An Analytical Method for Decomposing the External Environment Representation Task for a Robot with Restricted Sensory Information

Abstract

The problem of modelling and reducing knowledge needed to build an internal representation of the environment is still a milestone in robotics. This representation task is crucial for both understanding perception in humans and programming advanced robots with reasonable navigation skills. In this chapter, we propose an analytical method for decomposing this representation task in terms of a set of primitive inferences. All these inferences are analytical transformations of the sensory data that expand the input space, use rules to compute the centre of areas and polygons of open space and, finally, build a topological graph with possibilities of being updated by learning and used for navigation. The methodological approach followed in this chapter is to search for a library of reusable modelling components that could be used to solve other similar problems of topological representation.