In practice a certain amount of checking out on aircraft autopilot is arguably tolerable; the operator generally has _some_ time to react to something happening. Cars aren’t like that.
Aviation is an exception because everything has autopilot, and so it is MUCH easier to predict what others on the road are doing. If every car is self driving and can communicate, then things like random braking etc aren't an issue, cars can communicate stuff like this in advance.
Planes also have the advantage of not having to deal with pedestrians, bicyclists and other moving obstacles that will never be part of this hypothetical network.
There are always going to be objects on the road that aren't "self driving" and can't communicate. Like pedestrians and cyclists.
Only a subset of aircraft have autopilots, and many autopilots have very limited functionality. You can't just set a destination airport and go to sleep. The pilot is actively flying the aircraft at all times, even if they're not directly actuating the flight controls.
Marine traffic and aviation are very different environments from where cars operate. You do have lot more reaction time, lot less traffic and hazards are often static and visible long way away. And in scenarios where they are not, well I hope lot less blind trust is put in.
Commercial aviation is a little different in that on an IFR flight plan (which encompasses pretty much the whole time the autopilot is active), ATC is maintaining a 3 mile lateral bubble around the aircraft, making sure that there are no conflicts. So even though the pilot may not be as engaged in see-and-avoid behavior (which is the primary responsibility during VFR), there should be no risk of mid-air collision.
Aviation practices are very much about ensuring that there are extra layers of protection that should all prevent an accident, that even if one fails the others continue to maintain safety. ATC, autopilot, pilot, TCAS, etc. Contrast this with automotive practices, which seem to be about achieving the minimum amount of safety possible.
> I realize commerical aviation is a bit of an exception
If flying in free air had the same number of variables as driving on oddly shaped roads filled with oddly shaped vehicles and obstacles and the occasional people crossing where they shouldn't, we'd have an air crash every hour. A flock of pigeons usually isn't fatal for an airplane engine, now imagine flying surrounded by people testing their new jet backpack.
I'm 100% convinced we'll send someone to Mars a lot before we'll have safe self driving vehicles.
First I agree the driver shouldn't have been on their phone no matter what.
I think before passing judgement though, I would like to know how a human would have performed. Aviation investigations often recreate scenarios in a simulator and then have other pilots fly them. I'd imagine something similar could be done for cars.
FSD is pretty good at motorcycle detection and has the advantage of cameras looking in all directions.
But..
It has been here for many years, overpromising, under delivering at full price. And all it did was ride the coat tails of the tonystark "still relevant"fantasy of men everywhere. No,the world is already carved up, nobody gets a slice by wits alone unless already owing a bone. Meanwhile waymo quietly walked the walk..
2024 Q1: “we recorded one crash for every 7.63 million miles driven in which drivers were using Autopilot technology. For drivers who were not using Autopilot technology, we recorded one crash for every 955,000 miles driven. By comparison, the most recent data available from NHTSA and FHWA (from 2022) shows that in the United States, there was an automobile crash approximately every 670,000 miles.” [0]
And for further info that includes a comparison to Waymo see [1].
Doesn’t seem right to call something that is 10x safer “terrible”, but maybe you have different metrics you are looking at.
> that is 10x safer “terrible”, but maybe you have different metrics you are looking at.
Of maybe your “metrics” are just slightly dishonest?
Presumably you can only use FSD under conditions where accidents are already significantly less likely to happen (per mile driven) like freeways etc?
Which makes this an apples to oranges comparison.
Turning “FSD” off just before an accident is likely to happen without necessarily giving enough time for the driver to react would also be very helpful (although I’m not sure how the cases would be recorded specifically )
> Doesn’t seem right to call something that is 10x safer “terrible”, but maybe you have different metrics you are looking at.
Quick question: are there roads or conditions Autopilot doesn't do well on, so can't or won't be activated?
I don't see Autopilot working well in a bitter New England winter storm at night. Or a torrential Florida downpour.
That's why those metrics suck. And Tesla knows. They compare "all drivers, all conditions" to the "idealist, best case subset of miles driven by Autopilot".
> I would like to know how a human would have performed
Well, his vehicle was the only one on that busy highway that afternoon that "lurched forward while accelerating" and crashed into the motorcycle in front of him, so I think we have the answer to your question.
I think it'd also be worth comparing it to cars that have more basic collision-avoidance.
I don't expect Tesla's FSD to be perfect. But I do expect it to not be worse than the simple and effective collision avoidance that nearly all new cars have.
Exactly, IMO the point of autopilot is to be a passenger in your own car with a guarantee that it won't cause and accident. If you have to be ready to take over, then what's the point?
And yesterday a human driver rear-ended and drove over an acquaintance who was on a motorcycle.
Somehow didn't see the motorcycle break. Not a Tesla.
Fortunately only broken bones and material damage.
That said I think it's shocking that Tesla hasn't been slapped hard by authorities for calling their current system "full self driving".
It might be the hardware is FSD-ready once they figure out the software, but as long as it requires supervision it just should not be called "full self driving".
