FSD Beta 10.69

[Z_Lynx]

I need to figure out how to record some of this nonsense without holding the phone to record and the car yelling at me for it.

It is SO helpful to have audio and video of both traffic and screen to analyse "what the hell was that!!!".
Can't read the tiny print messages on the video but can hear if there has been a ding indicating disengagement and more.
I notice I am swearing more often.

Delkin Devices Fat Gecko Dual Suction Camera Mount
Amazon.com
Track price on CamelCamel and look for a drop to ~$40 (got mine in June).

Get this inexpensive adjustable mount, REMOVE ARM (vibration) and only use the clamping part (you could even use it for your phone).
Ailun Tripod Phone Mount Holder
Amazon.com
Picture shows a couple strips of rubber to take up slack due to slim camera:

Almost any rectangular cheap dashcam. Mine was perhaps $20 a few years ago. Pay attention to how it connects to power. Ideally, you do not want to use a 12V charger adapter because (at least for me) the armrest won't close completely if you have a cable leaving the compartment (2020 MY).

 

[bhakan]

On a 2000 mile round trip and had to stop using FSD within first 30 mins. Annoyed the *sugar* out of my passenger trying to randomly turn left and right on a straight road where GPS was clearly indicating to go straight

TACC it is from here on

I cannot use after 6 pm when it gets darker inside the cabin. The instant I turn on AP gives me the red hand pay attention to road and each time I disengage. After such 5 disengagements I get a strike. Got two strikes and each time within the first half mile driving when it is darker. Evening is when I find it more useful to use AP. Emailed the early access twice and have not heard anything from them while they respond to YouTubers. When I approached my SC with cabin camera issues they washed their hands and said I have to reach out FSD crew. The last two weekend trips were in my Outback.

 

[bhakan]

It is SO helpful to have audio and video of both traffic and screen to analyse "what the hell was that!!!".
Can't read the tiny print messages on the video but can hear if there has been a ding indicating disengagement and more.
I notice I am swearing more often.

Delkin Devices Fat Gecko Dual Suction Camera Mount
Amazon.com
Track price on CamelCamel and look for a drop to ~$40 (got mine in June).

Get this inexpensive adjustable mount, REMOVE ARM (vibration) and only use the clamping part (you could even use it for your phone).
Ailun Tripod Phone Mount Holder
Amazon.com
Picture shows a couple strips of rubber to take up slack due to slim camera:

Almost any rectangular cheap dashcam. Mine was perhaps $20 a few years ago. Pay attention to how it connects to power. Ideally, you do not want to use a 12V charger adapter because (at least for me) the armrest won't close completely if you have a cable leaving the compartment (2020 MY).

View attachment 857729

It does not work well for a refresh Model S as the actual car lanes and alerts appear on the smaller dashscreen. That's why I am trying to record with my Pixel 6 Pro phone on a cup mount and try to record. The problem is during the daytime there is so much reflection it's difficult to get a good picture of the dash screen.

 

[PACEMD]

Ok, that's it, I'm changing the name from FSD, Full Self Driving, to Ludicrous Self Driving, you know, LSD, to account for the frequent feeling someone has slipped something in the car's Kool-Aid.......

 

[JulienW]

Yesterday I think my Beta regressed to a very early build and maybe even Alpha version 1. Frustrated on every drive. Hope it was restored back to 69.2 today.

And today I just had the most unbelievably great (Whole Mars Catalog great) drive from Midtown ATL out to Decatur (right next to the Decatur Tesla center) and back. I still can't wrap my head around it but I had 0 disengagements (did do one, my bad, at a flashing yellow/unprotected left turn but reengaged before the turn was made) on both 10 mile drives there and back. I wish I had a video of the 2 drives just so I could watch as proof to myself I'm not .

 

[Dewg]

I cannot use after 6 pm when it gets darker inside the cabin. The instant I turn on AP gives me the red hand pay attention to road and each time I disengage. After such 5 disengagements I get a strike. Got two strikes and each time within the first half mile driving when it is darker. Evening is when I find it more useful to use AP. Emailed the early access twice and have not heard anything from them while they respond to YouTubers. When I approached my SC with cabin camera issues they washed their hands and said I have to reach out FSD crew. The last two weekend trips were in my Outback.

