Grasping known objects with humanoid robots: A box-based approach

Kai Huebner, Kai Welke, Markus Przybylski, Nikolaus Vahrenkamp, Tamim Asfour, Danica Kragic, Rüdiger Dillmann

Abstract: Autonomous grasping of household objects is one of the major skills that an intelligent service robot necessarily has to provide in order to interact with the environment. In this paper, we propose a grasping strategy for known objects, comprising an off-line, box-based grasp generation technique on 3D shape representations. The complete system is able to robustly detect an object and estimate its pose, flexibly generate grasp hypotheses from the assigned model and perform such hypotheses using visual servoing. We will present experiments implemented on the humanoid platform ARMAR-III.