Infant development sequences for shaping sensorimotor learning in humanoid robots

Early infancy is a time of remarkable learning and cognitive growth. Developmental psychologists have observed that very distinct stages in behaviour are related to such growth spurts. This paper presents the findings of a detailed review of the first 12 months of infant sensory and motor development, with the aim of providing an explicit framework that can guide the design of similar developmental processes in robotics. We show how human sensor and motor development sequences can be mapped on to robotic platforms and how constraints on perception and action can be applied or released in organized sequences so that staged behaviour and learning may take place. This approach, of using constraints as a scaffold, can effectively manage the complexity of learning a set of increasing sensorimotor skills, as in the human infant. We give examples of constraint structures that we are applying to an iCub humanoid robot.