Try switching your screen from nighttime to daytime mode, which will brighten the cabin with some light. You can adjust the brightness so it's not as blinding, but the camera should see you better. The goal is still to replace the camera, but you may be able to use FSD at night while you work with the SC team.

 

[GregalseT]

This reminds me of all of the "sudden unintended acceleration" reports from people who INSIST they pressed the brake and yet the car zoomed forward. If anyone has FSDb just turn off with no notification, get it on a video, or it didn't happen. Extraordinary claim requires extraordinary proof. And if that evidence surfaces, I will acknowledge that that is a more dangerous situation, though still manageable if you're paying attention.

Example: there have been times on long highway commute stretches where I was not on AP but somehow I thought I was (I had disengaged but forgot to turn it back on, and later I just assumed AP was on), so my torquing the wheel was actually me trying to anticipate the steering of the system. And then when I got lax for a second and the car started to drift from center, that's when I would realize I wasn't on AP. If you're paying attention at all times as you should be, there should be no real danger.

It doesn't stop "with no warning whatsoever." The car is blaring the critical alert sound, and the screen shows the red wheel.

Every one of us FSDb testers accepted a warning via the center screen that stated "it may do the wrong thing at the worst time." That covers situations when FSDb has gotten itself in a state where it cannot continue operating and suddenly disconnects with the red wheel warnings.

If you want to keep pushing that this particular scenario is very dangerous, then I'd say FSDb in any situation is dangerous, again because "it may do the wrong thing at the worst time." I don't really see why there's a distinction for FSD suddenly disengaging.

Either way, if you're paying attention and treating FSDb operation with caution, a sudden disengagement is quite easy to recover from. "you must always keep your hands on the wheel and pay extra attention to the road. Do not become complacent." So if you're doing this as Tesla instructs, then recovering from a sudden disengagement means just tapping the accelerator to maintain speed, something we all likely do very frequently while FSDb is in typical operation.

FSDb is as dangerous as YOU make it. You are still the driver, and you control what the car does.

Right...I'm not talking about when the system alerts you with warning sounds and displays a "red wheel with hands on it." I'm talking about the times when it simply disengages with no prior warning, likely because it feels FSD cannot properly process and handle the situation. I agree with the other poster that noted that wheel torque requirements to disengage FSD seem to vary widely, and it becomes very easy to disengage in the situations I'm talking about. I know that slight torque on my part has caused disengagements where my prior experiences would indicate that a disengagement was not imminent at that level of torque, but I also recall situations where it disengaged with no apparent torque.

For the record, I'm not pushing any agenda. Simply giving my two cents. Also, you seem to keep lecturing me on how to be a safe, observant driver. My wife and I collectively are safety score 100 drivers, if that's a valid metric to consider here. I also recall Musk's "worst thing at the worst time" comment, and go about my driving accordingly. All of this said, if it's true that sudden disengagements are a thing, there is only one argument one could make that would indicate that it is not a more dangerous situation than not suddenly disengaging; namely, that it would have been even more unsafe had the car not disengaged.

This reminds me of all of the "sudden unintended acceleration" reports from people who INSIST they pressed the brake and yet the car zoomed forward. If anyone has FSDb just turn off with no notification, get it on a video, or it didn't happen. Extraordinary claim requires extraordinary proof. And if that evidence surfaces, I will acknowledge that that is a more dangerous situation, though still manageable if you're paying attention.

Example: there have been times on long highway commute stretches where I was not on AP but somehow I thought I was (I had disengaged but forgot to turn it back on, and later I just assumed AP was on), so my torquing the wheel was actually me trying to anticipate the steering of the system. And then when I got lax for a second and the car started to drift from center, that's when I would realize I wasn't on AP. If you're paying attention at all times as you should be, there should be no real danger.




