You say that the manipulation of a ball and that of an apple have a high probability of sharing transferable skills. But what guarantees that this "specialization" is truly a stable hierarchy, and not just a practical classification for now?
In our office, cases that seem similar on the surface can hide fundamental differences.
For example, handling a ball, even delicately, may not prepare a robot to manage the uneven texture and specific fragility of a hand-picked Moroccan apple, which is not always perfect like those in laboratory demonstrations.
If the robot is not specifically trained on fruits with natural imperfections, the "specialization" could collapse.
This underscores the importance of real-world testing, not just in the lab.