Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 20: Focus Session: Modelling of Non-linear Dynamics in Biological Movement (joint session BP/ DY)
DY 20.2: Talk
Wednesday, April 2, 2014, 14:30–14:45, ZEU 250
Learning Motor Skills with Information-Theoretic Approaches — •Jan Peters1, 2, Christian Daniel1, and Gerhard Neumann1 — 1Technische Universität Darmstadt — 2Max Planck Institut für Intelligente Systeme
Synthesizing new motor skills from data has been a long standing vision of robotics, artificial intelligence, and the cognitive sciences. A first step towards this goal is to create approaches that can learn tasks triggered by environmental context or higher level instruction. However, learning techniques have yet to live up to this promise as only few methods manage to scale to high-dimensionality of humans and anthropomorphic robots. In this talk, we investigate a general framework suitable for learning motor skills in robotics and for explaining human movement learning which is based on the information-theoretic principles, such as movement organization, representation and acquisition by information entropy. As a result, the framework involves generating a representation of motor skills by parameterized motor primitive policies acting as building blocks of movement generation, and a learned task execution module that transforms these movements into motor commands. We discuss task-appropriate information-theoretic learning approaches for movements and illustrate their effectiveness on human movement data and in robot motor skill learning on both toy examples (e.g., paddling a ball, ball-in-a-cup) and on playing robot table tennis.