Difference between RoomPlan and ARKit plane and mesh classification

Hello,

ARKit already have ways to identify walls, doors, chairs and other objects:

Here there seems to say that RoomPlan is powered by a distinct (although similar) algorithms and AI model than ARKit scene understanding. Is it correct? I was wondering what was exactly the point of designing a new algorithm from scratch and also which of them present better performance and more reliable results? Why are two different methods designed for the same purpose integrated in the framework?

Difference between RoomPlan and ARKit plane and mesh classification
 
 
Q