Incremental mapping of large cyclic environments

TLDR

Mobile robots rely on geometric or topological maps for navigation, yet automatic map creation remains unrealized, especially in large cyclic environments. The authors present a method, local registration and global correlation, to reliably reconstruct consistent global maps from dense range data. The method incrementally updates the map with each new sensor input and runs in constant time regardless of map size, except when closing large cycles. Real‑time implementation demonstrates the method’s effectiveness in several indoor environments.

Abstract

Mobile robots can use geometric or topological maps of their environment to navigate reliably. Automatic creation of such maps is still an unrealized goal, especially in environments that have large cyclical structures. Drawing on recent techniques of global registration and correlation, we present a method, called local registration and global correlation, for reliable reconstruction of consistent global maps from dense range data. The method is attractive because it is incremental, producing an updated map with every new sensor input; and runs in constant time independent of the size of the map (except when closing large cycles). A real-time implementation and results are presented for several indoor environments.