In most Autopilot-related incidents, wasn't the driver somehow intentionally defeating safety features, such as by hanging a weight on the steering wheel? IMO, if you knowingly and intentionally defeat any safety system, you should be 100% on the hook for anything that happens that the safety system might have prevented, no matter how misleading any messaging may be.
> In most Autopilot-related incidents, wasn't the driver somehow intentionally defeating safety features, such as by hanging a weight on the steering wheel?
It is full-auto-pilot... so it shouldn't be an issue. But either way, I don't think that it was the case in most malfunctions.
> It is full-auto-pilot... so it shouldn't be an issue.
That's not how autopilot works even in planes, and Tesla makes it very clear that you need to pay attention before it lets you enable it for the first time.
No, the issues NTSB was looking into stem from Tesla's terrible stationary object detection. 10+ people using autopilot as directed have plowed into the back of stopped emergency services vehicles.
NTHSA requested docs on this issue from 12 automakers in total, as all of their emergency braking systems have this limitation when utilizing driver assistance systems.
My theory is that it’ll eventually be a requirement to detect emergency vehicles optically using a similar system to how traffic lights detect emergency vehicle strobe lights and give them priority (Opticon system). This is trivial with a front facing camera, but not possible with radar.
If this is so trivial with front facing camera, why has Tesla not implemented it?
Stationary object detection appears to be totally missing, whether it's Tesla repeatedly trying to smash itself into the concrete support posts for the Seattle Monorail, pedestrians at an intersection, emergency vehicles flashing a standardized set of lights, etc
In some cases, they have. My Model Y has no front facing radar, but supports automatic emergency braking with vision only (Autopilot currently limited to 80mph, it’s 90mph in our older S and X vehicles with front facing radar; the limitation will be raised once their models are reliable at higher speeds). I expect stationary object detection based on vision eventually, which is not possible with radar alone due to inherent technology limitations.
It looks like the latest Tesla OS release contains the feature I mentioned (emergency vehicle detection using emergency light detection).
New in #Tesla 2021.24.12 Owners Manual for #Model3 #ModelY
"If Model 3/Model Y detects lights from an emergency vehicle when using Autosteer at night on a high speed road, the driving speed is automatically reduced and the touchscreen displays a message informing you of the slowdown. You will also hear a chime and see a reminder to keep your hands on the steering wheel. When the light detections pass by or cease to appear, Autopilot resumes your cruising speed. Alternatively, you may tap the accelerator to resume your cruising speed.
Never depend on Autopilot features to determine the presence of emergency vehicles. Model 3/Model Y may not detect lights from emergency vehicles in all situations. Keep your eyes on your driving path and always be prepared to take immediate action."
It's hard to pay proper attention when a nearly-good-enough L2 system fails. They work just well enough so you're not paying attention and then the car hands you control without warning.
I have the basic autopilot (not the paid FSD) and trust it very little. I love traffic aware cruise control where I'm still steering, but normally don't ever use autopilot. I turned it on briefly today for a km on a long, straight freeway so that I could more safely switch audio sources while on a multi-hour drive. Switched it right back off once I was done fiddling.
Google studied driver attention and decided advanced L2 and L3 systems were quite dangerous because drivers simply are not ready to take back over right away when needed. That's why they jumped right to L4.
>They work just well enough so you're not paying attention and then the car hands you control without warning.
There's a book, Our Robots, Ourselves by David Mindell, which describes the aviation experience with auto-land. Auto-land is no longer used for this reason, and the related one that the automatic system gives up when the situation is bad and you only have seconds to recover, which is when you want the pilot to have been paying attention for quite a while.
Planes use pilot assistance (head-up displays of where the plane thinks it's going, and where it thinks it should be going), rather than automatic landing systems these days.
The NTSB's attitude towards safety has always been that data from accidents should be used to improve systems and processes in order to minimize the possibility of further similar incidents in the future.
This approach has a much better track record than looking to assign blame to the driver and insisting that in the future the human simply needs to act more perfectly.
a yellow light or a car in front is very well predictable, if you feel you should be slowing down and the car isn't, you have time for apply correction, car not slowing down is an implicit warning of autopilot malfunction you can act upon.
a emergency vehicle blocking a tiny fraction of the lane doesn't require as much slowing, you can just slightly move left of your lane to compensate. car not slowing down doesn't trigger the same "brain" warning, because you don't expect it to.
