Antipodal Grasp

A two-finger grasp where the contact points face each other and their connecting line lies inside both friction cones.

An antipodal grasp — “antipodal” meaning “located at opposite ends” — grips an object with two contact points from opposing directions, such that the line connecting them falls inside both points' friction cones (the range of directions a contact force can push without slipping). According to Modern Robotics, in the planar case, two frictional contacts meeting this condition form a force closure, able to resist an external force or torque in any direction; in 3D, two ideal point contacts can't resist rotation about the line connecting them, which in practice is handled by the twisting friction of the finger pads. Chen and Burdick gave an algorithm in 1993 for finding antipodal point pairs on irregular objects. This is exactly the grasp a parallel-jaw two-finger gripper performs, which is why many learned grasping methods first sample candidate antipodal point pairs from a point cloud or depth image, then score them with a network.

ExampleDex-Net 2.0 (2017) finds antipodal point pairs in a depth image, generates hundreds of candidate parallel-jaw grasps, and scores them with a convolutional network called GQ-CNN, reaching a 93% grasp success rate on known objects with an ABB YuMi arm.

Also called

Antipodal Grasping, Two-Finger Opposed Grasp

Related

Force Closure · Friction Cone · Parallel Jaw Gripper · Grasp Pose Detection · Dex-Net 2.0 · Grasp Quality Metric

Sources

Modern Robotics（Lynch & Park, 2017 预印本）第 12 章 Grasping and Manipulation (Chinese)
Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics (arXiv 1703.09312)