[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.

[PAClearner]

I don't know if people do this for vision yet but I know some people doing this for RL for robotic control.

[annerajb]

Google has a simulation for their cars that does this. Nvidia has a hw suit for this purpose too and for testing autopilot hardware by feeding it data from another rack/server