It’s vague because they have no solutions for this (although they are probably trying something internally). What you are asking for completely possible with the xArm but you will need to figure out your own solutions for computer vision to direct the xArm and suction effector to the appropriate position.
I think some of the big industrial robot companies have canned solutions for this but be prepared to pay a hundred thousand dollars.
Actually based on their website, they offer a camera module to the xArm. With such a module, it is implicit they should have way for you to see and pick an object.