Questions regarding Normal Map Baking

[Baelkin]

I'm currently working on an armour and I really would like to make the normal textures by "baking" the normals from a high resolution source model onto a low poly model, and the principle it seems simple enough and it seems easy enough to get *something* baked as normals. However, there are a few things I'm a bit confused about, which I hoped someone could clarfiy.

 

First question: I'm using a method where you make the high poly model first and the low poly based on the high poly - how much of the high poly height topology should I take into account when making the low poly? For instance, if I'm modelling a flat surface which has small nobs and rivets poking out of it, should I make rudimentary nobs and rivets on the low poly model, or should I just leave it as a flat surface and hope the baking takes care of making normal map that adjusts the height according to the bumps and nobs?

 

Second question: When making the two models, should the low poly model be small enough to fit inside the high poly model? The following diagram of how normal baking works in that regard, has left me a bit confused, so I hoped someone could clarify:

 

http://www.gameartisans.org/gamecon/tutorials/images/joao_6.jpg

 

As for the baking technique itself, I'm following Ben Mathis' tutorial on the subject with João Costa's tips and tricks as supplement. Hopefully they cover the subject sufficiently, but if anyone got better "educational" material I'd like to get my hands on it. :)

[Baelkin]

Just thought of a supplemental question, according to João, you can float objects on the surface of the high poly, have mesh objects intersect other mesh objects mesh on the high poly and so on, but how does normal baking respond to mesh objects overlapping eachother? For instance, if you have a shoulder on which you float a shoulderguard, does the normal bake just ignore whatever the shoulderguard overlaps, or will the process behave oddly in this situation? I figure the bake to texture just ignores the issue and only uses the surface for the shoulderguard (due to the way the surface of the high poly is projected onto the UV map of the low poly through the rays cast), but I thought I'd ask instead of modelling a ton of extra polys that wouldn't mean anything in the end (except it'd be a nice modelling excercise).