Goal Emulation in a Drummer Robot

Abstract: Learning from demonstration represents a powerful type of learning and a promising alternative to programming robots. However, even after years of research in the field the current level of robot imitation skills can be compared to a 2-year old child. Most of the presented works up to now have been exact imitation of trajectories without any particular understanding about the rudimentary goals of the movements. This paper presents our approach to developing a robot that imitates a human drummer, where the robot is able not to perform a scaled copy of the movements of a human demonstrator but to emulate the goal of the instructor’s actions through observation of action and goal extraction based on tests with human subjects. The intention is effectively reproduced and adapted to the robot’s situation and capabilities.