Very interesting post.... I think Andrej's approach (whatever that entails) combined with the super detailed maps engine and new processor news is likely the biggest clue as to how we are going to get to FSD. I still also wonder why Lattner didn't mesh... I know that's been talked about extensively in general terms, but I wonder specifically, what made someone at Tesla think, we aren't going to get to FSD fast enough with this guy's approach?
The only other thing I wonder about is why they haven't considered dual cameras on every single camera placement for redundancy and depth perception? I am sure this is not a novel idea.... and there is probably a very good reason for this besides cost but it seems to me that everything would be better in Karpathy's accuracy approach with dual cameras.... (you know... like a pair of eyeballs). The only thing I feel fairly certain about is that the processor will likely have to be updated on AP2.0 cars.
