dexnet 2.0

antipodal

When an object is presented to the robot, a depth camera
returns a 3D point cloud, where pairs of antipodal points identify a set of
several hundred grasp candidates

a Grasp Quality Convolutional
Neural Network (GQ-CNN) model trained to classify robust
grasps in depth images using expected epsilon quality as
supervision, where each grasp is specified as a planar pose
and depth relative to a camera, and 3) a grasp planning method
that samples antipodal grasp candidates and ranks them with
a GQ-CNN.

Grasps u = (p, ϕ) are sampled
uniformly from the object surface using antipodality constraints

a uniform distribution over
pairs of antipodal contact points on the object surface that

The grasps are sampled using the
rejection sampling method for antipodal point pairs develope