ABSTRACT
“Human teaching robots” is one of the most common hierarchical interactions between human and robots. This chapter focuses on the applications of hierarchical interaction enabled by imitation learning and analogy learning, respectively. This learning scheme serves as a powerful tool to transfer implicit skill models from a human operator (teacher) to a robot (student). We first discuss the framework of remote lead through teaching for implementing imitation learning. The human demonstration data can be captured remotely and then utilized in the robot reproduction. This imitation approach is verified in both an assembly scenario and a grinding scenario. Then we discuss the applications of analogy learning, including robotic grasping and motion re-planning. For robotic grasping, the robot can exploit the grasp examples provided by human and efficiently generate feasible grasps for a new object by finding the analogical mapping between objects. Similarly, for the robot motion re-planning, the planned motion used in the previous scene can be transferred to the novel scene by analogy learning.
