Learning from Demonstration in Robots using the Shared Circuits Model

Abstract

Learning from demonstration presents an alternative method for programming robots for different nontrivial behaviors. Various techniques that address learning from demonstration in robots have been proposed but those do not scale up well. Thus there is a need to discover novel solutions to this problem. Given that the basic idea for such learning comes from nature in the form of imitation in few animals, it makes perfect sense to take advantage of the rigorous study of imitative learning available in relevant natural sciences. In this work a solution for robot learning from a relatively recent theory from natural sciences called the Shared Circuits Model, is sought. Shared Circuits Model theory is a comprehensive, multidiscipline representative theory. It is a modern synthesis that brings together different theories that explain imitation and other related social functions originating from various sciences. This paper attempts to import the shared circuits model to robotics for learning from demonstration. Specifically it: (1) expresses shared circuits model in a software design nomenclature; (2) heuristically extends the basic specification of Shared Circuits Model to implement a working imitative learning system; (3) applies the extended model on mobile robot navigation in a simulated indoor environment; and (4) attempts to validate the shared circuits model theory in the context of imitative learning. Results show that an extremely simple implementation of a theoretically sound theory, the shared circuits model, offers a realistic solution for robot learning from demonstration of nontrivial tasks.