Hhhh true.
VW's system takes longer to kill someone but therefor kills more. Given their history and production environment it might be something they want to acchieve. No one knows but them.
IMO the code runing in cars (and other types of vehicles) should be made available to inspection agencies, pref. written in languages that can make certain guarantees, or at the very least be required to be written in clear, legible manners with certain standards.
This really feels like a non-story to me. It seems obvious that Tesla would have to create special builds or settings for crash testing.
Their crash avoidance features (automatic braking, steering correction, etc) can't be disabled by the user and it's rather hard to perform, say, a frontal crash test if the car refuses to crash.
This is my feeling as well, but to play devil's advocate; what settings could Tesla tweak to cheat on a crash test? I'm struggling to think what software tweaks Tesla could be implementing during a test that would game the results of the crash.
Airbag/seatbelt tensioner timings? Maybe lowering the threshold for their deployment? Maybe something sort of software-triggered fuse to make the battery safe?
Things like that, yes. It probably takes a certain amount of domain expertise to know exactly what, and how.
But it seems fairly reasonable to guess that certain settings that are normally generalized to work appropriately well across a wide range of crashes can also be optimized for certain cases if you know in advance which cases are being tested for. We see this sort of optimization-to-the-test being possible in most other domains so why not this one as well.
Another commenter [0] highlighted the answer to this.
Crash testing isn't just crashing into barriers, it also tests automatic emergency braking. For crash testing, they could give a software build that's more sensitive to seeing pedestrians and bicycles, but not use that build in customer cars because it's prone to false positives.
> It seems obvious that Tesla would have to create special builds or settings for crash testing.
Isn't obvious that you turn off the "crash avoidance" stuff if you want to test the car body only? Sure maybe the regular driver can't turn it off but I am expecting a Tesla mechanic can turn it off, rather then 10 engineers creating new software builds, deploy them on the cars and then ensure that when is the time to test the "Crash avoidence" you use the car with the correct build.
This is interesting because ANCAP stated the following
>The autonomous emergency braking (AEB) system is capable of detecting and reacting to pedestrians and cyclists. The AEB system showed GOOD performance in pedestrian test scenarios in both daylight and low light, with collisions avoided or mitigated in most scenarios including in turning scenarios and some reverse (AEB Backover). In cyclist test scenarios, the AEB system offered GOOD performance with maximum points scored.
It wouldn't be difficult for Tesla to just bump up the sensitivity for the test - ensuring that it acts extra cautiously for the test, and then have it lower in real world scenarios where they're worried false positives would make it unusable.
I could completely see Tesla making the system more sensitive to pass tests and then in the real world drop the setting down do that it's not emergency braking due to fall positives.
There have been a few reports of motorcyclists getting rear ended by Teslas cars, which if the ANCAP test was GOOD in cyclist settings and hitting motorcycles in real world scenarios, I can see this being the case.
16 comments
[ 3.4 ms ] story [ 45.3 ms ] threadTheir crash avoidance features (automatic braking, steering correction, etc) can't be disabled by the user and it's rather hard to perform, say, a frontal crash test if the car refuses to crash.
Airbag/seatbelt tensioner timings? Maybe lowering the threshold for their deployment? Maybe something sort of software-triggered fuse to make the battery safe?
But it seems fairly reasonable to guess that certain settings that are normally generalized to work appropriately well across a wide range of crashes can also be optimized for certain cases if you know in advance which cases are being tested for. We see this sort of optimization-to-the-test being possible in most other domains so why not this one as well.
Because then you're doing junk science (and possibly active harm) in all domains in the name of improving economic metrics.
Crash testing isn't just crashing into barriers, it also tests automatic emergency braking. For crash testing, they could give a software build that's more sensitive to seeing pedestrians and bicycles, but not use that build in customer cars because it's prone to false positives.
[0] https://news.ycombinator.com/item?id=32837954
Isn't obvious that you turn off the "crash avoidance" stuff if you want to test the car body only? Sure maybe the regular driver can't turn it off but I am expecting a Tesla mechanic can turn it off, rather then 10 engineers creating new software builds, deploy them on the cars and then ensure that when is the time to test the "Crash avoidence" you use the car with the correct build.
https://auto.howstuffworks.com/car-driving-safety/accidents-...
"A pulley, mounted in a track, pulls the car down the runway. The car hits the barrier"
>The autonomous emergency braking (AEB) system is capable of detecting and reacting to pedestrians and cyclists. The AEB system showed GOOD performance in pedestrian test scenarios in both daylight and low light, with collisions avoided or mitigated in most scenarios including in turning scenarios and some reverse (AEB Backover). In cyclist test scenarios, the AEB system offered GOOD performance with maximum points scored.
It wouldn't be difficult for Tesla to just bump up the sensitivity for the test - ensuring that it acts extra cautiously for the test, and then have it lower in real world scenarios where they're worried false positives would make it unusable.
There have been a few reports of motorcyclists getting rear ended by Teslas cars, which if the ANCAP test was GOOD in cyclist settings and hitting motorcycles in real world scenarios, I can see this being the case.