Robotics research has a potential to yield new insights about infant development. Investigating neural models for robots to acquire social behaviors could lead to a better understanding of underlying mechanisms for human cognition. We suggest from a perspective of robotics that predictive learning of sensorimotor information plays a key role in infant development. Various types of cognitive abilities such as self/other cognition and joint attention can be acquired by predictive learning, which is a process to identify internal models to represent sensorimotor coordination by minimizing prediction errors.
My talk will present our computational models to support our hypothesis. The importance of caregivers' scaffolding for infant development is also demonstrated in human-robot interaction experiments. Furthermore, our hypothesis is extended to explain a cause of social deficits in autism spectrum disorder (ASD): An abnormal tolerance for prediction errors may produce different internal models for ASD and thus result in difficulties in social communication.
Keywords: Cognitive developmental robotics, infant development, self/other cognition, joint attention, human-robot interaction