Designing Human-Robot Interaction based on/toward Understanding Parent-Infant Interaction
The goal of this project is to develop human-robot interaction through which a robot learns actions and languages from human partners as an infant does from his/her parents. Although the infants’ abilities to sense, process, and act are limited, which is comparable to robots, they are gifted at learning new skills. They can learn to perform goal-oriented actions and to use the mother tongue through interactions with the environment, especially with their parents. The key interest is in the evidence that parents greatly scaffold the infant development while infants induce such scaffolding. The immaturity of the infants’ abilities motivates their parents to properly teach them, e.g. adjust the difficulty of a taught task. The adjustment in parental teaching can result in the facilitation of infants’ learning. The project intends to build such an interaction system, by which a robot learner and a human teacher synergistically support robot learning.
There are three main issues in the project: (1) The first one is the analysis of parental teaching. Although the importance of parental teaching in infant learning has been suggested, it has not been revealed yet. We address the question of how parental teaching can aid infant and robot learning from the computational perspective. (2) The second issue focuses on the design of human-robot interaction. It is hoped that human partners properly teach a robot as parents do an infant. We intend to find out the factors that affect the interaction and to build a robot system that can induce parent-like teaching. (3) The third issue is to develop learning models for a robot. As the learning mechanism of infants seems to be adapted to parental teaching and vice versa, a robot should be able to take advantage of parental teaching. We aim at developing robotic models inspired by the infant abilities.