Building Each Other Up: Using Variation Theory to Build and Correct Human Mental Models of Robots

As we begin to coexist with robots, we must be able to anticipate each other's future actions to avoid dangerous or inefficient situations. How can we build up human and robot conceptual models to predict how each other will behave? To do this, I study how to apply human concept learning theories, like variation theory, to human-robot interaction. I have found that the contrast step of variation theory (showing contrasting examples of policies when learning about robot motions) improves people's ability to predict motion in unseen settings. For my proposed PhD, I seek to connect this idea of improving a human's model of the robot with enabling the robot to correct the human's model or alter its own behavior based on its model of the shortcomings in the human's understanding.