Grasp Quality Metric

A single number scoring a grasp by how well it can resist outside disturbances, used to rank candidate grasps.

A grasp quality metric maps a set of contact points, or a hand pose, to a single number, with a higher score meaning a better grasp — used to rank large numbers of candidate grasps. The most common is the ε metric proposed by Ferrari and Canny in 1992: with contact forces limited in magnitude and constrained to lie inside their friction cones, collect the achievable wrenches (force plus torque) at each contact point and take their convex hull; ε is the radius of the largest ball centered at the origin that fits inside that hull — the smallest disturbance, in the worst direction, that the grasp can still resist. A grasp that fails to achieve force closure scores zero or negative. Because force and torque have different units, torque is usually rescaled using a characteristic length of the object. Roa and Suárez's survey sorts these metrics into ones based on contact-point locations and ones based on hand configuration. Dex-Net 2.0 computes a robust ε metric under pose and friction uncertainty to label about 6.7 million synthetic samples, used to train the grasp-quality network GQ-CNN.

ExampleGripping the same disk with three fingers: when the fingers are evenly spaced around the circumference, the wrench convex hull is roughly symmetric and its inscribed ball is large, giving a high ε; when the three fingers are bunched on one side, the hull skews to that side, some direction of push is almost unresisted, and ε is very low.

Also called

ε Metric, Epsilon Quality, Ferrari-Canny Metric

Related

Force Closure · Wrench · Grasp Matrix · Grasp Planning · Dex-Net 2.0 · Friction Cone

Sources

Lynch & Park, Modern Robotics（预印本 PDF，12.1.7.3 节 Measuring the Quality of a Form-Closure Grasp） (Chinese)
Ferrari & Canny, Planning optimal grasps（ICRA 1992）
Mahler et al., Dex-Net 2.0（arXiv 1703.09312）