learning_dynamics TODO: Nonlinear encoder and decoder (try combination) Train curriculum learning with more steps (1to3 3to5) Update forward function for curriculum learning LQR!!!!!!!!!! Better naming for checkpoints and losses