pull down to refresh

I wonder if they would need fewer frames if they did some hand crafted intermediate representations rather than just trying to do it all in one step. As I understand it they recently deleted millions of lines of their AI and its basically pixels in -> accelerator and wheel position out now