Mary's room seems uninteresting, in that robot-Mary can predict pretty well what bit-pattern she's going to get upon seeing color. (To the extent that the human case is different, it's because of cognitive architecture constraints)

Regarding the zombie argument: The robots have uncertainty over the bridge laws. Under this uncertainty, they may believe it is possible that humans don't have experiences, due to the bridge laws only identifying silicon brains as conscious. Then humans would be zombies. (They may have other theories saying this is pretty unlikely / logically incoherent / etc)

Basically, the robots have a primitive entity "my observations" that they explain using their theories. They have to reconcile this with the eventual conclusion they reach that their observations are those of a physically instantiated mind like other minds, and they have degrees of freedom in which things they consider "observations" of the same type as "my observations" (things that could have been observed).