Comment by mrfusion

Speaking of that. I’d think we could possibly train on simulations. We’d have the 3d information plus the rendered 2d camera view.