Learning by Scaffolding

I propose to build a robot that can engage in simple but meaningful social exchanges with humans. In contrast to current works in robotics that focus on robot-robot interactions (Billard & Dautenhahn 1997), this work explores human-robot interactions whereby a socially sophisticated human assists the robot in acquiring more sophisticated communication skills. In addition, the human helps the robot learn the meaning these acts have for others. Toward this end, my approach is inspired by the way an infant learns how to communicate with his caregiver. An infant’s emotions and drives play an important role in generating meaningful interactions with the caregiver (Bullowa 1979). These interactions constitute learning episodes for new communication behaviors. In particular, the infant is strongly biased to learn communication skills that result in having the caregiver satisfy the infant’s drives (Halliday 1975). The parent, in turn, is strongly biased to provide scaffolding acts for the infant which structures the interaction in a way that facilitates learning. Examples of scaffolding acts include drawing the infant’s attention to relevant aspects of the task, showing the him the desired goal state before he can reach it on his own, showing a positive emotional response when he’s progressing towards the goal state before he understands why he’s on the right track, and so forth. The infant’s emotional responses provide important cues which the caregiver uses to assess how to satiate the infant’s drives, and how to carefully adapt her scaffolding acts to best promote his learning. The former is critical for the infant to learn how his actions affect the caregiver’s behavior. The later is critical for establishing and maintaining a learning environment of suitable complexity for the infant where he is neither bored nor over-stimulated. Given these key insights from developmental psychology, this work proposes to incorporate them into a behavior engine that will enable an autonomous robot to behave socially and learn from social interactions with people. I present a framework for the behavior engine architecture that integrates perception, attention, behavior, motivation, and motor acts, and highlight the design aspects of each. I then discuss how learning is incorporated into the system and identify some key skills that can be learned with this approach. The proposed experiments involve teaching the robot these skills through face-to-face exchanges with the robot.