Study of Step Perception using Space Recognition based on Topological Clustering

Abstract

This paper proposes a step perception method for autonomous mobile robots using a graph structure built by Adaptive resonance theory-based Topological Clustering. While our previous method utilizes the result of path planning, in this paper, the step perception method performs principal component analysis on local node groups in the graph structure and determines the presence or absence of steps based on features of curvature and normal vectors. Furthermore, by introducing a screening process focusing on cluster structures, it aims to reduce misjudgments near walls. The effectiveness of the proposed method was verified through experiments using 3D LiDAR data measured in real environments.