Grasp Estimation
Grasp estimation is the task of estimating the 6D gripper pose in camera coordinates. Grasps can be categorized into the 2D planar grasp, or the 6DoF grasp. For 2D planar grasps, the grasp is constrained from one direction, so the 6D gripper pose can be simplified into a 3D representation, which includes the 2D in-plane position and 1D rotation angle. For 6DoF grasps the gripper can grasp the object from various angles.
Evaluating Grasp Contact Points
The first category of methods sample candidate grasp contact points, and use analytical methods or deep learning methods to evaluate the quality of the grasp. This relies on prior knowledge such as object geometry and physics models. Dex-net 2.0 uses deep learning methods to plan grasps NO_ITEM_DATA:mahlerDexNetDeepLearning2017.
Other methods such as in NO_ITEM_DATA:morrisonClosingLoopRobotic2018a generate pixel-wise grasp affordance maps