ML approach for FSD was critized right from the start by mobileeye CEO who gave a very thorough talk on why that couldn't possibly work. I think the talk is still on youtube.
The guy was critized for being old school and not having adapted to the latest tech (including by elon musk when he dumped mobileeye), but in fact he was just plain right.
Until any FSD solution delivers, it's far too early to vindicate the mobileye CEO. As it stands, even with additional (expensive) sensors, it seems very likely that ML will be required regardless. Running purely off sensors is not going to be feasible when braking distances alone are greater than the sensors will be able to accurately measure.
He did give a second talk a few years later explaining how the purely cognitive aspect may be handled by ML, and the total system probably going to end up being a mix.
But his fundamental opposition to ML for things related to safety was, IMHO absolutely correct.
The guy was critized for being old school and not having adapted to the latest tech (including by elon musk when he dumped mobileeye), but in fact he was just plain right.
i think the tech community owes him an apology.