Training AI datasets for visual recognition plays a role, certainly, but the ability of a Festo robotic hand to grasp a apple without crushing it depends 80% on its mechanical design and sensors, and only 20% on AI.
Imagine a car with a perfect navigation system (AI), but defective brakes; it could identify the destination 99% of the time but crash upon arrival, which is a cause-and-effect ratio of 4 to 1 between mechanics and AI for the final execution.
AI provides the target, but the precision of the gesture comes from the finesse of the hardware and haptic control algorithms, which is a 60 percentage point difference in influence.