Two Different Approaches to Natural Indoor Landmark Recognition for Robot Navigation

In this paper, we address the problem of how to recognize natural landmarks for robot navigation (in an unknown office-like environment) given a description of the actual robot neighbourhood acquired through a ring of ultrasonic sensors. Such a proximity information is coded in a local map describing obstacles around the robot by means of occupied cells. The problem lies in recognizing, starting from the map, patterns of places that we want to utilize as landmarks in the building a global topological map of the whole explored environment. We show two different techniques. The first one relies on a wavelet representation of the local map and in the cross-correlation of wavelet coefficients of the input case with old cases extracted from a library, while the second one exploits an abstract description of the patterns through geometric constraints. About the latter, the experimentation work is still in progress, but we can already show very good results based on a large number of trials.