Grasp quality evaluation done right: How assumed contact force bounds affect Wrench-based quality metrics

Abstract

Abstract-Wrench-based quality metrics play an important role in many applications such as grasp planning or grasp success prediction. In this work, we study the following discrepancy which is frequently overlooked in practice: the quality metrics are commonly computed under the assumption of sum-magnitude bounded contact forces, but the corresponding grasps are executed by a fully actuated device where the contact forces are limited independently. By means of experiments carried out in simulation and on real hardware, we show that in this setting the values of these metrics are severely underestimated. This can lead to erroneous conclusions regarding the actual capabilities of the grasps under consideration. Our findings highlight the importance of matching the physical properties of the task and the grasping device with the chosen quality metrics.

Computing a Task-Dependent Grasp Metric Using Second Order Cone Programs

“…Irrespective of whether the grasp metric is task-dependent or task-independent, a key assumption in the computation of grasp metrics is that either (a) the magnitude of the normal force that can be applied at each robot-object contact is bounded by the same value (usually chosen to be 1) or (b) concatenating all the normal forces in a vector, some norm (either 1, 2 or ∞) of the vector is bounded. However as demonstrated experimentally in [12], this assumption does not reflect reality. In practice, the maximum amount of force that can be applied at a contact depends on the configuration of the manipulators (or fingers) holding the object as well as the torque limits of the joint motors.…”

“…All the metrics described above are task-dependent metrics since they take into account the task-specific TWS while evaluating the grasps. In general, it is very difficult to accurately compute the GWS or the TWS [7], [12], [3].…”

“…Krug et. al studied the effects of these underlying assumptions in [12] while evaluating various grasps and found that they do not accurately reflect the quality of the grasp. Therefore, we do not assume the magnitude of independent contact wrenches or the sum of the magnitude of the contact wrenches to be upper-bounded by 1.…”

“…This set has been geometrically described as an ellipsoid [10] as well as a polytope [11]. However, it is difficult to accurately obtain a geometric description of task wrench space [10], [12]. In our formulation we define a task as a unit screw, about which the fingers/manipulators need to apply a wrench to generate a desired motion.…”

Simultaneous Tactile Exploration and Grasp Refinement for Unknown Objects

“…To this extent, we follow a probabilistic approach for grasp quality based on force closure, which we aim to maximise during the exploration. Various grasp quality metrics have been proposed in the literature, most focusing on analytic approaches [38]- [40]. A common metric, based on force closure is presented in [41], which requires to compute the space spanned by the friction cones generated by all contacts, the Grasp Wrench Space (GWS).…”