Bottom-up social development through reproducing contingency with sensorimotor clustering

This paper presents a learning mechanism that nds a reasonable segmentation to achieve social behavior as well as that incrementally acquires it by reproducing the contingency in interactions with a caregiver. The robot autonomously categorizes sensorimotor activity according to a contingency measure based on transfer entropy. The advantage of adaptive categorization is tested in a task of acquiring joint attention behaviors. The results of computer simulations of human-robot interaction indicate that a robot acquires a series of joint attention behaviors such as gaze following and alternation and nds suitable segmentation over time that improves gaze following performance.