I. Introduction
Robotic mapping has been a field of study for more than thirty years [1] and the last ten have seen particular success. With widely-available platforms and sensors, two-dimensional occupancy grid maps can now be built with off-the-shelf tools like DP-SLAM [2] or GMapping [3]. Such maps support navigational planning and building highly accurate plans of buildings. Recent years have seen a focus on maps that operate at a higher level: maps of objects, not just free and occupied space.