Motor development facilitates the prediction of others' actions through sensorimotor predictive learning

Recent studies in psychology revealed that the emergence of infants' ability to predict others' action goals is correlated with the development of their motor ability to produce similar actions. In this regard, studies in neuroscience suggest that perception and production of actions share the same neural architecture (i.e., mirror neuron system). However, it is not yet clear what learning mechanisms are involved in the co-development of these abilities. Here we proposed a computational model to explain the development of prediction of others' action goals in synchronization with the development of action production. We adopted the concept of predictive learning of sensorimotor information as the key mechanism for it. Predictive learning intends to associate motor signals with sensory signals in a predictive manner during action executions. Thus, sensory signals perceived during action observations induce corresponding motor signals obtained in previous experiences. Our experimental results showed that our approach facilitated a robot to develop the ability to predict action goals in synchrony with the development of action production. Furthermore, our experiments demonstrated that the integration of goal-directed motor signals improved the accuracy to predict sensory signals and consequently action goals.