If tens of thousands (or really, even a handful) of people were dying from car crashes caused by, for example, brakes failing, then people would absolutely be attacking the manufacturer of the car. If you sell a feature that _does not work_ and it causes deaths, you should be held liable.
Reading the article or the NHTSA documents would clarify that blame is not placed on Tesla here for the crashes, but blame is placed for not ensuring the driver was paying attention. Drivers had plenty of time to react, but they didn't.
This is not trying to say the car quickly jerked into a pedestrian on the side of the road.
This article is about “FSD”/Autopilot and compares what others in the industry do with similar features. This isn’t Tesla autopilot vs Toyota camry drivers.
Other manufacturers limit their features to more controlled environments in motorways. Tesla allows it in a lot more contexts of mixed traffic, road conditions etc.
This is true for BlueCruise and SuperCruise, the only other ADAS'es that market a completely hands-free experience.
However, Lane Keeping and Autosteer systems, like Volvo's DrivePilot, have existed for a while. These usually activate whenever they detect lanes. However, they don't consider VRUs or obstacles in their paths at all. As long as they can detect lanes, they will happily drive you into a wall or a person at speed.
The biggest difference between them and FSDb is that FSD can be activated on any road surface. FSD can make decisions based on the environment it's in as perceived through its cameras (with a very short delay), but it doesn't disable itself when it doesn't know what to do. It tries very very very hard to not ram itself into anything, but it is definitely capable of it if it guesses wrong.
As a Tesla owner, for me Autopilot refers to highway / freeway driver assistance and Full Self Driving ( Supervised ) is for local streets. The only thing FSD provided previously was auto lane changing.
What the hell is with their naming? Self Driving sounds like an absolute and then they add the Full specifier to really drive it home but also add (supervised) meaning you must supervise the super duper full self driving.
>Even the brand name “Autopilot” is misleading, NHTSA said, conjuring up the idea that drivers are not in control.
Ironic how the Verge forms a union of Autopilot and FSD crashes in their headline to further convolute the major differences. As much as we blame Elon for naming Full Self Driving, I also think the media is responsible for tacking all Tesla modes together to get clicks.
For some context, my brother won't even use FSD because the driver monitoring system nags him SO much. It literally tracks his eyes and if he's not paying attention for a few seconds (even glancing out the window, looking down at his phone at all), it has a loud annoying alert.
Feels hard to say their being complacent with their messaging. You still need to pay attention all the time.
I know people are retorting “non assisted drivers are dangerous too, you know” and that’s true. But:
> In 59 crashes examined by NHTSA, the agency found that Tesla drivers had enough time, “five or more seconds,” prior to crashing into another object in which to react. In 19 of those crashes, the hazard was visible for 10 or more seconds before the collision. Reviewing crash logs and data provided by Tesla, NHTSA found that drivers failed to brake or steer to avoid the hazard in a majority of the crashes analyzed.
If an unassisted driver had reactions like they'd be arrested for reckless driving at minimum.
Tesla’s systems are obviously not safe. I suspect naming them things like “autopilot” and “full self drive” really don’t help perceptions and the systems themselves are clearly not able to get a useful gauge of driver attention. They’re overselling ability and under checking driver compliance.
NHTSA has the ability to recall every single Tesla ever put on the road, for any safety reason, including the recent 'warning light size is too small to read'
Therefore, NHTSA also have full authority to shut down FSD, or force Tesla to rename 'Full Self Driving' to something more fitting.
Yet NHTSA allows it to be what it is. If it were truly unsafe, NHTSA would immediately put the brakes on it. (Yes, have a Friday pun.) Before anyone can call FSD not safe, a full investigation needs to occur. These are just the first steps.
Importantly, FSD has had very aggressive cabin monitoring for years, while autopilot gained this feature from a recall sometime last year.
This article notes that the NHTSA will launch a new investigation to see if the new driver monitoring features are sufficient.
Part of the problem with investigations that take many years to complete is that the conclusion is meaningless due to over-the-air updates, as this issue was solved at least last year.
I agree that this was possible with a device to apply force to the steering wheel before they added the cabin camera to autopilot / FSD.
However, since adding cabin camera to autopilot/FSD retroactively to all vehicles that can support it (2018+), stuff like this simply isn't possible without the car kicking you out of FSD or autopilot.
This change would have been made sometime mid-last year for autopilot and far before that for FSD.
