Abstract
This paper presents a segment-detection and grouping scheme that allows incremental and on-line learning of indoor environment maps by mobile robots. The proposed algorithm is based on authors' earlier work of modeling the environment by Enhanced Adaptive Fuzzy Clustering (EAFC) algorithm along with Noise Clustering (NC). In this study, the modeling is refined by first dividing the world into discrete regions as local models. Then, the line segments in local and global models are grouped together by a hierarchical fuzzy system. Adjusting the membership functions that establish the grouping criteria controls the degree of approximation in such combination. The presented map building method for modeling an indoor office environment has successfully been implemented in real-time and tested on Pioneer 2DX mobile robot equipped with sonar sensors. The experimental studies demonstrate improved performance through continuous maps that are obtained by adopting the fuzzy system during the grouping of line segments within local and global models.