What does the normalization of the inputs mean in the context of PPO? At each time step of an episode, I only know the values of this time step and of the previous ones, if I take track of them. This means that for each observation and for each reward at each time step I will do:
value = (value - mean) / std
before passing them to the NN, right? Specifically, I compute mean and std by keeping track of the values for the whole episode and at each time step, I add the new values to an array. Is this a valid approach?
Also, how can I handle negative rewards, such that being positive?