I am trying to create an environment for RL where the size of my input (observation space) is not fixed. As a way around it, I thought about padding the size to a maximum value and then assigning "null" to those values that do not exist. Now, these "null" values are meaningful in a certain sense, because they are related to the shape and size of the input.
If these "null" values were zeros, would neural networks be able to distinguish between these zeros (nulls) and the zeros that are actually part of the picture? If that's not the case, should I assign a different number for the padding? What should I be mindful of in these scenarios? Is there any example I can look at with a similar situation?