I agree a lot of edge cases for an AI are things people are able to cope with easily but the reverse is usually also true that there are lots of things people mess up on while driving that a computer would easily handle correctly all the time.I don't think we'll ever see FSD (FSD as in the original definition of autonomous level 4 or even 5) in present cars.
FSD as in beta, possibly but unlikely outside the US. I don't think Tesla had much idea of just how different places have wildly differing rules and traffic conditions - blimey the 50 US states have different traffic rules!
And what has become fashionable to call "edge cases" are 99% of the time no more than what we as drivers cope with every day without conscious effort - and when we can't work something "edge casey" out we can reverse out of it or decide to do something unorthodox. I see no likelihood of the car being able to do this sort of thing, clever though they appear in videos.
Realistically we probably won't see level 4 or 5 autonomy FSD in current cars, but I'm confident that the FSD currently being tested in the US will eventually roll out here. Whether that's in a year, or four or more is anyone's guess.
One thing that I've often thought about is - would it be more advantageous to use the same training data for FSD worldwide and have one general model that understands all the rules and conditions in different countries, or is it better to have separate data for each region?
I'm kinda leaning on the latter, I think you'd get better results with a narrower set of rules. And for places like Australia, the vast majority of cars here will never be used overseas anyway. The problem for us then is a much smaller data set to train with compared to overseas.