Evolution of combinatorial structure in learned forms through embodied iterated learning in a robot collective

Abstract

In this study, we use mobile robots as physical entities to model the iterated learning of collections of forms that consist of randomly generated movement sequences. The robots implement an abstract model of embodied iterated social learning in which the forms evolve due to limited perceptual abilities of the robots during multiple learning cycles. It is shown that shared chunks that consisted of similar movement sequences emerge in the learned forms, and as these emergent shared sequences can be learned with high accuracy, they cause a cumulative increase in the learnability of the collections. Therefore, we are able to present robotic experiments in which embodied learning on robots leads to combinatorial structure as a result of cultural interactions in the form of iterated learning without a communicative task.