A Novel Method to Estimate the Position of a Mobile Robot in Underfloor Environments Using RGB-D Point Clouds

Abstract

This paper is focused on the design of a mobile robot whose objective is to apply thermal insulation spray in underfloor voids, to improve the energy efficiency of buildings. Solving robustly the mapping and localization problems is crucial to achieve a high degree of autonomy during the development of this task. Nevertheless, underfloor voids constitute specially challenging environments mainly owing to the extreme unevenness of the terrain and the changes the environment experiences as the insulation process is carried out. Taking these issues into account, this work presents the implementation of the localization module of the robot, which is equipped with a laser scanner and an RGB (Red, Green and Blue) camera. The data captured by both sensors is combined to build point clouds that describe the appearance of the environment. While the robot traverses the a priori unknown environment, several point clouds are built and an alignment between each pair of consecutive clouds is carried out. From this information, the current position of the robot is estimated with respect to the previous one. The method has been tested with several datasets captured in real underfloor environments (building crawl spaces) and under real operating conditions.

Keywords

• Building crawls
• global appearance descriptors
• localization
• point clouds
• registration