Robot Segmentation Mask for real depth images

[Dhanushvarma]

MuJoCo can render segmentation masks in sim, but is there a way to mask the robot out of a real depth image given a camera pose and qpos?

cuRobo does this with RobotSegmenter — unprojects depth to points, computes distance to robot collision spheres, returns a mask + filtered depth.

MuJoCo already has the FK and collision geometry, so it seems like it should be possible without maintaining a duplicate robot model elsewhere. Is there an existing path for this, or is it something that could be added?

Thank you.