论文笔记之抓取:Efficient Grasping from RGBD Images Learning using a new Rectangle Representation

7D

The full 7-dimensional gripper configuration—its 3D location, 3D orientation and the gripper opening width.

Method

A two-step learning algorithm to efficiently learn this representation.

• Describe a certain class of features that makes the inference in the learning algorithm fast

• we describe certain advanced features that are significantly more accurate but take more time to compute. Each step is learned using the SVM ranking algorithm.

Representation

Rectangle
blue lines: where the two jaws of the gripper are to be positioned for grasping
red lines: the gripper opening width and the closing direction
2D rectangle \(G (r_G , c_G , n_G , m_G , θ_G )\):

• the upper-left corner of the rectangle;
• the dimensions of the rectangle;
• the angle between the first edge and x-axis
• the normal to the image plane indicates a good direction to approach.

Algorithm