FSD beta 11.4.3?

[legendsk]

I went for a test drive after last night's update to 2023.7.15 / v11.4.3
It didn't fix any of the problems that 2023.7.10 created - it still wants to drive in the oncoming traffic lane instead of the correct lane. It slammed to a dead stop on the highway, because we met someone, who was well in the left lane and we were in our lane correctly.

On the good side(?) it did fix phantom braking on flashing yellow caution lights. I'm not sure I really approve of the solution. Flashing yellow caution lights no longer appear in the visualization. So yeah, I guess if you pretend they aren't there, the problem of braking or not is solved?

 

[honeyfist301]

I got 2023.7.15 yesterday and used it just now and keep having cabin camera error and cannot use it. Is it only me? I checked the cabin camera on the service tab and it was working fine.

 

[Dewg]

I got 2023.7.15 yesterday and used it just now and keep having cabin camera error and cannot use it. Is it only me? I checked the cabin camera on the service tab and it was working fine.

When you get the error, press the voice button and say "bug report". Then open a service ticket and reference the bug report date/time. Service will connect remotely and get the telemetry from the car.

 

[honeyfist301]

When you get the error, press the voice button and say "bug report". Then open a service ticket and reference the bug report date/time. Service will connect remotely and get the telemetry from the car.

Thanks for the solution! Will try that.

 

[Eddie123]

I got my 11.4.3 two days ago. To me, it's the 2nd most dangerous version ever released since I beta tested FSD daily from Jan 2022. Just yesterday, while on FSD on the same route I've been testing, it prematurely moved into the left turn lane on three intersections and attempted to move back into the main traffic to follow the correct route. I had to override each time to prevent an accident. Another dangerous feature is the random lane change chimes, alerting you an impending lane change. The problem is: there was no intention of any lane change, and it made me look at the screen each time. In addition, even if you select 'Minimal Lane Change,' it signals to the right lane, then back to the left lane, and then repeat for no apparent reasons. All of these features are dangerous and it's far from being ready. All in all, the last 5 updates have more negative impacts than positive features.

 

[rlsd]

I got my 11.4.3 two days ago. To me, it's the 2nd most dangerous version ever released since I beta tested FSD daily from Jan 2022. Just yesterday, while on FSD on the same route I've been testing, it prematurely moved into the left turn lane on three intersections and attempted to move back into the main traffic to follow the correct route. I had to override each time to prevent an accident. Another dangerous feature is the random lane change chimes, alerting you an impending lane change. The problem is: there was no intention of any lane change, and it made me look at the screen each time. In addition, even if you select 'Minimal Lane Change,' it signals to the right lane, then back to the left lane, and then repeat for no apparent reasons. All of these features are dangerous and it's far from being ready. All in all, the last 5 updates have more negative impacts than positive features.

The random lane change chime exists in 11.4.2 too.

 

[JB47394]

All of these features are dangerous and it's far from being ready.

Correct. It's not supposed to be ready for anything but testing. You are testing the software, not using it, and certainly not relying on it.

I suspect that people are letting their guard down with the most recent versions of FSDb because it's actually starting to work. Or, at least, giving the impression that it should work. Tesla has worked on the smoothness, and the software is getting more confident, and I think that has gotten some drivers to let down their guard. When the software makes bad choices, it's particularly jarring.

This may be the most dangerous stage of testing for the software. Though it could get progressively more dangerous as it gets progressively more competent.

 

[uscbucsfan]

Correct. It's not supposed to be ready for anything but testing. You are testing the software, not using it, and certainly not relying on it.

I suspect that people are letting their guard down with the most recent versions of FSDb because it's actually starting to work. Or, at least, giving the impression that it should work. Tesla has worked on the smoothness, and the software is getting more confident, and I think that has gotten some drivers to let down their guard. When the software makes bad choices, it's particularly jarring.

This may be the most dangerous stage of testing for the software. Though it could get progressively more dangerous as it gets progressively more competent.

