Are we allowed to shape the reward if the reward shaping is not dependent on the state, and no new rewards are introduced? For example, can we rescale the rewards or remove some reward entirely during training?
1 Like