The standard should be: "Safer than human driving" shouldn't it?

And when you realize that humans are already terrible drivers it doesn't seem so reckless to hold AI driving to a realistic, achievable standard instead of an ideology.

Another way to look at it: imagine two different human drivers, one is clearly "better" and faster than the other, but wrecks occasionally, but the other is a "bad" driver, but never does.

Which is a more desirable model for an AI to emulate?

The only bar to clear, and I mean the only bar is, "Do autonomous vehicles kill fewer human beings than humans on the road do?"

That's it. When it decreases by even 1 human being below what humans cause, its time to switch over, because like every single mechanical, electronic, and computerized tool, it will only continue to reduce the number of human deaths as we progress with its development.

I get that autonomous vehicle deaths freak people out because they do things a human would never do. But? So what? If 53,000 people die in automobile accidents where they drive into each other because they stop paying attention to the road to reach back and slap their 10 year old who won't leave his sister alone, it makes absolutely zero difference than if the car drives into the ocean and drowns its driver. Same result. Dead person. We only think it matters because we like to think we have control over our lives. You don't and you never will. You could have a massive aneurism the next time you're driving and slam your vehicle right into an oak tree. The Universe is a random uncaring system.

Your life is just a series of mostly random events that you shape into a coherent story because accepting that shit just happens and you have to deal with it is, like, a real bummer, man.