About input depth image

[Fitz135]

Hello,

I obtain RGB and depth images in IsaacGym, where the depth images are stored in uint16 format. The depth data obtained from IsaacGym cannot be used to correctly determine the grasp action. Is it possible to achieve a correct grasp without using the depth images from a RealSense camera?

[Fang-Haoshu]

Hi, why the depth cannot be used to correctly determine the grasp action? There are some grasp detection based on RGB by not that accurate

[Fitz135]

Hi, thanks for the reply, I have solved this problem. However, I wonder the detected grasp translation and rotation have undergone mirror flipping. I used the view matrix of the camera to transfer the translation and rotation in the camera coordinate system to translation and rotation in the world coordinate system. But it did not work. Do you have any suggestions?