3D perception and modeling for manipulation on Care-O-bot® 3

This paper proposes an algorithm that can be used to reconstruct a 3D environment or learn models for object recognition on a mobile robot. As sensors, color and time-of-flight cameras are used. 2D features are extracted from color images and assigned 3D coordinates. Those are the input for a modified fastSLAM algorithm that is capable of rendering environment maps or object models. The method is evaluated on the service robot Care-O-bot® 3.