A family member recently got her first Tesla. I was in the passenger seat when she was trying out AP. She disengaged so many times via steering wheel without noticing. Dropping to TACC is now a single chime that can be easily missed. Or someone hears it and doesn't recognize the significance of it, because Teslas do chime a lot in general. If I were to ask her if the car notified her of disengaging AP, she would confidently say no. And she'd be wrong.

 

[Dewg]

I'm talking about the times when it simply disengages with no prior warning, likely because it feels FSD cannot properly process and handle the situation.

That feels like the software crashing for some reason. If FSD needs to disengage, it should always alarm and show the red wheel, so you know that it's not going to drive and you have to.

I've had that happen a time or two when a camera goes out - I suddenly get the red wheel and have to take over, and the visualizations go out. When I press the camera button to show the three cameras, one of them is black with a loading circle.

I've never had it crash/fail without giving the red-wheel and alarm - so that's disturbing that it can just stop with absolutely no warning.

 

[willow_hiller]

I also recall situations where it disengaged with no apparent torque

Do you happen to have "Joe Mode" on to reduce the volume of chimes? I've got it on, and occasionally I'll disengage and just barely hear the chime over the radio.

 

[ironwill96]

This reminds me of all of the "sudden unintended acceleration" reports from people who INSIST they pressed the brake and yet the car zoomed forward. If anyone has FSDb just turn off with no notification, get it on a video, or it didn't happen. Extraordinary claim requires extraordinary proof. And if that evidence surfaces, I will acknowledge that that is a more dangerous situation, though still manageable if you're paying attention.

Example: there have been times on long highway commute stretches where I was not on AP but somehow I thought I was (I had disengaged but forgot to turn it back on, and later I just assumed AP was on), so my torquing the wheel was actually me trying to anticipate the steering of the system. And then when I got lax for a second and the car started to drift from center, that's when I would realize I wasn't on AP. If you're paying attention at all times as you should be, there should be no real danger.




A family member recently got her first Tesla. I was in the passenger seat when she was trying out AP. She disengaged so many times via steering wheel without noticing. Dropping to TACC is now a single chime that can be easily missed. Or someone hears it and doesn't recognize the significance of it, because Teslas do chime a lot in general. If I were to ask her if the car notified her of disengaging AP, she would confidently say no. And she'd be wrong.

Exactly - have had numerous times where I accidentally disengaged from too much wheel torque but it leaves TACC on so you don’t realize until it wanders out of lane that it’s not actually on FSD on autopilot. I do think this should be a setting we can control - should wheel override turn off autopilot fully or just drop to TACC.

 

[sleepydoc]

Exactly - have had numerous times where I accidentally disengaged from too much wheel torque but it leaves TACC on so you don’t realize until it wanders out of lane that it’s not actually on FSD on autopilot. I do think this should be a setting we can control - should wheel override turn off autopilot fully or just drop to TACC.

I’ve had occasions where it was in TACC instead of FSD but it’s always evident pretty quickly and if you’re paying attention you should catch it just as quickly. For me it’s usually something like “what the heck is FSD doing now??? … oops! It wasn’t on!”

 

[GregalseT]

Do you happen to have "Joe Mode" on to reduce the volume of chimes? I've got it on, and occasionally I'll disengage and just barely hear the chime over the radio.

I don't use Joe Mode and I do hear the chime, but of course that alerts me that FSD is not engaged after it has already disengaged.

Here's an example of when I believe it's disengaging with no prior warning, though I guess this would be AP, and not FSD. I'm driving down the highway and see that it narrows down to one lane because of road work. I wonder to myself how the car is going to handle the situation. As it approaches the narrowing point, I see, and I believe the car now realizes, that it has gotten itself into an untenable situation and it just disengages, with no prior warning, and the car is now in my control. Granted, I've got my hands on the wheel so maybe it's primed in this type of situation to respond to virtually any amount of input with a disengagement. If true, then this is practically tantamount to my original point, as I would not expect the car to disengage with such little torque.

