Really glad you are okay, thus far anyway. It makes me mad that someone else’s inattention causes harm to another.
My Tesla sales guy mentioned that about AP...but isn’t that where the anti collision system comes into play? I’ve had a car pull in front of me, brake , and my non Tesla car slammed the brakes on. The car was in adaptive cruise control. If the idea is to maintain a distance of 25 feet from the object in front of you it doesn’t seem like it should matter whether the object is moving or stationary. Radar signal bounces back a target whether it or you are moving (on ships anyway). I’d love to understand the engineering a bit better if anyone knows why the tech functions as it does.
It's not the braking in response to an object slowing part that's hard.
The part that makes stationary cars problematic is the defining of the objects.
Tesla is using chirping continuous wave radar to do adaptive cruise and emergency braking, with limited sensor fusion with the AP camera's neural networks.
Continuous wave radar is constantly putting out a signal. The computers analyze the frequency shift to determine the relative velocity of anything it bounces off of, and direction finding on the return for azimuth (and ascension if they measure that.)
To get distances, the radar chirps - twenty times per second, it does a frequency hop and then it listens for how long it takes the reflected signal to hop.
So in this scheme (very common in automotive radar,) objects are recognized by their Doppler shift (relative speed to the car,) and then assigned a direction and distance by further analysis.
But there are a whole lot of things in the world that don't move but do reflect radar. Enough that even if the car wanted to track them all, it's doubtful that the direction finding part could build an accurate picture. What's more, the strength of a radar return doesn't necessarily have much correlation with an object's size - geometry and materials make a huge difference.
So the solution for first generation systems was to just throw out all the stationary returns. That's why those systems can't work all the way to a stop or start again from a stop.
Second generation systems use memory as well - objects that were identified by Doppler shift which then come to a stop are remembered and it tracks them as they get closer through the chirp timing. This lets the car do full stop and go in adaptive cruise, but it still can't see a stationary car unless the car was moving earlier.
Tesla went a step further. The neural networks analyze the camera image and identify things that look like cars. The car then does some sort of sensor fusion negotiation to mark them as objects even though they were always stationary - azimuth matching on the chirps, maybe?
This usually allows the Tesla to see and correctly respond to stopped cars in the lane that it never saw moving, but it depends on the network seeing the car's image. Which may be why most of the cases we see in the news have been odd angles and unusual paint jobs (bright red fire truck at a thrity degree angle to the road with flashing lights that change its profile) - they are harder for the neural network to see as cars.
One additional path forward Tesla said they were starting implementation on a couple years ago but I've never seen evidence actually happened is the radar whitelist. The theory was that the cars would collect a list of things they saw that looked like cars but don't cause the driver to react, geotagged and categorized for strength. At some point in the future, Tesla would declare the list mature and the cars would download tiles for the areas they are driving in - and start braking for large stationary returns that aren't on the list until drivers override it. I have seen the car brake for no apparent reason near things that could confound it, but they haven't been new things, so they should have been on the list if Tesla has deployed a list.
One possible brute force solution to the problem with firetrucks and police cars would be to push a red hands takeover immediately when the car sees flashing red lights.