Abstract
Despite many recent advances both in terms of analytical and learning based approaches, grasping remains a challenging open problem in robotic manipulation. The major-ity of the research focuses on enhancing grasping capabilities by designing strategies characterized by different degree of intelligence for a given gripper. Our proposed approach to effective grasping is different: instead of optimizing a policy given a single gripper geometry, we search for the tool in order to grasp a given set of objects. To do so, we first introduce a parametrization for the geometry of two common families of grippers: two-fingers parallel-jaw and suction-cup vacuum. Then we present a novel grasp score discussing its properties for gripper design. Thanks to these we can formally cast the gripper design as an optimization problem, tackled with existing global optimization frameworks. Numerical findings on a set of industrial objects show effectiveness of our proposed approach.