search

ZhangHanbo / Visual-Manipulation-Relationship-Network-Pytorch

MIT License
41 stars 13 forks source link

questions about the paper #12

Open gyxx323 opened 2 years ago

gyxx323 commented 2 years ago

Hello, I would like to ask you what you mentioned in the paper.

  1. When the robot is actually used for grasping, the approach direction is along the normal direction of the grasping point. What is the normal direction from the RGBD camera or a point? Do you need to compute a point cloud?
  2. Why is the best grab point chosen for the lowest depth point rather than the center point?

Thank you!

ZhangHanbo commented 10 months ago

Hello, I would like to ask you what you mentioned in the paper.

  1. When the robot is actually used for grasping, the approach direction is along the normal direction of the grasping point. What is the normal direction from the RGBD camera or a point? Do you need to compute a point cloud?
  2. Why is the best grab point chosen for the lowest depth point rather than the center point?

Thank you!

Sorry for the delay. This repo is no longer maintained.

For your question:

  1. We typically fit a plane using N points around the grasping point with least squares method. The normal vector will be then straightforward to get from this plane.
  2. This is because we do not hope that the executed grasp collide with any of the objects, because in our grasping detection module, there is not collision-detection or collision-avoidance part.