The Right Policy for Offline Data Sets

The idea of which we should go through augmented levels is the thing that I think is the most exciting but done aside from Excel in the last year or two. So imagine you have a data set and train your existing world model that can predict the change in states for giving action. It takes in a state of action and it predicts the change in the state. And then at test time, you want to deploy the policy on a cheetah that's got like slightly heavier torso, if there's awful. You don't just want to throw up a data Set because I just use it in a wasteful way.