Abstract
This report describes a robot-mediated task knowledge transfer framework for pick-and-place tasks. Such a task transfer is divided into two steps: (1) a person teaches task knowledge to a robot; and (2) the robot teaches the learned knowledge to another person. In the first step, the person performs a task and the robot observes it. The task treated here is to make a structure of boxes like an arch. The robot detects geometrical relationships between the boxes to make a model of the task; three relationships, On, Against, and Coplanar, are considered. The robot then executes the same task based on the learned knowledge, where the positions of boxes are automatically calculated according to their relationships. In the second step, the robot demonstrates the task to another person.
