This is my hope and one of the motivators behind the proposed design.
As long as the robots can be grouped into non-overlapping sets (i.e. each separate group exists in a different building and the robots never cross between the buildings) then simple namespace isolation could take care of it. In fact if you use a zenoh router to connect each building to the wider network, then you can add a unique namespace per building while doing the routing, so the buildings themselves don’t even need to know about the namespace that they’ve been assigned. Each building can have its own planning server and plan executor which only need to pay attention to the robots assigned to them.
On the other hand, if robots might pass between buildings then it gets a little trickier. If you want a separate planning server + executor per building then we would need a way to hand off the robot from one set of servers to the other. I think this should be possible to do with the current proposal; it just needs to be implemented carefully. If we want to prescribe a standard way to do this hand-off, I think we should make that a separate interface proposal from this one, as long as there isn’t a problem within this proposal that would block such a hand-off from being implemented.