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

FSD Beta Videos (and questions for FSD Beta drivers)

This site may earn commission on affiliate links.
E787CAD0-2D99-4F60-8396-03C6CB2C80A0.jpeg

I had some issues at this intersection. Hoping they work on it.
 
The silly thing is the logic should already exist. It’s the exact same logic that‘s used when you have a sweeping right turn lane with a yield onto the target road. Of course FSD seems to have trouble with that, too.

Maybe Chuck needs to start making UPR videos.
I don't know how much this has changed, but AFAIK most of the logic is still hand coded.

Even if there is some NN being used for it, roundabouts are relatively rare in the US, so some targeted testing may be necessary. There are suggestions Tesla might actually be sending test cars to Chuck's turn to do targeted testing of UPLs:

Maybe some representative roundabouts can be targeted in a similar way. There is some risk of overfitting to a given situation, but I think overall it'll still lead to significant improvements in the general case (as it seemed to have done for many types of turns with the latest update).
 

another UPL video from Chuck for the 69.2 version. 100 percent success today, but mostly low traffic scenarios. Interesting on turn number 4 or 5 where there was a car making an UPL ahead of Chuck’s car, and you can see how far the human driver crept up to see traffic coming from the left, which looked to be further than the Tesla normally goes.
 

another UPL video from Chuck for the 69.2 version. 100 percent success today, but mostly low traffic scenarios. Interesting on turn number 4 or 5 where there was a car making an UPL ahead of Chuck’s car, and you can see how far the human driver crept up to see traffic coming from the left, which looked to be further than the Tesla normally goes.
But, but visibility is not restricted at Chuck's ULT!

/s
 
  • Funny
Reactions: AlanSubie4Life

another UPL video from Chuck for the 69.2 version. 100 percent success today, but mostly low traffic scenarios. Interesting on turn number 4 or 5 where there was a car making an UPL ahead of Chuck’s car, and you can see how far the human driver crept up to see traffic coming from the left, which looked to be further than the Tesla normally goes.
Interesting observation on #5, I hadn't noticed how far forward that car went. Have to know your car! (Also it's not necessary since vision is not obstructed to that degree.)
We'll see a much higher failure rate when he does this in denser traffic (should be Monday if he's not working). One major issue is that the car is only getting up to 12mph (slowly) to cross to the median (not proceeding), and that's just likely not going to be fast enough unless they have also made the required gap larger. It’s got a weird inconsistency where traffic is allowed to be uncomfortably close (for the chosen crossing speed, which currently means it takes ~5 seconds to complete the maneuver). We'll also see if they have the continued issues with localization of the car in the median seen in other videos.

but visibility is not restricted at Chuck's ULT!
That would be a weird thing for someone to say, for sure! Visibility is restricted at the stop line and becomes very good after sufficiently advancing.
 
Last edited:
  • Like
Reactions: Daniel in SD
I just took my one year old pup to the vet for some vaccinations using FSD beta 10.69.2. I had my first intervention free round trip. No red lights were run and the car followed the navigation perfectly. Three yields, one URT and 2 ULTs, all with compromised visibility were handled perfectly on the outbound trip, as was a PLT at a controlled intersection.

This version looks pretty solid in comparison to 10.69.1.1.
 
Changing your tune I see? Visibility will always become very good after sufficiently advancing.
I think you are thinking of someone else.
No tune change. I bring receipts. I have been crystal clear on the issues on this turn, and even provided pictures to show the required “window.”

The point here is whether the vehicle needs to proceed to an unsafe position to be able to see on this turn. And clearly it does not. On some turns it is not possible to obtain excellent visibility without exposing the vehicle to a collision. This is not one of those cases.
 
Last edited:
  • Like
Reactions: Daniel in SD
I've pontificated at length about the benefits of accelerator application in various other threads, so thought it would be good to show an example of the "slalom" issue that happens if you're a bit too vigorous, which I have also mentioned.

