Use of speech and motion cues for bootstrapping complex action learning in iCub

Abstract

Parental scaffolding is an important mechanism that speeds up infant sensorimotor development. Infants pay stronger attention to the features of the objects highlighted by parents, and their skills develop earlier than they would in isolation due to caregivers support. Parents are known to make modifications in infant-directed actions, called “motionese”, which is characterized by a wider range of motion, repetitive actions, and longer and more pauses between movements. Inspired from motionese, we previously realized a robotic system [1] where the affordances and effect prediction capabilities that are learned in the previous stages of development are used to bootstrap complex imitation and action learning with the help a cooperative tutor through motionese. With this system, a robot could learn new skills via imitation learning by extracting the important steps from the observed movement trajectory, and then encoding them as subgoals that it can fulfill. Considering the affordances provided by the objects in the environment, it found and sequentially executed the actions that are predicted to generate the desired effects and achieve the subgoals; achieving the overall goal of complete imitation. We showed that motionese can be used to bridge the gap between the interacting agents with different movement capabilities, such as the human tutors and the arm-hand robot we employed. Furthermore, our experimental data indicated that naïve tutors who are not informed about the imitation mechanisms of the robot, changed their teaching strategy, and started to display motionese.