[Daishichimaru]

Easy trick:
1. Make two boxes
2. Make one a child of the arms that grab
3. Position the other exactly where the robot will be grabbing the box
4. Put a visibility track on both boxes
5. Animate the track so that the unparented box on the floor fades to transparency right when the arm would latch on.
6. Make the box that's a child of the arm become visible on the frame where the unparented box becomes invisible.

This way you don't have your box jumping all over the place trying to keep it still and then in time with the robot. If he's dropping another box off just use three. Or, if you're really worried about polycount, you can make a keyframe for the box sitting still and copy that into every frame where the box would be still but leave it free where it would be carried by the robot. Same thing with a bit less flexibility and a lot more annoyance.