A rapid and accurate grasping planning method depends not only on the deep neural network architecture but also on the grasping model used by the network. For example, Mahler et al. provided a Dex-Net series [[5],[20]]. The dataset they used to train the net is composed of a set of...
view and thus can be applied to cases where the full 3D model of an object is not known. Other methods focus on specific cases of robotic grasping. For example, [24, 3, 28] assume that objects to be grasped belong to a particular set of shape primitives or compositions ...