Questions about point cloud processing in simulation and reality

[misaka0502]

Thanks for your amazing work!

I'm very interested in how you process the point cloud observations, but I'm having some troubles that I hope will be resolved!
I noticed that you use mesh models to directly compute synthetic point clouds, which is interesting. However, I also noticed that the example point clouds you provided in your paper and website (especially the real point clouds) only have a few target objects, obstacles, and grippers. I'm curious why they're so clean. Are point clouds from other parts (like the desktop) discarded?

I haven't noticed the relevant code or instructions. Could you please clarify my question? Thank you very much！

[misaka0502]

In addition, the robot body (beyond the gripper) will inevitably enter the camera’s field of view—would this have any impact?