It seems like they made a lot of untruths and are selectively correcting them when they have no other choice but to correct them.
-
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
CyberGus
Not Just a Member
The real problem with this test was the safety cones, which prevented the car from swerving.
SAVE THE CHILDREN, BAN THE MURDER CONES
SAVE THE CHILDREN, BAN THE MURDER CONES
How do you feel about having that factoid in your google search history…..
Tam
Well-Known Member
I'm glad that you found a solution to drive like the case of NCAP kid mannequin tested at 30Km/hr or 19MPH but this time not on a test track but in 85MPH speed zone rural highway.
Next example is sun glares. My FSD and FSD beta consistently shut down with no prior warnings (that means to manually take over immediately) in early morning or evening when facing the sun. The only way to get the function back is to turn north/south.
LIDAR and 4D radar will take over seamlessly in sun glare problem.
Tam
Well-Known Member
No. But at least it's less dangerous when driving at 85MPH zone without experiencing Tesla red steering wheel icon.
Waymo set their cameras at different angles and elevations to solve the sun glares issue.
The driver's isn't. It says that he used FSD as the instructions say for the entire test, but then he says that he never had his hands on the wheel. (The instructions require hands on the wheel, correct?) Then there is the time the the car told him to take control immediately and he didn't. So he lied in his affidavit.
Daniel in SD
(supervised)
Tesla's cameras aren't even close to state of the art. What percentage of human caused accidents are perception errors? My only point is that if you could achieve human level perception using only cameras then you could achieve far greater than human performance which would be fine for L4. Right now that technology doesn't exist so radar and lidar are good solutions until it does.
Tam
Well-Known Member
Agreed! Tesla software and hardware are sub par. That's the heart of the problem. Owners might think it's the state of the art like the dude who drove from the back seat.
AlanSubie4Life
Efficiency Obsessed Member
Admittedly I only read the part of the affidavit that was interesting to me, but it says: “
During the June 21, 2022 test drive, I brought the 2019 Tesla Model 3 with FSD Beta
10.12.2 up to a speed of forty (40) miles per hour and then put it in full self-driving mode,
in the manner instructed by the owner's manual, once the car entered the lane defined by
the cones.”
That seems correct. He did put it in FSD as instructed, with a double pull on the gear stalk.
He specifically said that he kept his hands off the wheel, which is not in keeping with the overall FSD instructions, so I doubt very much that any claim about using FSD correctly was made!
It’s possible that there are pieces of the affidavit that are wrong that I have not read. I was most interested in seeing how many trials they did, so only read a few sections. (In the affidavit they do not say how many trials were done, of course.) All they say is: “
During three (3) of the tests, the 2019 Tesla Model 3 with FSD Beta 10.12.2 was put into
full self-driving mode and it hit the mannequin three (3) times.” So no mention of how many trials were done.
Edit: went to the prior page, which I had not read, to see if I missed anything, and once again, there is just this statement, which makes no claim about FSD actually being used correctly. It is only a statement about the mode the car is in, not whether it is being used correctly. Very carefully worded!
“The 2019 Tesla Model 3 with FSD Beta 10.12.2 was in full self-driving mode, in the
manner instructed by the owner's manual (as provided in Exhibits A & C) during all
portions of the June 21, 2022 video shoot where the car was being tested to see if it
recognized the child-size mannequin in its path.”
Last edited:
Is your confidence and fact the cameras and hardware are subpar based on there being better options available or that you have facts that they don’t work? There will always be better computers and cameras Forever.
Tam
Well-Known Member
Please take a look at the clip on message 17:
FSD fails to detect children in the road
(Not to be confused with green poles, school buses, and guys on motorcycles.) https://stocks.apple.com/A4xrG0pGKT820ZC7m-MixWA
teslamotorsclub.com
Better hardware/software means better desirable results.
So by your fact this has nothing to do with software it’s for certain subpar cameras and hardware.Please take a look at the clip on message 17:
FSD fails to detect children in the road
(Not to be confused with green poles, school buses, and guys on motorcycles.) https://stocks.apple.com/A4xrG0pGKT820ZC7m-MixWA
Better hardware/software means better desirable results.
Does Tesla Full Self-Driving Beta Really Run Over Kids?
As expected, a contrived, pointless, and terribly dangerous test by Whole Mars Catalog using real kids in the street. Their 'conclusion' is that because they didn't run over kids that FSD Beta is proven safe. Go ahead, click on their video and give them views, they aren't going to stop being stupid anyway.
Why is driving around stationary 'kids' in the street considered a good thing?? Do they really think that objects remain stationary? I'll bet that if the object moved towards the car as it sped around them, it would not further avoid them.
EuroNCAP test aeb seriously, and Thatcham research also. This can be done properly.
The most interesting thing here is that FSD beta does not seem to have acceptable aeb capability. I would love to see more tests so one can find the limits. Certainly aeb can not always stop depending on the circumstances.
Anyway, I question Teslas development priorities for FSD development. I think the basis they should solve is best in world aeb capabilities. Why would the not fix that before for example "Chucks turn"?
The most interesting thing here is that FSD beta does not seem to have acceptable aeb capability. I would love to see more tests so one can find the limits. Certainly aeb can not always stop depending on the circumstances.
Anyway, I question Teslas development priorities for FSD development. I think the basis they should solve is best in world aeb capabilities. Why would the not fix that before for example "Chucks turn"?
AlanSubie4Life
Efficiency Obsessed Member
The prime directive is to proceed with no disengagements.
As you say, they did not try nearly hard enough (a good thing, with real kids), or try enough scenarios! They have to do what Dan O’Dowd did (look for a way to make it fail) to get a failure! (Or just do a lot of trials - for all we know in O’Dowd’s tests they did 300 tries - and 1 in 100 would be a terrible result (for a fully autonomous system) of course.).
Anyway, no surprises here; Tesla says the system will hit children, and they are correct.
I’m not sure how flaggers in construction zones will feel about this test. Of course FSD should not be used there.
The main problem is that people don’t seem to understand (or just keep forgetting or are told otherwise) that FSD Beta does not drive the car. There’s a only one driver - the person behind the wheel.
Last edited:
Whole Mars Catalog seems to be under the impression that his weekend test has recreated Dan O'Dowd's test and given opposite results. He suggests that Dan O'Dowd's test was falsified and designed to show Tesla failing.
What Whole Mars has done is an entirely different test. If he wanted to negate Dan O'Dowd he needed to have done the same test. Dan O'Dowd's test was at least conducted on a closed track with a dummy manikin, but WMC's test was done on an open public street with humans (adults & children) and manikins. The tests were not remotely the same.
What Whole Mars has done is an entirely different test. If he wanted to negate Dan O'Dowd he needed to have done the same test. Dan O'Dowd's test was at least conducted on a closed track with a dummy manikin, but WMC's test was done on an open public street with humans (adults & children) and manikins. The tests were not remotely the same.
| Dan O'Dowd's Test | Whole Mars Test |
| Closed course - Willow Springs Track | Public residential street |
| Straight course, cones on either side | Straight street, random cars parked on the sides |
| Car speed 40mph | Car starting from 0-15mph |
| Car start distance far away | Car very close |
| Object barely visible at start | Object already visualized by FSD at start |
| No shadows | Heavy shadows |
Whole mars catalog is about a Tesla shill as you can get. I wouldn’t be surprised. You shouldn’t use children to do dumb tests. He should be removed from the beta at the least.
Similar threads
