[ ] Add torch as engine to API. So, if an input is MX, then process as Casadi MX, if an input is torch.tensor, then process as torch
[ ] After all the calucations, whatever is passed to the "outside" (logger, animator, sim step printer etc.) must be converted to ndarray. But this must be done at the controller level. Say, at the end of compute_action_sampled, do a type conversion
HIGH PRIORITY
LOW PRIORITY