> FSD has had very aggressive cabin monitoring for years
When FSD was first released, the steering wheel checks were every fifteen MINUTES (and there are fatal collision reports that mention that in one incident, the car only checked for attentiveness 6 times in 37 minutes), and NHTSA had to push Tesla into reducing that to a sane number.
The question is how that compares to human drivers.
Imagine everyone would have been using autopilot in the past and steering wheels would have been introduced just recently. We would see horrifying reports about humans causing deaths by speeding, driving drunk, falling asleep etc. It would seem as if letting humans drive is an insane idea.
But would that really mean we have to return to autopilot immediately?
We don't know until we have some numbers. Like miles driven per fatal crash. For humans and for software.
Wikipedia says:
For driving, the rate was 150 per 10 billion vehicle-miles
I'd argue both are relevant - just like in medical research they look into both efficacy (performance under ideal and controlled conditions) and effectiveness (performance under actual real-world conditions).
Tesla gave me, totally out of the blue, a one-month free trial for Full Self-Driving for my Model Y. I have it until the end of April.
My thoughts:
- It feels like a very cool tech demo.
- It feels like a very stupid driver.
- I have to watch the road, because the cabin camera tracks my eyes, and my car will scream at me if I look away for more than a few seconds.
FSD feels absolutely magical to me: I'm in a car! Driving itself! It's making lane changes, putting on blinkers, stopping for pedestrians, dynamically reacting to traffic. Unbelievably cool. Totally feels like the future.
At the same time: I would not trust this thing to drive me to the grocery store down the street without supervision. It's both dangerously cautious and foolhardy. It once stopped for a red light 3 feet past an intersection stop line. It has also spent 30 seconds waiting to make a left turn, with no cross traffic, only to abruptly make the turn right in front of an oncoming car.
My verdict overall: it is nowhere close to replacing a human. It will cause dangerous situations if left to drive without human interferance. *AND YET:* it's a great technology when supervised, it is probably safer to drive FSD + human than just human alone, and it is absolutely & inevitably the future for all driving.
If my commute were much longer & more stressful, then most likely yes. My commute is 25 minutes in relatively stress-free suburban traffic, so driving doesn't bother me much.
> it's a great technology when supervised, it is probably safer to drive FSD + human than just human alone
No, because FSD + human is very difficult to maintain for any length of time. That's because if the FSD is reasonably ok in non-challenging situations, it is human nature to start paying less and less attention.
That's what the article talks about, crashes where the human had plenty of time to react but didn't. The fact that they didn't tells us they were not paying attention.
We can (and yes, should) blame those humans for not paying attention, but at the same time we (society, and laws) need to recognize that human nature is what it is, so asking a person to be 100% attentive and ready to react when they get used to not having to do anything is not a realistic expectation.
> In 59 crashes examined by NHTSA, the agency found that Tesla drivers had enough time, “five or more seconds,” prior to crashing into another object in which to react. In 19 of those crashes, the hazard was visible for 10 or more seconds before the collision. Reviewing crash logs and data provided by Tesla, NHTSA found that drivers failed to brake or steer to avoid the hazard in a majority of the crashes analyzed.
This tells you everything about autonomous driving. It is either working 100% all the time, or it does not work at all. Problem here is that if human brain is not in the driving loop, it will just disconnect. You can have open eyes and watching road, but you will just stop processing what is going on around you.
Just open a YouTube video of some boring long dashcam drive on highway and see how long you will manage to keep alerted and able to intervene by just watching. 5 minutes? 10 minutes?
I’m not laughing at fatalities I’m laughing at the way reporters refuse to drop the spin on FSD and AP kills people as if people don’t die daily from just… driving
As someone else said when seatbelts came around people also botched about them killing people until it became understood that sure people can die from them but it saves A LOT more.
People hate elon as should be, but using Tesla as a punching bag as “unsafe” because you can trick it to let you not pay attention is a joke when shit like bluecruise is hands free completely and A LOT less safe having driven with it than in a Tesla even just talking about on highways
All of these articles on Autopilot and FSD are hit pieces.
Every single other LKAS/ADAS out there today doesn't disengage at traffic signals, road obstacles or vulnerable road users, like cyclists or pedestrians, yet is more than happy to disengage at the slighest bit of road curvature, usually with minimal or no warning.
Said differently, these systems will happily run over people if they detect lanes and AEB (if equipped!) fails or veer you straight into a median/ditch when the road curves.