No big deal, but it will readily do this on surface streets too (crossing into the oncoming traffic lane if unmarked, on the onset of the slalom). Obviously it would be better if the car would just blast around corners rather than slowing down to 16mph (way too slow in this case - even the massive work truck caught up to me, and I didn't want him caving in the rear of my vehicle).

Obviously, both hands on the wheel at all times. A few minor traffic violations here (specifically the lack of signaling, for two lane changes), of course, but it's enough of an issue that I think it is worth the risk, to just ensure people understand the behavior.

This is 10.69.1.1, but this feature is not a new one. I think it just slaloms more smoothly now.


Normally I would not let the initial slowing take place (earlier application of accelerator), and then the corner would be taken at a normal speed, and there would be no need for more assertive application.
 
Last edited:
I've pontificated at length about the benefits of accelerator application in various other threads, so thought it would be good to show an example of the "slalom" issue that happens if you're a bit too vigorous, which I have also mentioned.

No big deal, but it will readily do this on surface streets too (crossing into the oncoming traffic lane if unmarked, on the onset of the slalom). Obviously it would be better if the car would just blast around corners rather than slowing down to 16mph (way too slow in this case - even the massive work truck caught up to me, and I didn't want him caving in the rear of my vehicle).

Obviously, both hands on the wheel at all times. A few minor traffic violations here (specifically the lack of signaling, for two lane changes), of course, but it's enough of an issue that I think it is worth the risk, to just ensure people understand the behavior.

This is 10.69.1.1, but this feature is not a new one. I think it just slaloms more smoothly now.


Normally I would not let the initial slowing take place (earlier application of accelerator), and then the corner would be taken at a normal speed, and there would be no need for more assertive application.
I didn't see you tap the report button for the un-signaled lane change. Shame!
Not sure why people disagree with your copious accelerator usage though, accelerator usages is a direct error signal sent to Tesla that can be used to train the system and clearly it's more valuable for Tesla to get reports of errors when the car is going the correct speed versus too slow.
Arguably it would be even better to have an "FSD Pro" mode so that brake and steering corrections could be sent to Tesla without disengaging FSD.
 
I didn't see you tap the report button for the un-signaled lane change. Shame!
I've reported this on this specific turn several times before, as well as on others. While I didn't report this incident, I did report a merging slowdown on the freeway about a minute after this, so it's probably captured, for their detailed review.

Arguably it would be even better to have an "FSD Pro" mode so that brake and steering corrections could be sent to Tesla without disengaging FSD.

That would be so awesome.
 
Last edited:
Arguably it would be even better to have an "FSD Pro" mode so that brake and steering corrections could be sent to Tesla without disengaging FSD
Wait so how would you disengage FSD if braking didn't disengage it in an emergency? That sounds like the most horribly dangerous thing haha. The brake pedal should always be your last resort "red button" disable option
 
Wait so how would you disengage FSD if braking didn't disengage it in an emergency? That sounds like the most horribly dangerous thing haha. The brake pedal should always be your last resort "red button" disable option
The brake would still work it just wouldn't disengage FSD! When you released the brake pedal the car would be able to accelerate again. The same way the accelerator works now just in reverse. I feel like this would help Tesla get training data to fix slow response to speed limit signs, red lights, and stopped traffic.
You're right though that FSD should disable after manual ABS engagement as that indicates some sort of emergency.
I was also suggesting that this mode would only be used by expert FSD users looking to provide as much feedback as possible to Tesla, like @AlanSubie4Life.
 
  • Like
Reactions: AlanSubie4Life
Wait so how would you disengage FSD if braking didn't disengage it in an emergency? That sounds like the most horribly dangerous thing haha. The brake pedal should always be your last resort "red button" disable option

The only thing about this plan that wouldn't work very well is that you could not use the accelerator to modulate regen. That's where the brake would have to come in (in addition to actual emergencies). This would be a departure from normal (Tesla) driving habits, unfortunately. But at least you could get the speed & steering roughly correct. Tesla would have to pull out their blended regen/braking tricks for sure, though, since it would make sense to use more regen when you press the brake pedal rather than using the brakes (many cars do this).

And disabling could be done with the stalk as usual.