn't have any problem with the same routing and intersection before 10.8. (Worth mentioning that the intersection hasn't changed in many years.) I'll re-visit that intersection on occasion and will be interested to see how it does.

Could be a 10.8 issue as well.

I’ve a left turn lane it sometimes tries to get away near a traffic signal.

May be the planner gets stuck in some kind of local minimum at times ?

 

[thewishmaster]

2 T-junction intersections with 3-way stops today, FSD beta insisted on letting cars parked at the top of the T (e.g. the “dead end“) go first. One was perpendicular so it sort of made sense as it pointed into the intersection. The other was parallel parked between 2 others and the beta literally waited 5 seconds before it decided the parked car wasn’t going to go.

 

[Marc_K_inNJ]

My friend finally got FSD beta yesterday. Score of 98 in NJ. Gave me a ride. Truly amazing!
It is “leaves season” in NJ now. (We blow our leaves into the street curb for the township to pick up…) So there are piles of leaves all along the street. FSD did a great job of avoiding them by going around them and coming back to the right side of the street. It didn’t show them in the visualization but it still avoided them.

 

[MAHahh]

Could be a 10.8 issue as well.

I’ve a left turn lane it sometimes tries to get away near a traffic signal.

May be the planner gets stuck in some kind of local minimum at times ?

I've had very good experiences, so far, on 10.8. About 40 miles on a combination of residential neighborhoods, rural roads, and city streets (so, no highway/AP), and only 2 required disengagements.

But, yesterday, at night, had it not understand a right only turn lane, that it confronted immediately after making a right turn onto that road. There was no traffic around, so I let it play out. It just went straight through the intersection (which was the right direction for my destination, but obviously illegal). It found itself in a "no go" painted out area (no cross hatching, just a white line border all around it). The area was the same width as a lane, so it seemed to be happy to enter and just keep going. Eventually, the pavement ended, and it moved over a lane to the left, and back into a proper lane, and proceeded as if nothing had happened.

 

[lascavarian]

Oh good, I thought that was just me having that issue last night.

When I look at the notifications, (Menu-> Service -> Notifications) the explanatory text says that "Cruise control unavailable" (DI_a175) followed with a notification of "Autosteer temporarily unavailable" (APP_w207), each with a link to the owners manual. That referenced that it could be a sensor error if it keeps occurring. Some googling shows that the cruise control unavailable notification was also a bug in previous software releases as well.

Fingers crossed that it is just a software bug since I have not seen it at all today. I did do a power reset process that seemed to help some people (Google) that involved disabling sentry, unplugging USB and letting the car sleep.

I've been waiting for reports of other people on 10.8 to see this in the hopes it is just software. If it kept happening with it only happening to me, I was going to schedule a service visit.

Are you seeing the same notifications?

TL;DR Disable sentry, let car sleep, cross fingers and hope that fixes it.

”cruise control unavailable” is best reported with the Bug Report process when it happens. This will help Tesla figure this out. My opinion is that this is unliky to be SW related but again only my opinion.

Losing cruise control is a serious problem IMO a bit deeper than FSD beta code flakiness. It may be that the FSD beta code is revealing a rough spot in general functions. Deep sleep may help but it may be only a bandaid and not a real solution. Bug reports capture data that Tesla can use to drive to a root cause solution.

 

[Gilliland]

There is a section of I271 near my home where the road has recently changed and Tesla's map has not been updated. Specifically, the road divides into two parts (basically express and local, though they don't use those terms). Since my exit is in that area, I need to exit twice, once to the local part of the road and again to my own exit. The Tesla navigation and NOA directs me to my exit, and I've solved the problem of getting into the local section by simply forcing the car to drive in the exit lane. It takes the exit without complaint.

On the ramp between the two sections, NOA drops out several times because it thinks it is off the freeway, but it becomes active again when I get to the local section of the freeway.

Yesterday, I drove through that section for the first time since receiving the FSD beta. I did what I always do, moving into the exit lane and reducing my speed to 50MPH. Again NOA took the exit without complaint.

But at the point where NOA drops out, FSD took over. It immediately changed my max speed to 75 and began a high rate of acceleration. I dialed the max speed back down to 50, and FSD immediately set it back to 75 and kept accelerating. I dialed it back down again, and it raised it again. I continued this fight with FSD for the entire time it had control - probably no more than an eighth of a mile, but long enough that I went through about five cycles of this up/down/up/down argument with FSD. As the ramp merged back into the local section of the freeway, NOA regained control and the fight was over.

 

[Sporty]

There is a section of I271 near my home where the road has recently changed and Tesla's map has not been updated. Specifically, the road divides into two parts (basically express and local, though they don't use those terms). Since my exit is in that area, I need to exit twice, once to the local part of the road and again to my own exit. The Tesla navigation and NOA directs me to my exit, and I've solved the problem of getting into the local section by simply forcing the car to drive in the exit lane. It takes the exit without complaint.

