Training expert modules for a mobile robot using mnSOM

A promising approach to performing the complex task of navigation of a mobile robot is to decompose it into simpler subtasks by task segmentation. Proposed is a new method for task segmentation in navigation of a mobile robot by a modular network SOM (mnSOM). mnSOM has both segmentation and interpolation abilities. In the standard mnSOM algorithm, data classes need to be known in advance. In mobile robot applications, however, data classes are unknown. Hence, we propose to decompose sequence data into many subsequences, supposing that a class label does not change within a subsequence. During learning, modules in mnSOM compete with each other, generating a winner in each subsequence called an expert. The resulting mnSOM demonstrates good segmentation performance of 94.05% for a novel dataset.