I remember that, I also remember having to confirm for it to continue through, and I think it would chime for that. Maybe I'm remembering scenarios where I confirmed and then the light changed to yellow afterwards. It's also possible there was another version that allowed it to continue without a leading vehicle, but outside of FSD beta. One final possibility is that earlier versions of FSD beta were better at it, but that seems unlikely unless they were using the TACC version with programmed timers and switched to an ML version. Regardless, I'm definitely certain that there was a time when it consistently was spot on, deciding to stop at the last moment or clearing the intersection before the light went from yellow to red in scenarios where I would have been hard-pressed to judge appropriately. That was months ago, probably prior to 2022.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Wiki MASTER THREAD: Actual FSD Beta downloads and experiences
- Thread starter jebinc
- Start date
I was on TACC yesterday and it seems the same. I had to use the stalk to continue at a light but it also followed a car through it. I also felt that on TACC it was doing a better job at stop signs which for me means a slow down further out instead of running up to the stop sign and hard braking.
sleepydoc
Well-Known Member
It's hard to say - back before I was on FSDb I had stoplight detection on for TACC and It was perfect; I never had it get the signal wrong. Even with FSDb I never had an issue before the current version (10.12), but even now I've only had it try to proceed through a red light once, maybe twice.
The decision whether to proceed on a yellow is a completely different issue. Nominally, FSD should be pretty accurate at calculating the distance and timing relative to the car's speed, but even then it depends on how long the light stays yellow and how much deceleration is 'reasonable.'
Lights in Europe turn yellow before turning green but I've never seen a light in the states do that.
Had an anti Dan O’Dowd event this morning. Beta did a full on emergency stop for a J-waker.
I didn’t even see him but as he crosses he is shaking his head and didn’t really intend to cross in front of me also appeared to be angling to go behind me. I guess Beta would have ignored stopping and went for him if he was a kid.
I didn’t even see him but as he crosses he is shaking his head and didn’t really intend to cross in front of me also appeared to be angling to go behind me. I guess Beta would have ignored stopping and went for him if he was a kid.
AlanSubie4Life
Efficiency Obsessed Member
Plenty of these events.Had an anti Dan O’Dowd event this morning. Beta did a full on emergency stop for a J-waker.
The point of course is that it won’t always work. For example, see Chuck’s Memorial Drive drive yesterday where it stopped late for a pedestrian (after the crosswalk), because it never seemed to even visualize the guy until he was in the crosswalk.
Obviously plenty of other cases too. Just have to have the right tricks to defeat the system. Some might even work on humans (most won’t of course).
sleepydoc
Well-Known Member
But the real point is currently, the point of any AEB system is to supplement the driver and by definition, any case where it succeeds is a case that the driver missed and would have had an accident (or a worse accident.)
If/When FSD is truly autonomous it needs to be judged differently. Where the bar should be is open to debate. Logically, it just needs to be as good as the average human but we all know that won't suffice. If doesn't matter if it catches 100 cases that humans would miss; as soon as it misses a case a human would have caught the system will be crucified in the public square, because that's the level of rationality we have right now.
colettimj
Member
MarcG
Active Member
Yep got this on Monday, so far haven’t noticed any improvements
PACEMD
Active Member
Improvements? You mean software updates are supposed to include improvements? I did not know that.......
old pilot
Member
sleepydoc
Well-Known Member
I had an ‘interesting‘ experience on the way home from work a few days ago - I entered ‘home’ as my destination while in the parking ramp. As I exited the ramp, the car’s position and orientation on the map was clearly off, presumably because it couldn’t get a signal in the ramp and it takes a bit after exiting.
I engaged FSD right as I exited the ramp. It proceeded to drive straight through a stop sign. I disengaged and stopped at the last second, then reengaged Only to have it barrel towards a red light. I disengaged again and as I was waiting for the light, the GPS presumably found a signal and oriented itself and FSD worked normally for the remainder of the trip.
I assume this happened because GPS was wrong and there was discordant data between the cameras and the GPS. I still find it odd and troubling that it would let GPS totally overrule the cameras, though.
I engaged FSD right as I exited the ramp. It proceeded to drive straight through a stop sign. I disengaged and stopped at the last second, then reengaged Only to have it barrel towards a red light. I disengaged again and as I was waiting for the light, the GPS presumably found a signal and oriented itself and FSD worked normally for the remainder of the trip.
I assume this happened because GPS was wrong and there was discordant data between the cameras and the GPS. I still find it odd and troubling that it would let GPS totally overrule the cameras, though.
Agreed. The master-decision-maker is sometimes odd to me too. We've seen it make turns the opposite way to the displayed map route and sometimes the opposite way to the turn signals. Not even a spur of the moment decision, it even gets into the wrong turning lane to set up the wrong turn
We've also seen before where it shows the car on a different road because of GPS error and seems to drive accordingly, not to what it's seeing. And sometimes it sees things, displays them, but tries to drive through them anyway.
Yes, It's odd.
sleepydoc
Well-Known Member
Just to clarify, this was with 10.12, not 10.69. I just updated to 69 so I'll have to see if it happens again with 69.
valem
Member
I have actually noticed regressions with 10.69... roundabouts in my neighborhood were handled smoothly with 10.12, now 10.69 keeps randomly braking in the roundabouts.
Phantom braking in general around residential streets seems to be more frequent.
It's now ignoring speed bumps that it used to recognize before.
In my local plaza parking lot it used to slow speed to 15mph, now it keeps the main road speed of 30mph when entering the parking lot. GET OUT OF MY WAY!
Also driving MUCH closer to parked cars than 10.12 was.
So far (to me) it seems like the only improvement is the creep boundary, and the unprotected left turn, stop in the median.
I have only driven in my neighborhood since 10.69, but so far it is handling the neighborhood much more poorly compared to 10.12
old pilot
Member
Did a second short drive today using 10.69.2 and was pleased with its behaviour. Of special note, to me, was the car moved out into the intersection when making an unprotected left. I was taught to do this, especially when at an intersection on a two lane road with heavy traffic in both directions and no turning lanes present. By pulling into the intersection you are letting others know you are committing to make the turn either when there is a sufficient break in traffic or after the light changes. This maneuver also shortens the time and distance needed to make the turn which allows for smaller breaks in traffic to complete the turn as well as allowing those behind you who are going straight to move to the right and get around your car and keep traffic moving. I give a +1 to Tesla for this positive change.
old pilot
Member
Did a third drive today. At an intersection where the car should make a right turn as per nav data and blue arrow and voice announcement, it instead decided to go straight through the intersection just like 10.12 did, so -1 for tesla. It is still doing unprotected left turns correctly. I'm really beginning to believe it has something to do with the intersection being at an oblique angle.
Similar threads
