Three-Dimensional Truss Modelling for Biped Climbing Robots

Abstract

Recognizing and modelling the three-dimensional (3D) truss-style environment, consisting of cylinder members, is a challenging task for climbing robots. State-of-the-art segmentation algorithms always suffer from incomplete member-segmentation. To solve this problem, in this paper we propose a truss environment modelling (TEM) method using a sliding bounding box algorithm. TEM can obtain accurate poses of the members in truss environments by two strategies. First, the remaining incomplete segmented members are localized by a sliding bounding box along the axis of the members. After that, members are merged based on the new endpoints detected after projecting points in the new bounding box onto the axis, reducing the uncertainty and complexity of members merging. Our proposed method was verified by experiments with steady performance, which satisfied the requirements of the climbing robots.