https://fortune.com/2022/06/10/elon-musk-tesla-nhtsa-investigation-traffic-safety-autonomous-fsd-fatal-probe/

Elon Musk’s regulatory woes mount as U.S. moves closer to recalling Tesla’s self-driving software

Initially the probe started last year in response to Tesla vehicles mysteriously plowing into the scene of an existing accident where first responders were already present. 

On Thursday, NHTSA said it had discovered in 16 separate instances when this occurred that Autopilot “aborted vehicle control less than one second prior to the first impact,” suggesting the driver was not prepared to assume full control over the vehicle. 

CEO Elon Musk has often claimed that accidents cannot be the fault of the company, as data it extracted invariably showed Autopilot was not active in the moment of the collision. 

While anything that might indicate the system was designed to shut off when it sensed an imminent accident might damage Tesla’s image, legally the company would be a difficult target. 

All of Tesla’s current autonomous features, including its vaunted Full Self-Driving tech, currently in beta testing, are deemed assistance systems in which the driver is liable at all times rather than the manufacturer. 

https://arstechnica.com/tech-policy/2022/06/us-expands-probe-of-tesla-autopilot-crashes-in-step-toward-possible-recall/

Tesla Autopilot safety faces bigger questions as US upgrades crash probe

Might need appropriate legislation to get those deathtraps either upgraded or off the road.

possibly there are paradoxical requirements involved in analysis of the situation of convergent vehicles, by convergent I mean (also) any two objects traveling in the same line in the same direction which also includes the possibility or range extending to zero speed/stationary for the object (possible path obstruction) ahead

if you consider what’s involves in distinguishing between a stopped vehicle ahead and a slower moving vehicle there’s a range of speeds for a vehicle ahead of which situationally the object ahead might be categorized as good-as-stopped, what’s interesting though is how that is arrived at, and importantly dealt with by a suitably cautious human driver

as the traveling vehicle approaches a slower-could-be-stopped potential path obstruction ahead, a cautious suitable assumption might be that anything could be stopped ahead, anything could be a stationary path obstruction

analysis of convergence rate (ignoring radar etc for a moment) requires convergence, and the certainty about the rate increases as the distance closes, in fact it might be argued the certainty is at its highest when two vehicles collide, the computation is made physical in kinetic energy altering the shape of the vehicles or whatever that collide

thinking of it in basic terms it’s instructive to consider how you know the speed of a vehicle ahead you’re approaching, and the only way to know the speed of a vehicle ahead is (to read your speedometer after) stabilizing at a speed which the distance between the two vehicles remains the same, which is known with greater certainty or ease when you are closer, and in that is a potential trap, troubles

now run that through your head as to how you arrive at the speed of a potential obstruction ahead is zero, that it’s stationary

It’s still in beta testing and one of the requirements to use the autopilot system is that the driver has to be ready to take over immediately if it doesn’t react correctly.
Hence all the crashes are the fault of the driver.