Optimizing the parameters of tilting surfaces in robotic workcells

Abstract

Pick-and-place regrasp is an important grasp and manipulation skill for a robot to rearrange the world around it. This paper discusses how to design a tilted work surface for robotic workcells so that the robot inside the cell can improve its pick-and-place regrasps. In particular, the workcell design is guided by two heuristics, which simultaneously increase the number of stable placements of the objects to be grasped and the gripper's manipulability. We leverage regression techniques to compute a continuous function for fitting the multiplication of these two heuristics, and then maximize the fitted function for optimizing the parameters of tilted surfaces in a workcell. Experiments on using tilted surfaces to improve the stability and manipulability of various objects validate our approach.