Been a week so far, it's still terrible.Does it happen consistently for you folks? I've seen similar regressions in past updates (e.g. 2019.5.15 as well as 2019.20.x) for the first few drives, but it seems to improve over time with the same version. Ready to bite my tongue when I get 24.4 and see for myself
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Phantom braking will get a lot worse before it gets better
- Thread starter conman
- Start date
-
- Tags
- autopiliot autosteer braking
Ah, no idea then. I never liked TACC, it’s felt pretty rough to me. Regular autopilot is usually smoother for me on all versions.
Uhh, autopilot uses TACC and just adds autosteer, so there's no separation of the two...
Autopilot is Autosteer on top of TACC (Autopilot = TACC + Autosteer). In other words, when you use Autopilot you activate TACC as well. In fact, you can cancel the Autosteer part of Autopilot without deactivating TACC. When you are in Autopilot, the TACC behavior is no different than if you are only in TACC without being in Autopilot.
Yes, but doesn’t TACC rely heavier on radar? In my experience the two do not drive the same, as in, AP to follow a car vs just TACC to follow a car. TACC will keep distance way more aggressively for me and allow itself to get closer to the car in front when stopping while AP is smooth in more situations.
TACC does rely on radar. Heavier? There is no evidence.
No, TACC uses exactly the same inputs that autosteer uses, and as autopilot matures it is relying more and more on camera.
Yeah, same inputs, but processed / biased differently for sure, else they would drive similarly other than human vs. machine steering. Anyway I try not to use TACC as I prefer AP car-following behavior, but I can see how TACC can exhibit phantom braking more frequently. As a result my experience with TACC in recent versions is limited.
If designed correctly (designed my way); TACC should have a lot less false braking events.
The reason is TACC can make the assumption that the human is paying attention, and the system can solely rely on radar data that ignores stopped objects. I don't want TACC to attempt to be smarter than the bare minimum that it needs to be. It really only needs to use the cameras in corners (to make sure it's not reacting to a car in the other lane).
Where with AP one expects the car to take more responsibility of reacting to it's surroundings as after all it's taken over steering. So it has to assume the human oversight might be lacking. I expect AP to try to be smart even if there might be some false braking. As long as it's not sever false braking it's okay.
That's generally what I want out of those two systems where NoA adds a navigational aware decisions to so it can do autolane changes.
In my experience however the smoothness of adaptive cruise control has more to do with the SW update than which mode I am.
I've had SW updates that were really quite smooth over hundreds, and hundreds of miles and I've had SW updates that were absolutely ridiculous. Things like slowing down by 1-2 mph after passing someone even when no one was in front of either the lane I was in or the lane I'm in now. Or times it tried to be telepathic about what the car in front, and next to me was going to do.
I really wish Tesla would optimize for smootheness for TACC. Where we had two settings. One setting for minimum distance, and one for maximum distance. Where the car simply tried to stay within that range. Like if a car suddenly got in front of me it wouldn't immediately slow down (unless minimum wasn't met), but would simply slow down slightly to go back to the middle. Something like that.
By the way, I ONLY use autosteer and never use TACC by itself, and have furious phantom braking issues.
Just as phantom braking has improved dramatically in V10, we now see this happen so be prepared for a new round of phantom braking to get worse as AEB failed to detect pedestrians in this test:
Tesla Model 3 Failed To Save Dummies In AAA Emergency Braking Test
Tesla Model 3 Failed To Save Dummies In AAA Emergency Braking Test
willow_hiller
Well-Known Member
What I don't understand is that this directly contradicts the findings of Euro NCAP. Official Tesla Model 3 2019 safety rating
They got 5.6/6 for AEB pedestrian test, and 6/6 for AEB cyclist test.
And you can see a video of their AEB tests here. Seemed to stop just fine at 60 kph, which really calls into question the AAA test where the Model 3 never stopped for a single pedestrian dummy.
It is weird isn't it. Those AEB tests look great. One thing I did notice was the model 3 was coming around a bend. My own experience tells me our vehicles are terrible at seeing around bends.
My phantom braking has gotten much better on 2019.32.12.2 , HW3 Model 3 LR AWD . It almost never happens anymore.
diplomat33
Average guy who loves autonomous vehicles
Daniel in SD
(supervised)
It’s much easier to pass a test when you know the questions beforehand. Haha.
AlanSubie4Life
Efficiency Obsessed Member
It is not really weird at all. The Model 3 was designed to pass the Euro NCAP test. Unlike with humans, performance in a specific test using a computer is not easily extended to extremely similar situations with a different color dummy, different shape, different approach angle, etc. So it fails very similar tests with small changes in the conditions.
This is completely normal, expected behavior for such a system.
Remember it only has to work some of the time to improve safety. (With some caveats...
)
SalisburySam
Active Member
Since update to v32.12.2 I noticed no phantom braking events for the first several hundred miles. Then yesterday on a 120-mile trip on I-85 it’s baaaaack. Three events during the trip, three bug reports, and a lot of disappointment as I thought this plague was behind me. It’s not.
That's a good point and reminds me of someone's diesel emissions test (no I'm not equating them on that level at all.) Though it does mean it doesn't extrapolate well to the real world since most humans don't exactly behave like an NCAP dummy.It is not really weird at all. The Model 3 was designed to pass the Euro NCAP test. Unlike with humans, performance in a specific test using a computer is not easily extended to extremely similar situations with a different color dummy, different shape, different approach angle, etc. So it fails very similar tests with small changes in the conditions.
This is completely normal, expected behavior for such a system.
Remember it only has to work some of the time to improve safety. (With some caveats...
Tesla probably tuned it specifically for those tests. All manufacturers do that, and as soon as you present a different scenario it doesn't work so well.
Similar threads
