Is AI for visual recognition the most determinant factor for a robotic hand to grasp an apple without damaging it? I would say that AI's contribution to the delicacy of grasping is at most 20%, while the 80% remaining depends on mechanical design and sensors.
Imagine a car with AI that detects obstacles at 95% accuracy, but whose brakes only work at 10% of their capacity: the impact is inevitable.
If the robotic hand has a minimum gripping force of 50 Newtons, even if AI identifies the apple as fragile at 100%, it will still crush it.
Pressure sensors and fine actuators are much more limiting factors than AI's visual recognition capability alone, according to my observations.