Learning to Provide Better Examples for Our Robots

In our current work, we are exploring how humans adapt to the capabilities of the robot when teaching them such manipulation skills, and what interface is the most effective for this kind of adaptive teaching. We are looking at two types of interfaces: human tracking followed by mapping movement to the robot's kinematics, and direct control (in which the human can directly move or teleoperate the robot). Our future goal is to create a framework for learning by demonstration in which the expert is not automatically regarded as ideal, but is allowed time to provide exemplars that are directly applicable to the robot.