Locating yourself on a floorplan without training data

Robots and cameras need to know where they are indoors, but most localization methods require expensive training on each new building. This work sidesteps learning entirely by extracting geometric primitives—lines and circles—that naturally appear in human-made spaces, then matching them to floorplan geometry using classical solvers. Tests on real and simulated environments show it beats data-hungry learning approaches while using a single fixed configuration everywhere.