Investigation on human environmental recognition from 2D and 3D point clouds by simulator and its comparison with self-localization by experimental robot

Abstract

When given a map, a human can select an appropriate route and reach a destination. In this research, we focused on human smart ability of self-localization. Recently point cloud information acquired by 2D or 3D laser range finder (LRF) is often used for mobile robot self-localization. In this research, human self-localization ability when given point cloud information (not RGB image information) is investigated by questionnaire using a developed computer simulator. The results show that 3D information is more effective than 2D information especially in case of given only sparse points. It is because the road surface information can be obtained from 3D point cloud. An experiment in which a developed mobile robot equipped with 2D and 3D LRF travels a given outdoor course, shows the superiority of 3D point cloud information to 2D one.