Trained robots can have issues handling objects that are deformable and soft, such a fruit. This is due to challenges with teaching machines the appropriate level of force and dexterity, as well as difficulties with object modeling.
At the University of Tokyo, researchers developed a machine learning system for a robot with two arms that was successfully able to pick up and peel a banana without damaging it (news from Future Timeline). The robot’s total training time was 13 hours; watch it in action below!