-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD fails to detect children in the road
- Thread starter 2101Guy
- Start date
Let's hope the entire industry continues to improve, including Tesla. According to IIHS, 144 children were struck by cars in 2020. That number is dropping year after year, thankfully. But it still shows that AEB needs improvement. Computers and sensors are much faster than human reaction time, and can hopefully mitigate collisions and reduce injury or, in a perfect world, avoid the collision all together. We're not there yet, but the trend is promising.
Yes, I’m sure it isn’t. It says so when you buy it and the manual is quite clear on that. Also it would instantly become clear to anyone using it….Are you sure?
Elon Musk says Tesla's Full Self-Driving tech will have Level 5 autonomy by the end of 2021
Tesla's CEO also confirmed that customers won't be able to transfer previous FSD purchases between old and new cars.www.cnet.com
Yes, because Musk pissed on liberals and liberals run the media with a few notable exceptions.
Daniel in SD
(supervised)
Or put your hands on the wheel. I'm guessing people wearing helmets are not part of the training data for the attention monitoring system.
It will be interesting to see if Tesla uses its "hardcore" legal team to sue O'Dowd...
You can see here the expensive professional camera (though they could have used an iPhone and had the whole thing look a lot more professional!)
Operator64
Member
The driver needs a helmet, however the cameraman in the back does not and dont worry about the seatbelts either.
That's the context people are missing.
People assume AEB on any vehicle will stop for children, but that simply isn't true.
AEB systems differ from one system to the next. Some systems are only intended to be a crash mitigation system where they reduce the speed, and I don't think any system has tested well for stopping for kids or pedestrians at night. Thankfully the industry has finally woken up to the need to do standardize testing for pedestrian detection in various conditions.
Tesla mostly differs in they attempt to do fast iteration of SW to push the limits of what it can do. Like they had automatic lane changing before anyone else, and with the FSD beta they did city streets before anyone else.
When you're the one pushing boundaries then you're going to get a lot of push back.
FSD Beta being put in the hands of 100K plus owners was about the dumbest thing Tesla could have done because it was way too premature for that widespread of "testing". They didn't even have things like NoA working properly (for 99.9% of the userbase), and AP still did unsettling things like re-centering in a merge lane.
It opened Tesla up to scrutiny from anyone with a camera, a YouTube channel, and a lot of time. If something like Pedestrian detection works 90% of the time then it gives 10% of the time to take all the videos that make it look bad.
I haven't watched any of the videos from this test because I don't trust the source just like I don't watch FSD videos from Tesla Fanboi's who purposely try to make it look good.
_Redshift_
Member
Oh sure, make this tribal and kill the thread, eh?
Please note that the exact argument can be used for/against. Kindly stay on the subject.
NCAP has done some night time testing and the Model 3 passed:
Though it seems they only do one of the tests at night.
Daniel in SD
(supervised)
Easy to pass a test when you know the exact questions in advance!
I am talking about the constant negative news stories. The fact is the issues we are talking about here, and the other issues that recently came to light have been going on for a long time. But the media has decided to cast a light on them now because of politics. They are true, but the media chose not to cover them then, now they are.
The fact that problems exist is a separate issue from what I was responding too, which is the newcomer topical nature of the complaints; as above. Elon has been saying autonomous driving was a solved problem for at last five years; the media waited till now to call him out on his bullshit. If is what it is.
Tam
Well-Known Member
The problem with the ad was: Its speed was set at 40MPH. Tesla's system did slow down to 22MPH and then gave controls to the driver at the moment of collisions.
NCAP night time was set at 60Km/H or 37.5MPH, but the big difference for that night was: The mannequin was traveling in the same direction as the car. Tesla has more problems with stationary (first responder NHTSA investigation) and cross-path obstacles (2016 Autopilot fatal accident). Still, its chance of stopping for obstacles traveling in its direction is vastly increased.
NCAP kid mannequin was tested at 30Km/hr or 19MPH. The slower the speed, the better chance for an AEB system to work. The NCAP speed was lower than the ad's speed of 22MPH when the system handed the control over to humans.
The ad must adhere to the above factors to prove that Tesla works.
I suspect the way they were able to get it to hit the dummy was by "putting it on rails" by putting the cones up and that if you took the cones away and had just painted lines, or no lines at all, it would have gone around the dummy. But yes, this is a case where Tesla can improve FSD/AEB, but they probably haven't spent as much time working on construction zones yet, since you aren't supposed to be using it in construction zones. (I'm not positive that the FSD beta has that restriction, but it makes sense that its performance is not as good in a construction zone.)
_Redshift_
Member
What a conspiracy theory!
Except, that’s not quite true.
I canceled my subscription to NYT in early 2013 because of John Broder’s false reporting about running out of juice in an early Model S, titled ‘Stalled out on the electric avenue’ or something like that. Some liberal papers had a bias against Tesla even in those days.
Note, at the same time, the conservative media was dead set against Tesla. Until recently.
Like I said, for/against arguments can be made.
Don’t think for one second you are persuading anyone. So, don’t waste your time.
_Redshift_
Member
A contrived test that "proves" the car will hit a child is no more accurate than a contrived test that "proves" that the car won't hit a child.
These tests prove nothing, despite each side claiming "there that proves what we've been saying". It's scientifically and statistically meaningless to draw conclusions this way. Also there are clearly ways that these tests can be falsified.
It is more worrying to see the failures - assuming the test has not been falsified. Not directly hitting things really should be Job One.
Hitting a fake dummy, cardboard cutout, or empty plastic bag on purpose (because the car 'knows' the thing
) is perhaps debatably ok in the far future, but let's start with not hitting all things first.
These tests prove nothing, despite each side claiming "there that proves what we've been saying". It's scientifically and statistically meaningless to draw conclusions this way. Also there are clearly ways that these tests can be falsified.
It is more worrying to see the failures - assuming the test has not been falsified. Not directly hitting things really should be Job One.
Hitting a fake dummy, cardboard cutout, or empty plastic bag on purpose (because the car 'knows' the thing
) is perhaps debatably ok in the far future, but let's start with not hitting all things first.
Last edited:
There’s a difference between a one off article about a particular subject and a series of stories release on multiple platforms on a regular basis about multiple subjects. If you can’t see the difference, well… as they say, ignorance is bliss.
Daniel in SD
(supervised)
The 2019 Autopilot crash was the same with HW2.5.
I would say it proves that FSD is not ready to go out of beta (which of course we all know!). Definitely not ok for a robotaxi to hit objects that large unless there's a really good reason.A contrived test that "proves" the car will hit a child is no more accurate than a contrived test that "proves" that the car won't hit a child.
These tests prove nothing, despite each side claiming "there that proves what we've been saying". It's scientifically and statistically meaningless to draw conclusions this way. Also there are clearly ways that these tests can be falsified.
It is more worrying to see the failures - assuming the test has not been falsified. Not directly hitting things really should be Job One.
Hitting a fake dummy, cardboard cutout, or empty plastic bag on purpose (because the car 'knows' the thing
Similar threads
