Human-inspired robotic grasping of flat objects

In this paper, we propose a human-inspired framework for grasping domestic flat objects placed on planar support surfaces. In particular, three grasp strategies are proposed which aim to pinch small flat objects from different scenes. The framework uses representations of the robotic hand, the support surface and the target object which encapsulate rough information for the scene. Furthermore, the strategies exploit the environmental constraint of the support surface by establishing compliant contact with it, which leads to increased robustness against object geometry uncertainties as well as pose estimation errors possibly introduced by the perception system. This is inspired by how humans perform relative grasping tasks with object pose and geometry uncertainties by using compliant contact with the support surfaces. Finally, the strategy selection is determined by a decision making procedure which uses the current scene representation.