You say that holding a ball and an apple are distinct uses and not just hierarchical specializations, because of the regular geometry of the ball versus the intrinsic variability of the apple. I understand the idea that the task is not the same, but I think the distinction is even more fundamental than that.
It's not just a matter of shape or fragility, but of real-time adaptation challenge. For a ball, we have a predictable model, we can anticipate all movements; for an apple, with its irregularities and variable pressure points, the robot must constantly adjust its grip, like when holding a fragile egg.
For example, imagining a robot picking up a bruised or small-sized apple versus a perfect large apple, completely changes the sensors and algorithms needed. Grasping a ball remains a basic skill, while the apple requires continuous learning and much more advanced situational intelligence, making it a technological leap rather than just a specialization.