Grasping with Soft Hands

Despite some prematurely optimistic claims, the ability of robots to grasp general objects in unstructured environments still remains far behind that of humans. This is not solely caused by differences in the mechanics of hands: indeed, we show that human use of a simple robot hand (the Pisa/IIT SoftHand) can afford capabilities that are comparable to natural grasping. It is through the observation of such human-directed robot hand operations that we realized how fundamental in everyday grasping and manipulation is the role of hand compliance, which is used to adapt to the shape of surrounding objects. Objects and environmental constraints are in turn used to functionally shape the hand, going beyond its nominal kinematic limits by exploiting structural softness. In this paper, we set out to study grasp planning for hands that are simple — in the sense of low number of actuated degrees of freedom (one for the Pisa/IIT SoftHand) — but are soft, i.e. continuously deformable in an infinity of possible shapes through interaction with objects. After general considerations on the change of paradigm in grasp planning that this setting brings about with respect to classical rigid multi-dof grasp planning, we present a procedure to extract grasp affordances for the Pisa/IIT SoftHand through physically accurate numerical simulations. The selected grasps are then successfully tested in an experimental scenario.