Conceptual imitation learning based on functional effects of action

Abstract

In General, Learning new skills by imitation is faster, safer, and more efficient. In robotics research, imitation also provides an implicit and user-friendly mechanism for robot programming. But, according to the research in neuroscience and cognitive science, true imitation is accompanied by abstraction and conceptualization. This paper presents a method for autonomous acquisition, generalization, recognition, and regeneration of abstract (relational) concepts through perception of spatiotemporal demonstrations and identification of their functional effects. In fact, the effects help to classify the concepts based on their functional properties. As a result, the concepts are represented by prototypes which abstract different perceptual variants of a concept but make similar functional effects. Performance of the proposed algorithm is evaluated in a task of imitating a bunch of behaviors based on their emotional effects. Results of the experiments on a humanoid robot show that our model is successful for extraction, abstract representation, accurate recognition, and reproduction of the learned concepts.