Growing Neural Gas Based Space Structure Learning for Feature Vector Composed of Multiple Properties

Abstract

Space recognition is one of the most important capabilities for intelligent autonomous robots that work in unknown environments. For realizing the space recognition, Growing Neural Gas (GNG) based approaches have been proposed by many researches since the GNG can learn a space structure and generate a topological structure simultaneously. However, GNG cannot preserve the space information if the input vector is composed of multiple properties. For solving this problem, this paper proposes GNG with Different Topologies (GNG-DT) that has the multiple topological structures according to the number of properties. In addition, the learning result of GNG-DT does not depend on the scale in the input vector. Finally, we conduct on several experiments for evaluating our proposed method by comparing to other conventional approaches, and discuss the effectiveness of our proposed method.