Robotic grasping of deformable objects: a simulation benchmark

Kurzfassung

The grasping action usually represents the first interaction between the end effector and the grasped object in the context of a robotic manipulation task. A crucial aspect of synthetizing grasping actions with robotic arms and multi-fingered end effectors (robotic hands) is predicting the outcomes of a grasping action before the successive steps of the manipulation task occur. This difficult job is performed by a set of quantitative parameters called Grasp Quality Metrics (GQM), representing the assumed most crucial physical and geometrical characteristics of the configuration reached by the hand and the constrained object at the end of the grasping transient. In the past decades, researchers have concentrated their studies on formulating models and consequent parameters in order to judge grasps acting on objects approximable as rigid, due to the intrinsic simplifications this hypothesis allows. In order to fulfil the request for more flexible and human-like performances of robotic manipulators, in the most recent years the scientific community has increased its interest towards grasping actions applied on deformable bodies, to allow the exploitation of the flexibility and dexterity properties typical of multi-fingered robotic hands. This Thesis work has been conducted under the collaboration with the Robotics and Mechatronics department of the German institution Deutsches Zentrum fur Luft- und Raumfahrt (DLR), located in Wessling (D), with the objective of studying the subject of grasping and manipulation of deformable objects with a variable-stiffness robotic hand. A particular interest has been posed on deformable objects mimicking the behaviour of meat and fish cuts. After an introduction on the Mechanics governing the grasping action, grasped objects are categorized to depict the properties of the bodies of interest. Successively, an overview on the fields of grasping action synthesis and planning approaches is proposed. A list of Grasp Quality Metrics present in literature and applied to deformable bodies is reported and commented, ranging from wrench-based, to geometry and energy-based formulations. A grasping simulation benchmark has been set up on the Nvidia Isaac Gym software platform, utilizing an opportunely simplified model of the DLR’s end effector CLASH 3F. Output grasping data are discussed and future development horizons are proposed, having this work as a starting point