For some reason this also seems to happen sometimes when it's making a protected left turn and on the outside lane of a two lane turn. It's either disengaging on its own or the torque needed is very minimal, which then begs the question as to why torque requirements to disengage would vary so widely.

My theory is it's

1) disengaging with no input and with no prior warning, or
2) disengaging with much lower torque requirements than normal because
3) it finds itself in a situation that it does not feel comfortable handling.

It would be hard to convince me this aspect of FSD is not dangerous. Overall I do not believe FSD is dangerous, and I use it routinely with my family in the car, etc.. I also have about 90% of my net worth tied up in Tesla, so I definitely have no ulterior motive here. Just passing along my observations and opinions.

Remember when we used to debate about the fonts and typefaces available on computers? How pricey they were, how the Macintosh system had this typeface available for use, while Windows had this typeface available? About how typeface packages cost so much, etc? Now of course we look back at that rather sheepishly and wonder what all the fuss was about as now typefaces are essentially free and ubiquitous.

I imagine all this hand-wringing about FSD will age just as well.

 

[Z_Lynx]

It does not work well for a refresh Model S as the actual car lanes and alerts appear on the smaller dashscreen.

The Fat Gecco can be put anywhere on the glass ceiling and is extremely adjustable. It may still be possible to position a cam to provide you with some information - though a more creative solution might be needed if you actually need to READ the alerts. With only one cam and a wide angle, I can't read them.

Two cams? One mounted on the door?

Then again perhaps it can be done with mirrors?

 

[nvx1977]

For the record, I'm not pushing any agenda. Simply giving my two cents. Also, you seem to keep lecturing me on how to be a safe, observant driver. My wife and I collectively are safety score 100 drivers, if that's a valid metric to consider here. I also recall Musk's "worst thing at the worst time" comment, and go about my driving accordingly. All of this said, if it's true that sudden disengagements are a thing, there is only one argument one could make that would indicate that it is not a more dangerous situation than not suddenly disengaging; namely, that it would have been even more unsafe had the car not disengaged.

I am not lecturing you to be a safe driver. 100 safety score is impressive in my book. Most people couldn't get it, even if they knew "tricks".

What I'm saying is for ANYONE who is paying attention as they should be, FSDb is not dangerous. Because ultimately we are still the ones making sure the car is making lawful and safe maneuvers. It shouldn't matter if FSD suddenly disengages, with warning or without; or it doesn't disengage and try to drive into a ditch or oncoming traffic. We humans should expect dumba** moves and catch them. And statistically the people in this internal beta are doing fine avoiding serious accidents.

I understand there's a complacency issue to contend with, esp when FSDb gets much better (I don't think we're there yet). I just don't agree with the characterization that FSDb is dangerous in any of the scenarios you presented. Or if we were to call it dangerous, then any of the hundreds of other errors (like running red lights, crossing double yellow lines, etc) are equally dangerous. The system is as dangerous as you allow it to be.

 

[nvx1977]

Here's an example of when I believe it's disengaging with no prior warning, though I guess this would be AP, and not FSD. I'm driving down the highway and see that it narrows down to one lane because of road work. I wonder to myself how the car is going to handle the situation. As it approaches the narrowing point, I see, and I believe the car now realizes, that it has gotten itself into an untenable situation and it just disengages, with no prior warning, and the car is now in my control. Granted, I've got my hands on the wheel so maybe it's primed in this type of situation to respond to virtually any amount of input with a disengagement. If true, then this is practically tantamount to my original point, as I would not expect the car to disengage with such little torque.

For some reason this also seems to happen sometimes when it's making a protected left turn and on the outside lane of a two lane turn. It's either disengaging on its own or the torque needed is very minimal, which then begs the question as to why torque requirements to disengage would vary so widely.

The other issue is that not a lot of people report AP/FSD suddenly disengaging without notification. Most people are going to say that when it does happen the car basically freaks out with visual and audio notifications. So you have to acknowledge you're quite the outlier here.