While they are more insistent about asking for driver input, these systems can easily be defeated. FSD will lock drivers out of using the system if the in-cabin camera detects phone use (which it is very good at doing according to @greentheonly's videos on the feature that were uploaded earlier last year). This code was merged into production Autopilot code last year, so non-FSD users are given the same treatment.
Meanwhile, you can buy a current model year Volvo XC90 or Porsche Taycan or whatever whose ADASes don't have cameras at all and are more than happy to let you keep thirsting on X or Insta as long as you tap the wheel every so often.
I'm not saying that these reports are wrong. FSD has a lot of work to do. Mine was within an inch of hitting a car in a parking lot the other day (I intervened). I'm saying that if we're going to roast Autopilot/FSD for not being attentive enough, let's also make sure that every other automaker's take on autonomy gets the same treatment.
53 comments
[ 5.0 ms ] story [ 105 ms ] threadTens of thousands of people die from cars every year. Yet we are not attacking Ford or Toyota.
This is not trying to say the car quickly jerked into a pedestrian on the side of the road.
However, Lane Keeping and Autosteer systems, like Volvo's DrivePilot, have existed for a while. These usually activate whenever they detect lanes. However, they don't consider VRUs or obstacles in their paths at all. As long as they can detect lanes, they will happily drive you into a wall or a person at speed.
The biggest difference between them and FSDb is that FSD can be activated on any road surface. FSD can make decisions based on the environment it's in as perceived through its cameras (with a very short delay), but it doesn't disable itself when it doesn't know what to do. It tries very very very hard to not ram itself into anything, but it is definitely capable of it if it guesses wrong.
Ironic how the Verge forms a union of Autopilot and FSD crashes in their headline to further convolute the major differences. As much as we blame Elon for naming Full Self Driving, I also think the media is responsible for tacking all Tesla modes together to get clicks.
Feels hard to say their being complacent with their messaging. You still need to pay attention all the time.
> In 59 crashes examined by NHTSA, the agency found that Tesla drivers had enough time, “five or more seconds,” prior to crashing into another object in which to react. In 19 of those crashes, the hazard was visible for 10 or more seconds before the collision. Reviewing crash logs and data provided by Tesla, NHTSA found that drivers failed to brake or steer to avoid the hazard in a majority of the crashes analyzed.
If an unassisted driver had reactions like they'd be arrested for reckless driving at minimum.
Tesla’s systems are obviously not safe. I suspect naming them things like “autopilot” and “full self drive” really don’t help perceptions and the systems themselves are clearly not able to get a useful gauge of driver attention. They’re overselling ability and under checking driver compliance.
https://www.goddardlawwv.com/2020/01/30/2019-united-states-c...
https://www.lendingtree.com/insurance/brand-incidents-study/
And of course, RAM & Subaru drivers are the exact opposite of Volvo drivers. Surprised absolutely no one :)
NHTSA has the ability to recall every single Tesla ever put on the road, for any safety reason, including the recent 'warning light size is too small to read'
Therefore, NHTSA also have full authority to shut down FSD, or force Tesla to rename 'Full Self Driving' to something more fitting.
Yet NHTSA allows it to be what it is. If it were truly unsafe, NHTSA would immediately put the brakes on it. (Yes, have a Friday pun.) Before anyone can call FSD not safe, a full investigation needs to occur. These are just the first steps.
This article notes that the NHTSA will launch a new investigation to see if the new driver monitoring features are sufficient.
Part of the problem with investigations that take many years to complete is that the conclusion is meaningless due to over-the-air updates, as this issue was solved at least last year.
Last time I heard that, it was from an actual Tesla employee. Commuting 1.5 hrs to work and doing emails during that time.
I agree that this was possible with a device to apply force to the steering wheel before they added the cabin camera to autopilot / FSD.
However, since adding cabin camera to autopilot/FSD retroactively to all vehicles that can support it (2018+), stuff like this simply isn't possible without the car kicking you out of FSD or autopilot.
This change would have been made sometime mid-last year for autopilot and far before that for FSD.
Can't find the OG post, but it was discussed on reddit, for example
https://old.reddit.com/r/jobs/comments/1cbh024/trending_on_l...
When FSD was first released, the steering wheel checks were every fifteen MINUTES (and there are fatal collision reports that mention that in one incident, the car only checked for attentiveness 6 times in 37 minutes), and NHTSA had to push Tesla into reducing that to a sane number.
