USF inventors have developed an approach whereby robots can learn hu-man-like motion and task-specific grasping strategies from human demonstration and analysis of a grasp quality metric based on distribution of task disturbance, and apply the learning results into planning, with respect to grasp qualities. Furthermore, they have developed a methodology to congregate object recognition and manipulation motion in an intelligent robot that connects interactive objects with their functional motions. These methods can be generalized to different robotic hand models.
Corresponding small wrap and lateral pinch of robotic hand to hu-man hand. They look similar to a human grasp but are different for a robotic grasp. Left: Small wrap grasps for a human hand (top) and a robotic hand (bottom). Right: Lateral pinch grasps for a human hand (top) and a robotic hand (bottom)