On the ramp between the two sections, NOA drops out several times because it thinks it is off the freeway, but it becomes active again when I get to the local section of the freeway.

Yesterday, I drove through that section for the first time since receiving the FSD beta. I did what I always do, moving into the exit lane and reducing my speed to 50MPH. Again NOA took the exit without complaint.

But at the point where NOA drops out, FSD took over. It immediately changed my max speed to 75 and began a high rate of acceleration. I dialed the max speed back down to 50, and FSD immediately set it back to 75 and kept accelerating. I dialed it back down again, and it raised it again. I continued this fight with FSD for the entire time it had control - probably no more than an eighth of a mile, but long enough that I went through about five cycles of this up/down/up/down argument with FSD. As the ramp merged back into the local section of the freeway, NOA regained control and the fight was over.

FSD is new so I get it might have less automated test coverage. But the same kind of thing happens on NoA .. cancel lane change. 1 second later.. “let’s change lanes”. It’s been that way for years, and clearly no test case for the problem.

The system is too complicated to be successful without a solid testing methodologies being applied. Gorilla testing will not cut it.

 

[d21mike]

10.8 still made the changing lanes for no reason. I.E. Need to turn right in 2 blocks. For no reason it decided to move to the left lane and then at the right turn it can no longer get back to the right lane because a car is not blocking. This move to the left lane happens when there is no cars on the road which in that case it can get back to the right lane. It did improve getting into the right turn lane to make a right turn near my house. But then it turn into the far right lane where cars are parked in front of their house. It is going pretty fast to the back of a car where I allowed for a bit then it suddenly jerked to the left to go around the car. It does not understand which lane to be in and it will not allow me to tell it to stay in a particular lane. Hard to understand after over a year of testing it still has some of these issues.

 

[Sandor]

Oh good, I thought that was just me having that issue last night.

When I look at the notifications, (Menu-> Service -> Notifications) the explanatory text says that "Cruise control unavailable" (DI_a175) followed with a notification of "Autosteer temporarily unavailable" (APP_w207), each with a link to the owners manual. That referenced that it could be a sensor error if it keeps occurring. Some googling shows that the cruise control unavailable notification was also a bug in previous software releases as well.

Fingers crossed that it is just a software bug since I have not seen it at all today. I did do a power reset process that seemed to help some people (Google) that involved disabling sentry, unplugging USB and letting the car sleep.

I've been waiting for reports of other people on 10.8 to see this in the hopes it is just software. If it kept happening with it only happening to me, I was going to schedule a service visit.

Are you seeing the same notifications?

TL;DR Disable sentry, let car sleep, cross fingers and hope that fixes it.

That was my experience last night; Autopilot would disengage with the panic red wheel almost instantly; there was some fleeting error on the bottom right of the screen underneath the car cartoon, something to the effect "FSD encountered an unknown error" or similar. After re-engaging it would quit again (DI_a175 error in the notifications) and after few attempts, I'd get the Autosteer temporarily unavailable error too (APP_w207) and at that point the steering wheel grey icon would completely disappear from the screen. TACC would be non-functional too and it would quit after engaging.

After researching some on TMC, I parked, removed the USB flash drive and then the FSD and NoA would work but only for 10-15 minutes before panicking again. Home I've left the USB unplugged and let the car sleep overnight while charging. That seems to have cured the issue for now. See this previous incident too.

This happened to me a week ago. I submitted a ticket to tesla and they replied fast with the fix. (I love being able to submit tickets to my car. I hate traditional auto dealers) Hope this helps others. This fixed it for me

Please restore your autopilot features by following the steps below.

1. Set Sentry Mode to OFF from Controls > Safety & Security menu
2. Remove USB Drive (if inserted)
3. Allow vehicle to sleep
4. Autopilot features should return after the sleep cycle

In order to prevent the issue from repeating, for now please set your Sentry Mode settings to exclude Home (Controls > Safety & Security > check the Exclude Home checkbox)

Note: You will need to have you Home address saved in Navigation for this exclusion to apply.

The Tesla provided solution seems similar and so far so good. I do not have Sentry normally enabled on a daily basis, just on demand. My 1.25h morning commute was perfect: fully FSD/NoA with no disengagements at all. I am still keeping the USB drive unplugged and I'll update after the night commute as that was when the problems occurred- at night. The USB drive was a 18 months old 64GB Samsung. I did check it and no corruption detected on a deep scan... I am replacing it anyways with a new one... $12.50 to be safe sounds reasonable.

2020 Model 3 LRAWD running 2021.44.25.6 (FSD 10.8)

I'll update later tonight.