Imagine everyone would have been using autopilot in the past and steering wheels would have been introduced just recently. We would see horrifying reports about humans causing deaths by speeding, driving drunk, falling asleep etc. It would seem as if letting humans drive is an insane idea.
But would that really mean we have to return to autopilot immediately?
We don't know until we have some numbers. Like miles driven per fatal crash. For humans and for software.
Wikipedia says:
https://en.wikipedia.org/wiki/Transportation_safety_in_the_U...In Teslas latest investor update, they state that FSD has driven about 1.2 billion miles:
https://digitalassets.tesla.com/tesla-contents/image/upload/...
But I think that does not include autopilot?
So I'm not sure if there is enough public data for a comparison.
My thoughts:
- It feels like a very cool tech demo.
- It feels like a very stupid driver.
- I have to watch the road, because the cabin camera tracks my eyes, and my car will scream at me if I look away for more than a few seconds.
FSD feels absolutely magical to me: I'm in a car! Driving itself! It's making lane changes, putting on blinkers, stopping for pedestrians, dynamically reacting to traffic. Unbelievably cool. Totally feels like the future.
At the same time: I would not trust this thing to drive me to the grocery store down the street without supervision. It's both dangerously cautious and foolhardy. It once stopped for a red light 3 feet past an intersection stop line. It has also spent 30 seconds waiting to make a left turn, with no cross traffic, only to abruptly make the turn right in front of an oncoming car.
My verdict overall: it is nowhere close to replacing a human. It will cause dangerous situations if left to drive without human interferance. *AND YET:* it's a great technology when supervised, it is probably safer to drive FSD + human than just human alone, and it is absolutely & inevitably the future for all driving.
Would you live in spider infested house with knowledge there will be only one deadly spider trying to bite you per week?
Would you eat in a restaurant where a badly prepared Fugu fish is dropped randomly into meals once a week?
No, because FSD + human is very difficult to maintain for any length of time. That's because if the FSD is reasonably ok in non-challenging situations, it is human nature to start paying less and less attention.
That's what the article talks about, crashes where the human had plenty of time to react but didn't. The fact that they didn't tells us they were not paying attention.
We can (and yes, should) blame those humans for not paying attention, but at the same time we (society, and laws) need to recognize that human nature is what it is, so asking a person to be 100% attentive and ready to react when they get used to not having to do anything is not a realistic expectation.
This tells you everything about autonomous driving. It is either working 100% all the time, or it does not work at all. Problem here is that if human brain is not in the driving loop, it will just disconnect. You can have open eyes and watching road, but you will just stop processing what is going on around you.
Just open a YouTube video of some boring long dashcam drive on highway and see how long you will manage to keep alerted and able to intervene by just watching. 5 minutes? 10 minutes?
As someone else said when seatbelts came around people also botched about them killing people until it became understood that sure people can die from them but it saves A LOT more.
People hate elon as should be, but using Tesla as a punching bag as “unsafe” because you can trick it to let you not pay attention is a joke when shit like bluecruise is hands free completely and A LOT less safe having driven with it than in a Tesla even just talking about on highways
Every single other LKAS/ADAS out there today doesn't disengage at traffic signals, road obstacles or vulnerable road users, like cyclists or pedestrians, yet is more than happy to disengage at the slighest bit of road curvature, usually with minimal or no warning.
Said differently, these systems will happily run over people if they detect lanes and AEB (if equipped!) fails or veer you straight into a median/ditch when the road curves.
While they are more insistent about asking for driver input, these systems can easily be defeated. FSD will lock drivers out of using the system if the in-cabin camera detects phone use (which it is very good at doing according to @greentheonly's videos on the feature that were uploaded earlier last year). This code was merged into production Autopilot code last year, so non-FSD users are given the same treatment.
Meanwhile, you can buy a current model year Volvo XC90 or Porsche Taycan or whatever whose ADASes don't have cameras at all and are more than happy to let you keep thirsting on X or Insta as long as you tap the wheel every so often.
I'm not saying that these reports are wrong. FSD has a lot of work to do. Mine was within an inch of hitting a car in a parking lot the other day (I intervened). I'm saying that if we're going to roast Autopilot/FSD for not being attentive enough, let's also make sure that every other automaker's take on autonomy gets the same treatment.