Socially Developmental Robot based on Self-Induced Contingency with Multi Latencies

Abstract

Early social development is a process that a human infant and his/her caregiver adapt to each other. This paper presents a learning mechanism to find the contingency of human-robot interaction in the real world, which is intended to enable similar process to the mutual adaptation in the infant-caregiver interactions. A contingency measure based on information theory is applied not only to acquire behavior rules but also to find suitable latency to observe the found contingency. Experimental results show that a robot can acquire a series of social behavior such as gaze following and utterance to a human subject through 20 minutes interaction. Mutual adaptation between them is discussed in terms of transition and synchronization of their behavior, based on the analysis of the interaction data.