this is what sets these crashes aside from the other, you get no explicit and no implicit warning
Wikipedia: "In 1996 and 1997, Homendy worked as a government relations manager at the American Iron and Steel Institute. From 1997 to 1999, she was a legislative representative for the AFL–CIO Transportation Trades Department. From 1999 to 2004, she was a legislative representative for the International Brotherhood of Teamsters. From 2004 to 2018, she was a Democratic staff member for the United States House Transportation Subcommittee on Railroads, Pipelines, and Hazardous Materials.[5] In 2018, she was appointed as a member of the National Transportation Safety Board. On May 19, 2021, it was announced that Homendy would be nominated to serve as chair of the NTSB.[6]"
Given that Tesla will put many autoworkers out of a job due to fewer components in electric vehicles, and given that Full Self Drive when perfected will put many Teamsters out of a job, its easy to predict NTSB obstacles to Telsa innovations.
Good for NTSB however to publicly advocate that autonomous vehicles will be safe when eventually available.
Issue is in my opinion that Tesla deliberately doesn’t install features to closely monitor driver‘s attention. In a Mercedes , if you don’t pay attention , multiple systems detect this and system is shutting down auto pilot functions with pre-warning .
18 comments
[ 1.3 ms ] story [ 46.1 ms ] threadIt is full-auto-pilot... so it shouldn't be an issue. But either way, I don't think that it was the case in most malfunctions.
That's not how autopilot works even in planes, and Tesla makes it very clear that you need to pay attention before it lets you enable it for the first time.
My theory is that it’ll eventually be a requirement to detect emergency vehicles optically using a similar system to how traffic lights detect emergency vehicle strobe lights and give them priority (Opticon system). This is trivial with a front facing camera, but not possible with radar.
https://www.reuters.com/business/autos-transportation/us-ask...
Stationary object detection appears to be totally missing, whether it's Tesla repeatedly trying to smash itself into the concrete support posts for the Seattle Monorail, pedestrians at an intersection, emergency vehicles flashing a standardized set of lights, etc
In some cases, they have. My Model Y has no front facing radar, but supports automatic emergency braking with vision only (Autopilot currently limited to 80mph, it’s 90mph in our older S and X vehicles with front facing radar; the limitation will be raised once their models are reliable at higher speeds). I expect stationary object detection based on vision eventually, which is not possible with radar alone due to inherent technology limitations.
New in #Tesla 2021.24.12 Owners Manual for #Model3 #ModelY "If Model 3/Model Y detects lights from an emergency vehicle when using Autosteer at night on a high speed road, the driving speed is automatically reduced and the touchscreen displays a message informing you of the slowdown. You will also hear a chime and see a reminder to keep your hands on the steering wheel. When the light detections pass by or cease to appear, Autopilot resumes your cruising speed. Alternatively, you may tap the accelerator to resume your cruising speed. Never depend on Autopilot features to determine the presence of emergency vehicles. Model 3/Model Y may not detect lights from emergency vehicles in all situations. Keep your eyes on your driving path and always be prepared to take immediate action."
https://mobile.twitter.com/analytic_eth/status/1440537502272...
If you were using Autopilot as directed, wouldn't you have been paying attention, and so braked yourself when you noticed Autopilot wasn't?
I have the basic autopilot (not the paid FSD) and trust it very little. I love traffic aware cruise control where I'm still steering, but normally don't ever use autopilot. I turned it on briefly today for a km on a long, straight freeway so that I could more safely switch audio sources while on a multi-hour drive. Switched it right back off once I was done fiddling.
Google studied driver attention and decided advanced L2 and L3 systems were quite dangerous because drivers simply are not ready to take back over right away when needed. That's why they jumped right to L4.
There's a book, Our Robots, Ourselves by David Mindell, which describes the aviation experience with auto-land. Auto-land is no longer used for this reason, and the related one that the automatic system gives up when the situation is bad and you only have seconds to recover, which is when you want the pilot to have been paying attention for quite a while.
Planes use pilot assistance (head-up displays of where the plane thinks it's going, and where it thinks it should be going), rather than automatic landing systems these days.
This approach has a much better track record than looking to assign blame to the driver and insisting that in the future the human simply needs to act more perfectly.
a emergency vehicle blocking a tiny fraction of the lane doesn't require as much slowing, you can just slightly move left of your lane to compensate. car not slowing down doesn't trigger the same "brain" warning, because you don't expect it to.
this is what sets these crashes aside from the other, you get no explicit and no implicit warning
Given that Tesla will put many autoworkers out of a job due to fewer components in electric vehicles, and given that Full Self Drive when perfected will put many Teamsters out of a job, its easy to predict NTSB obstacles to Telsa innovations.
Good for NTSB however to publicly advocate that autonomous vehicles will be safe when eventually available.