Have you seen the thread with all the people using safety defeat wheel weights? All the technology can't fix stupid. People will just plug in their seatbelts and then sit on top of it. On the plus side, it gives us the Darwin Awards to watch.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Why are TACC, AP (and ?FSD) so bad?
- Thread starter mspohr
- Start date
-
- Tags
- ap fsd phantom braking tacc
Yeah there are other realities behind why this or speed governors aren't mandated. It's possible to fool eye tracking too but the regulators also just work to make things so onerous that fooling it isn't worthwhile because it requires something like this
Organizations like the NHTSA and NTSB of course don't just say you can't fix stupid, because their focus is protecting other people who responsibly use public roads and shouldn't be exposed to risks created by others being stupid -- if they want to build their own private road and be stupid on it, that's their prerogative.
There's a well-documented way to reduce the contributing factors
#1 Drive a Tesla
#3 Use FSD
That dramatically increases safety.
Any idea what would be the average millions of miles driven per one accident while using regular dumb cruise control?There's a well-documented way to reduce the contributing factors
#1 Drive a Tesla
#3 Use FSD
That dramatically increases safety.
From what I've heard, getting those numbers is impossible, as cars just don't report them.
But I'd guess that they may actually decrease safety, as they'll allow a car to run off the road or into another one easily.
Tesla collects a lot of data that allows them to provide this metric, which is great. And I think Teslas do have a lot of good design elements that lend to safety, as do EVs as a whole particularly in outcomes with a lower center of gravity etc.
I also think comparing accident rates via Autopilot miles to all miles driven is not at all apples to apples, the nature of highway driving blows up the miles driven per accident. And this is also why Teslas overall have so much more comparable rates of accidents than just Autopilot alone.
Not only road type but we can talk about how people only activate assist systems under certain conditons. For example, people are less inclined to activate cruise control in inclement weather or winter with icy roads, so that’s selection bias. People activate cruise controls when they expect to do large amounts of boring, monotonous driving and staring out the windshield, and that’s why cruise control exists in the first place.
E90alex
Active Member
Not really a great comparison without limiting for similar model years vs Tesla and other vehicles with similar ADAS technology.There's a well-documented way to reduce the contributing factors
#1 Drive a Tesla
#3 Use FSD
That dramatically increases safety.
Teslas are all basically 2013 and newer, and that data will skew even newer than that due to significantly more 3/Y on the road than any other Tesla model. Newer cars tend to be in better condition and therefore better able to evade an accident (better tires, suspension and brakes etc).
Cars with competing ADAS tech like lane departure warning, forward collision warning, active lane keep assist, adaptive cruise control, blind spot monitoring, rear cross traffic alert, automatic emergency braking etc etc will also reduce accidents.
zoomer0056
Active Member
. . . much of anything. The news article was very deficient. The only thing we know from the news article is that an accident occurred.
For safety comparisons, maybe try and compare cars that are in the same price range as well as age. Most cars in the tesla price range also include a lot of safety features and are probably statistically much better than the average. Or maybe compare to the safety of all cars that can go 0-60 in under 4 seconds?
And with driving the same mix of miles... AP miles are almost exclusively highway, and the majority of accidents happens outside highways. It's hard not to look good when Tesla compares this way.
Last edited:
No, that's a much simpler world of single camera, single frame AP highway. Definitely not FSD. Where is the equivalent FSD chart?There's a well-documented way to reduce the contributing factors
#1 Drive a Tesla
#3 Use FSD
That dramatically increases safety.
It be interesting to see how FSD's trend progresses once the single stack is on the road. For now FSD's disengagement rate is too high.
Yet the early numbers for FSD beta also look good, and that is purely off-highway driving. You are correct, of course, that comparing apples to apples is very hard in these cases. However, it's worth noting that it's pretty well documented that the majority of accidents are fender-benders on city streets, which are rarely recorded at all (no surprise there I think). The accidents that make it into the statistics tend to be major ones such as T-bones at red lights or accidents on freeways where the higher speeds make them more sever and hence reportable. Assuming this to be the case then I think the Tesla numbers probably are representative of a significant reduction in reportable accidents, though perhaps not by the margins the raw data might suggest. Meantime, we should all watch for better and more rigorous statistics, and not fall into the trap of jumping to conclusions based on individual incidents (as the press would love everyone to do, cause, well, that sells!).
