Occlusion
Physical AI
A condition in which part or all of an object is hidden from a sensor's view.
Occlusion limits the evidence available for annotation and review. Guidelines should distinguish visible observations from permitted inference and explain when a result is unknown. Multiple views or surrounding frames may resolve the uncertainty, but they should not be assumed available when they are absent.
In practice
A gripper hides the contact point, so a reviewer needs another view or later evidence before confirming a stable grasp.