Learning Action Primitives

Abstract

The use of action primitives plays an important role in modeling actions. Action primitives are motivated not only by neurobiological findings, they also allow an efficient and effective action modeling from an information-theoretic viewpoint. Different approaches for modeling action primitives have been proposed. This chapter overviews the recent approaches for learning and modeling action primitives for human and robot action and describes common approaches such as stochastic methods and dynamical systems approaches. Active research questions in the field are introduced, including temporal segmentation, dimensionality reduction, and the integration of action primitives into complex behaviors.