People tend to be very bad at relative risk .. everyone knows a friend who is nervous of flying, even though the drive to the airport they make with no worries is FAR more dangerous, statistically speaking, then the actual airplane flight. Meantime the press screams about every single ADAS indecent, real or imagined, (Andew J Hawkins at The Verge comes to mind) as if the world has ended.
Daniel in SD
(supervised)
Tesla has the data but doesn't report it. I want to see collision rate with TACC alone (they hardly sold any cars with dumb cruise control).
FirstInTown
Member
Took a trip with latest non FSDb software , so TACC cruise on interstate this weekend. Latest software installed on Friday. Lots of sun, clear roads. Had 6 Phantom Brakes in 4 hours of driving. 2 caused by overpass shadows, so that still exists.(Shadow Unaware Cruise Control System, or S.U.C.C.S. as we now call it) Other 4 for no reason I could see.
And one time is was so abrupt, slowed from 70 to 55 NOW, lucky no one behind us as wife even said "Jez! Thanks Elon" as the seat belt locked up.
Again, NONE of our others cars with Adaptive cruise do this! Especially at a rate of more than 1 fake brake per hour. So still crap.
And one time is was so abrupt, slowed from 70 to 55 NOW, lucky no one behind us as wife even said "Jez! Thanks Elon" as the seat belt locked up.
Again, NONE of our others cars with Adaptive cruise do this! Especially at a rate of more than 1 fake brake per hour. So still crap.
Yes FSDb. Just because the car passes a thousand speed limit signs and the car changes the speed limit in the car every time does NOT mean it "READ" the sign. That data is not good enough when it is also known that the car uses map data because now you have to be able to separate those two conditions.
Yeah, and when were those tests done? As far as I remember that was as soon as Tesla applied visual sign reading to the MODEL S using the Mobileeye and AP1 hardware. Show me a current test, something in the last 3 years on a model 3.
Your theory can be thrown into chaos by passing just one sign and the car doesn't change the speed limit accordingly. It doesn't mean that there wasn't some other reason, but now you have to figure that out.
My data shows that I have two different physical signs that the car does not react to in any way that the user can see. It doesn't change the round speed limit icon on the screen, it does not slow the car down when in AP(FSDb), AND it doesn't visualize the sign AT ALL on the display. So what is your theory on that?
The ONLY way to really attempt to test this is to get a stretch of public road and toss a speed limit sign in the ground that is say 5mph less than the road speed limit and see if the car changes the speed. I say that a Model 3 will NOT visualize that sign and will not react and apply that speed limit. I would be ecstatic to see someone carry out this test and prove/disprove me. Yes people did this in the past, but it only applied to the MODEL S on AP1.
OxBrew
Active Member
Non FSD, I drive by the same sign in AP where it goes from 65 to 50 all the time. If a truck is on my right and blocks the sign it stays at 65. If it sees the sign, it drops to 50. Just sayin.
(and it show the sign if it sees it, not if it does not)
Last edited:
zoomer0056
Active Member
Does the MPH drop like this while using TACC and does the car adjust speed?
My Feb 2022 MSLR (does not have FSD and is running 2023.2.12) does not acknowledge a different speed limit on a sign when TACC is on. However, with TACC off, the new sign is read and acknowledged. I have to reset TACC engagement (turn it off then back on) for the car to set the new speed.
OxBrew
Active Member
I can't say for sure, but I think the behavior is the same on TACC and AP. 2022.44.25.3, and not updating until I need to take it in for service, this version is working great for me, no PB ever, at all.
This is nonsense. Tesla didn't report non-highway numbers or any FSD-beta metrics / KPI:s. The main reason why the City Streets beta isn't in even more accidents is because it's too unreliable so far. Seems stuck at <10 miles / disengagement and <2 miles per intervention. If this improves to 10x accidents will likely skyrocket due to automation complacency and overtrust.
Similar threads
