Is there a paper? I think I get how they did their training, but I'd like to understand it more.

Does anyone know if this trained model would work on a different robot at all, or would it need retraining?