Sound-scapes for robot localisation through dimensionality reduction

Sound-scapes similar to landscapes, are geometric representations of an objects’ relative positions in the real world. In this paper we demonstrate how to obtain and use a sound-scape to assist the Aldebaran NAO with localisation. We apply dimensionality reduction techniques such as statistical learning methods which include neural networks, support vector machines, the recent graph based approximation technique isometric feature mapping to extract the NAO’s field co-ordinate from its recorded acoustic data. Results obtained includes visualisations of sound-scapes (robot’s positions on field) and positional accuracies of up to 80%.