This video ignores how bad humans are at seeing motorcycles. Every motorcyclist knows to assume they are invisible to cars aka cagers.
It's likely that the FSD system is better than a human at seeing and avoiding motorcycles but that's not a high bar.
Why is the relative risk not considered in a video? As a motorcyclist are you safer with a Tesla behind you with FSD enabled or with a human driver behind you?
How well do traditional adaptive cruise control systems, that use radars and which the likes of Mercedes-Benz and BMW have been using for a couple of decades, detect motorcycles ahead of the car?
I sometimes give the fsd a try for just keeping the car in lane. It's so wobbly and uncertain and does weird choices all the fucking time.
Based on my experience, I think people who use the full fsd for driving longer trips should definitely lose their license to drive. Perhaps some day the system will work reliably, but not today.
So much FUD whenever an article is about Tesla. As per the article -
"the driver told a trooper that he was using Tesla’s Autopilot system and looked at his cellphone while the Tesla was moving."
Does this mean that it wasn't actually FSD and was just Autopilot? The article, which mentions how people confuse Autopilot and FSD, seems to confuse them itself.
39 comments
[ 2.9 ms ] story [ 83.2 ms ] threadPeople completely check out, like this person, playing on their phone. And then an avoidable accident happens, this one being fatal.
Might be better to have the autopilot as the copilot and not the other way around.
Only a subset of aircraft have autopilots, and many autopilots have very limited functionality. You can't just set a destination airport and go to sleep. The pilot is actively flying the aircraft at all times, even if they're not directly actuating the flight controls.
Aviation practices are very much about ensuring that there are extra layers of protection that should all prevent an accident, that even if one fails the others continue to maintain safety. ATC, autopilot, pilot, TCAS, etc. Contrast this with automotive practices, which seem to be about achieving the minimum amount of safety possible.
If flying in free air had the same number of variables as driving on oddly shaped roads filled with oddly shaped vehicles and obstacles and the occasional people crossing where they shouldn't, we'd have an air crash every hour. A flock of pigeons usually isn't fatal for an airplane engine, now imagine flying surrounded by people testing their new jet backpack. I'm 100% convinced we'll send someone to Mars a lot before we'll have safe self driving vehicles.
I think before passing judgement though, I would like to know how a human would have performed. Aviation investigations often recreate scenarios in a simulator and then have other pilots fly them. I'd imagine something similar could be done for cars.
FSD is pretty good at motorcycle detection and has the advantage of cameras looking in all directions.
No buts,no ifs,just a stif
And for further info that includes a comparison to Waymo see [1].
Doesn’t seem right to call something that is 10x safer “terrible”, but maybe you have different metrics you are looking at.
[0] https://www.teslarati.com/tesla-releases-vehicle-safety-repo... [1] https://www.nextbigfuture.com/2024/07/tesla-autopilot-fsd-an...
Of maybe your “metrics” are just slightly dishonest?
Presumably you can only use FSD under conditions where accidents are already significantly less likely to happen (per mile driven) like freeways etc?
Which makes this an apples to oranges comparison.
Turning “FSD” off just before an accident is likely to happen without necessarily giving enough time for the driver to react would also be very helpful (although I’m not sure how the cases would be recorded specifically )
Quick question: are there roads or conditions Autopilot doesn't do well on, so can't or won't be activated?
I don't see Autopilot working well in a bitter New England winter storm at night. Or a torrential Florida downpour.
That's why those metrics suck. And Tesla knows. They compare "all drivers, all conditions" to the "idealist, best case subset of miles driven by Autopilot".
Check out the guidelines:
Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes.
Well, his vehicle was the only one on that busy highway that afternoon that "lurched forward while accelerating" and crashed into the motorcycle in front of him, so I think we have the answer to your question.
I don't expect Tesla's FSD to be perfect. But I do expect it to not be worse than the simple and effective collision avoidance that nearly all new cars have.
Somehow didn't see the motorcycle break. Not a Tesla.
Fortunately only broken bones and material damage.
That said I think it's shocking that Tesla hasn't been slapped hard by authorities for calling their current system "full self driving".
It might be the hardware is FSD-ready once they figure out the software, but as long as it requires supervision it just should not be called "full self driving".
This continues to be an excellent video.
For the same reason it viewed railroad crossings as a weirdly broken red traffic light with intermittent trucks.
For the same reason it still hits stationary emergency vehicles with lights going.
Because the software is nowhere near ready.
It's likely that the FSD system is better than a human at seeing and avoiding motorcycles but that's not a high bar.
Why is the relative risk not considered in a video? As a motorcyclist are you safer with a Tesla behind you with FSD enabled or with a human driver behind you?
Based on my experience, I think people who use the full fsd for driving longer trips should definitely lose their license to drive. Perhaps some day the system will work reliably, but not today.
"the driver told a trooper that he was using Tesla’s Autopilot system and looked at his cellphone while the Tesla was moving."
Does this mean that it wasn't actually FSD and was just Autopilot? The article, which mentions how people confuse Autopilot and FSD, seems to confuse them itself.