Abstract
This paper presents a method to build a hybrid map in on-line for autonomous robot using a Self-Evolving Modular Network (SEEM). The SEEM is the modular network that have a graph architecture. The module and path in the SEEM are generated in self-organizing manner via a learning process. The proposed method is possible to build a hybrid map in online while moving the autonomous robot. Aim of this work is to develop the system for autonomous building of the hybrid map using SEEM. For this purpose, we performed the following two works: (1) Selecting the backbone algorithm of the SEEM for building the hybrid map, (2) System design for building the hybrid map using the SEEM. In (1), a Growing Cell Structure (GCS), a Growing Neural Gas (GNG), and an Evolving Self-Organizing Map (ESOM) were compared by experiments in order to select the backbone algorithm of the SEEM. As the result, it was suggested that the algorithm of the ESOM is appropriate as the generation mechanism of the SEEM. Moreover, in (2), the hybrid map was built using the SEEM based on the ESOM. As a result, it was suggested the proposal method is possible to build the hybrid map from only vision information.
