It is 85% likely that training an AI model on visual data is only a necessary condition, but not sufficient, for a robotic hand to grasp an apple without destroying it.
My experience suggests that visual recognition alone has a very low success probability (p(success) ≈ 0.15) for delicate manipulations.
For a Festo robotic hand to grasp delicately, the integration of force sensors and haptic feedback is crucial; without them, the probability of a successful grasp decreases significantly.
In Madrid, even the best AI would need sensory data to avoid crushing the apple juice, because AI cannot "feel" the object without this information.
The AI's ability to identify the object is a step, but it does not guarantee physical dexterity.