I agree with this. Honestly on 11.4.2, I knew that anytime a turn lane was coming up on the right or a left turn was coming, I'd probably have to take over. Outside of that, it was really reliable and I let it go. I had maybe 30 drives without an intervention, but 11.4.3 is a completely different story for me. There have been at least 5 instances where it almost drove directly into another car. The first on this one, on a right turn, it almost happened, because I honestly thought it would continue to turn and I wouldn't have to worry. If I didn't disengage, it would have plowed right into the side of a car at a stop sign. It was crazy, it entered the turn pretty quick, around 30 mph, starting turning the wheel and then just stopped turning, but didn't slow down enough.

The driver at the stop sign honked and looked terrified. I slammed on brakes and finished the turn, but it was jarring.

 

[Supcom]

I agree with this. Honestly on 11.4.2, I knew that anytime a turn lane was coming up on the right or a left turn was coming, I'd probably have to take over. Outside of that, it was really reliable and I let it go. I had maybe 30 drives without an intervention, but 11.4.3 is a completely different story for me. There have been at least 5 instances where it almost drove directly into another car. The first on this one, on a right turn, it almost happened, because I honestly thought it would continue to turn and I wouldn't have to worry. If I didn't disengage, it would have plowed right into the side of a car at a stop sign. It was crazy, it entered the turn pretty quick, around 30 mph, starting turning the wheel and then just stopped turning, but didn't slow down enough.

The driver at the stop sign honked and looked terrified. I slammed on brakes and finished the turn, but it was jarring.

I expect to see a lot of these sorts of anomalies as Tesla furthers the transition of the planning software to neural networks.

 

[Eddie123]

I agree with this. Honestly on 11.4.2, I knew that anytime a turn lane was coming up on the right or a left turn was coming, I'd probably have to take over. Outside of that, it was really reliable and I let it go. I had maybe 30 drives without an intervention, but 11.4.3 is a completely different story for me. There have been at least 5 instances where it almost drove directly into another car. The first on this one, on a right turn, it almost happened, because I honestly thought it would continue to turn and I wouldn't have to worry. If I didn't disengage, it would have plowed right into the side of a car at a stop sign. It was crazy, it entered the turn pretty quick, around 30 mph, starting turning the wheel and then just stopped turning, but didn't slow down enough.

The driver at the stop sign honked and looked terrified. I slammed on brakes and finished the turn, but it was jarring.

Yes, I agree. I had something similar while waiting for right turn on red, it decided to take a chance even when an approaching car sped through the intersection.

 

[Eddie123]

Correct. It's not supposed to be ready for anything but testing. You are testing the software, not using it, and certainly not relying on it.

I suspect that people are letting their guard down with the most recent versions of FSDb because it's actually starting to work. Or, at least, giving the impression that it should work. Tesla has worked on the smoothness, and the software is getting more confident, and I think that has gotten some drivers to let down their guard. When the software makes bad choices, it's particularly jarring.

This may be the most dangerous stage of testing for the software. Though it could get progressively more dangerous as it gets progressively more competent.

I understand your logic. I don't mind the need to override due to delays in lane changes, hesitation on right turns, and the extra time it takes to process the sensor inputs. The underlying issue I see is the inability to obey traffic laws. Over the last 5 software upgrades, the car made dangerous moves including crossing double solid white lines, making right turn from the 2nd lane, crossing into opposing lanes on left, etc. I believe that no software or firmware should override these set boundaries unless it sees imminent danger.

 

[JB47394]

I believe that no software or firmware should override these set boundaries unless it sees imminent danger.

But that's the point here; the developers have to build the software to first understand those set boundaries, identify imminent dangers, and act on the whole thing in some useful fashion. They don't have that figured out completely. They have parts of it, which is why FSDb can sometimes drive a car around town without intervention. They don't have others, which is why FSDb can sometimes make an unsafe move or overreact to a shadow on a road.