Human-robot interaction learning using timed automata

Abstract

A new unsupervised learning algorithm of human-robot interaction for behavior planning unit of a scrub nurse robot is proposed in this paper. The algorithm constructs a composition of timed IO automata where each automaton represents behavior of an interaction party. The learning architecture of the scrub nurse robot and its incremental learning cycle are discussed. The automatic compilation of the interaction model is guided parametrically in the learning process that allows generating models of different level of abstraction and profile. The approach is illustrated with an example of learning a fragment of laparoscopic surgical procedure.