Replication of Infant Behaviours with a Babybot: Early Pointing Gesture Comprehension

Natural deictic communication with humanoid robots requires a mechanism for understanding pointing gestures. This mechanism should have a representation for space and time dynamics to accurately model joint covert attention. Here, we introduce a babybot that actualise a hybrid computational architecture for spatial covert attention which is embodied in the iCub humanoid robot. This developmental robotics architecture was an extension of our previous model that combines a connectionist model of pointing comprehension and a dynamic neural field model of infant saccade generation. In order to test the babybot's abilities, an attentional cueing design was built as it is a common methodology to study pointing gesture comprehension in the current developmental psychology literature. The babybot was evaluated by modelling two different age groups (i.e. 5- and 7-month-old infants) in two different attentional cueing experiments from Rohlfing et al.'s study where they have shown that a dynamic pointing covertly orients attention in early infancy as opposed to static pointing. These experiments were replicated by our babybot for all modelled age groups. The resemblance between infant and babybot behaviours supported the idea that motion information is important to disengage from the centrally salient stimulus to orient the attention to the distal ones. The current experimental setup and the model's ability to simulate different age groups provide a new platform to replicate other developmental studies on pointing comprehension to spotlight the reasons of the discrepancies between them.