Robotic Exploration: Place Recognition as a Tipicality Problem

Abstract

Autonomous exploration is one of the main challenges of robotic researchers. Exploration requires navigation capabilities in unknown environments and hence, the robots should be endowed not only with safe moving algorithms but also with the ability to recognise visited places. Frequently, the aim of indoor exploration is to obtain the map of the robot’s environment, i.e. the mapping process. Not having an automatic mapping mechanism represents a big burden for the designer of the map because the perception of robots and humans differs significantly from each other. In addition, the loop-closing problem must be addressed, i.e. correspondences among already visited places must be identified during the mapping process. In this chapter, a recent method for topological map acquisition is presented. The nodes within the obtained topologicalmap do not represent single locations but contain information about areas of the environment. Each time sensor measurements identify a set of landmarks that characterise a location, the method must decide whether or not it is the first time the robot visits that location. From a statistical point of view, the problem we are concerned with is the typicality problem, i.e. the identification of new classes in a general classification context. We addressed the problem using the so-called INCA statistic which allows one to perform a typicality test (Irigoien & Arenas, 2008). In this approach, the analysis is based on the distances between each pair of units. This approach can be complementary to the more traditional approach units × measurements – or features – and offers some advantages over it. For instance, an important advantage is that once an appropriate distance metric between units is defined, the distance- based method can be applied regardless of the type of data or the underlying probability distribution.