Training an AI model to recognize images is, in itself, a prerequisite for a robotic hand to grasp an apple, but presenting it as the main driver is a reversal of the real cause.
Simple visual recognition won't prevent an apple from being crushed if the mechanics of the hand and its pressure sensors are not perfectly tuned.
It's a bit like saying that having an address allows you to rent an apartment: it's just a start; you still need financial guarantees and a solid contract.
AI identifies, but it is the precision of physical engineering that ensures delicacy, as seen with specialized industrial gloves that do not use AI at all to grasp fragile objects.