Just caught up with a couple days and hundreds of posts. I do not mean to disparage AP in this post - if the driver was truly engaged at the critical moment, or if the driver was in the habit of disengaging AP in challenging areas like this funky intersection, it appears the accident could have been avoided. The exit was also poorly designed and maintained and the barrier should have been reset. With that said ...
Tesla’s blog says “the driver’s hands were not detected on the wheel for six seconds prior to the collision.” I think by saying this Tesla is trying to imply that he didn’t have his hands on the wheel, but the comment is very misleading. I’m sure that at most points in time
my hands wouldn’t have been detected for the prior six seconds simply because my hands, one of which is always on the wheel when I use AP, didn’t apply steering torque during the prior six seconds. I sometimes set off a nag by not applying torque, even though I believe it takes much more than six seconds of an absence of torque to produce a nag. So I don’t think we can infer anything about the driver’s state of attention from that fact. We can infer more from the simple fact that he hit the barrier.
It seems like all the incidents that set off long discussions here and in the press involve hitting stationary objects: semis, fire trucks, the collapsed barrier. Someone upthread correctly pointed out that radar can detect stationary objects but such detections are thrown out because a high false positive rate would result. Imagine the car slamming on the brakes at highway speeds when you are about to pass a building on the side of the road that is positioned in a certain way, or pass an overhead road sign. If this happened once it would put me off of AP for a while, and if it happened twice I would never use AP again. Such false positives that lead to hard braking, possibly with other cars close behind, are absolutely unacceptable, from which it follows that it is up to the driver to detect and react to hazardous stationary objects.
That is much of the reason why AP is not equal to Self Driving.
@Reciprocity posted a great description of what AP really is back at 1171. But the problems are 1) FSD, which some owners have paid for, is impossible by definition without shifting the responsibility of dealing with stationary objects to the car, and 2) as incidents like this one accumulate, the absence of stationary object detection without false positives may become unacceptable to regulators and customers, even for AP. I’m personally fine with accepting stationary-object responsibility, so I hope (2) does not develop.
I suspect the current radar plus camera sensor combination used in AP1 through AP2.5 is unable to do reliable stationary object detection without unacceptable false positives. It’s a big enough deal that if it could, they would already have the car doing it. To really do it I think you need either a single forward looking LIDAR around hood level, or stereo cameras and lots of compute power, or something like cameras and a projected grid of infrared points that could be seen by the camera and interpreted by software. And then you would need sensor fusion with the radar to confirm there is no false positive. Of course none of this could be realistically retrofitted to the existing fleet, including those cars where FSD has been paid for.
So I have no great conclusion other than that this is a worry, and we should all be aware of APs limitations when we use it.