As far as disengagement torque goes, in my experience, if you torque the wheel in the same direction as what FSDb is currently doing, it requires very light torque. Conversely if you take control by turning the wheel the opposite way as FSDb, particularly if FSD jerk at that instant was high, it takes a LOT of torque to disengage.

 

[KLam]

Actually, if you turn the steering wheel exactly as FSDb turns it, there is no sense of feedback torque on you hands and the system will complaint that your hand is not on the steering wheel! The system relies on your turning is slightly out of sync to create the torque for sensing your hand is there.

 

[nvx1977]

Actually, if you turn the steering wheel exactly as FSDb turns it, there is no sense of feedback torque on you hands and the system will complaint that your hand is not on the steering wheel! The system relies on your turning is slightly out of sync to create the torque for sensing your hand is there.

Below is how I rationalize why the discrepancy of disengagement torque exists. Probably completely wrong, but feels like it works like this:

Just making up numbers here, but if the threshold to disengage is +/-5Nm, and FSDb is applying -4Nm, I need to actually apply +9Nm to disengage. But if FSDb is applying +4Nm, I only need to apply +1Nm to disengage. Positive vs negative in this case just indicates the direction of the torque (left or right turn of the wheel).

Tesla should be looking at user-applied torque only, not the net torque on the system at that instant, so the user feels the resistance to disengage is constant.

 

[willow_hiller]

I know there have been some posts about upload volume before, but I just figured out how to track my Model 3's data, and I'm staggered by how much it uploads.

Just from a 4 mile drive, it uploaded 2.699 GB:

Does anyone know offhand the joint bitrate of the cameras? I'm wondering if it's even possible to generate that much video data on such a short drive.

Edit: Okay, found a tweet from Green that says 12-40 Mbps. So my drive was 16 minutes or 960 seconds. Multiply by an average 26 Mbps, and divide by 8 bits in a byte yields 3.1 GB. So it seems likely they just uploaded all 8 camera views of my entire drive.

 

[sleepydoc]

This reminds me of all of the "sudden unintended acceleration" reports from people who INSIST they pressed the brake and yet the car zoomed forward. If anyone has FSDb just turn off with no notification, get it on a video, or it didn't happen. Extraordinary claim requires extraordinary proof. And if that evidence surfaces, I will acknowledge that that is a more dangerous situation, though still manageable if you're paying attention.

Example: there have been times on long highway commute stretches where I was not on AP but somehow I thought I was (I had disengaged but forgot to turn it back on, and later I just assumed AP was on), so my torquing the wheel was actually me trying to anticipate the steering of the system. And then when I got lax for a second and the car started to drift from center, that's when I would realize I wasn't on AP. If you're paying attention at all times as you should be, there should be no real danger.




A family member recently got her first Tesla. I was in the passenger seat when she was trying out AP. She disengaged so many times via steering wheel without noticing. Dropping to TACC is now a single chime that can be easily missed. Or someone hears it and doesn't recognize the significance of it, because Teslas do chime a lot in general. If I were to ask her if the car notified her of disengaging AP, she would confidently say no. And she'd be wrong.

There’s definitely a bit of a learning curve but I actually find the various chimes to be quite useful. Of course if you haven’t learned what they mean they don’t help much.

 

[sleepydoc]

Below is how I rationalize why the discrepancy of disengagement torque exists. Probably completely wrong, but feels like it works like this:

Just making up numbers here, but if the threshold to disengage is +/-5Nm, and FSDb is applying -4Nm, I need to actually apply +9Nm to disengage. But if FSDb is applying +4Nm, I only need to apply +1Nm to disengage. Positive vs negative in this case just indicates the direction of the torque (left or right turn of the wheel).

Tesla should be looking at user-applied torque only, not the net torque on the system at that instant, so the user feels the resistance to disengage is constant.

I’m pretty sure it is looking at user-applied torque. Part of the problem is that during a turn the car-applied torque is not consistent and will change faster than a human can adapt so if you’re trying to apply torque on top of that you end up applying too little or too much.

One trick I’ve used is to grip the wheel enough to apply some drag as it’s turning but not actually stop it. That usually satisfies the system without me having to overthink it.