Learning movement sequences from demonstration

摘要: Presents a control and learning architecture for humanoid robots designed acquiring movement skills in the context of imitation learning. Multiple levels abstraction occur across hierarchical structure architecture, finally leading to representation sequences within probabilistic framework. As its substrate, framework uses notion visuo-motor primitives, modules capable recognizing as well executing similar movements. This is heavily motivated by neuroscience evidence motor primitives mirror neurons. Experimental results from an implementation are presented involving demonstrated synthetic real human data.