Exploring the difference between solving and teaching in sensorimotor tasks

Abstract

Considering a human is not only solving a task, but actively teaching how to solve it to a robot has not been extensively explored and is an important step to improve LfD algorithms. We explored the difference between solving and teaching in a sensorimotor task. In a first experiment participants first solved a continuous maze task and gave demonstrations for a robot afterwards. While teaching the participants could give negative demonstrations (how not to solve the task). In a second experiment we asked new participants to rate how informative they perceive the demonstrations from the first experiment. The results show that significantly more demonstrations from the Teaching-phase are perceived as informative than from the Solving-phase. Furthermore, significantly more negative than positive demonstrations were perceived as informative.