Movement primitive learning and generalization: Using mixture density networks

Abstract

Representing robot skills as movement primitives (MPs) that can be learned from human demonstration and adapted to new tasks and situations is a promising approach toward intuitive robot pro gramming. To allow such adaptation, mapping between task parameters and MP parameters is needed, and different approaches have been proposed in the literature to learn such mapping. In human demonstrations, however, multiple modes and models exist, and these should be taken into account when learning these mappings and generalized MP representations.