Personalized Robot Storytelling Companion

Could a social robot collaboratively exchange stories with children as a peer and help improve their linguistic and storytelling skills? We use machine learning algorithms to develop  Companion AI  to make robots learn to be helpful to young learners. Our robots learn actions that improve children's storytelling and keep them engaged.  We are also interested in how a social robot can personalize its interaction with each child over multiple encounters, because every child learns and engages differently. 

In Fall 2019, we sent 18 Jibo robots to kindergarten classrooms in Atlanta. Most children in these schools come from less privileged neighborhoods, and the main focus is in prepping children with basic literacy skills, so they're ready to learn when they get to each grade level. This ongoing project will last for the whole school year, during which time our robots will provide one-on-one, personalized story-time interaction for the young readers. Please inspect our website for teachers and parents to find detailed information about the activities.

Of course, this work would not have been possible without our robot pioneer, Tega, who in Spring 2017 went to twelve preschool classrooms in the Greater Boston area for three months, making a big impact to the field of long-term human-robot interaction. Using Q-learning, a policy was trained to tell stories optimized for each child’s engagement and linguistic skill progression. Tega monitored children's affect signals and asked dialogic questions during storytelling to gauge their engagement. Tega also invited children to tell it stories, which Tega used to assess each child's linguistic skill development. Our results show robot's interaction policy indeed personalized to each child. At the end of the sessions, the policy significantly differed from one child to the other. Children who interacted and built relationships with a personalized robot showed higher engagement, learned and retained more vocabularies, and used more complex syntax structure in their speech compared to where they had started.


MIT Media Lab/Personal Robots Group

As can be seen from our results, social robot learning companions hold great promises for augmenting the efforts of parents and teachers to promote learning, academic knowledge and the well-being of children. Social robots can physically, socially, and emotionally play and engage with children in the real world. They can be designed to interact with children in a collaborative, peer-like way during playful educational activities. The interactions between a child and a robot resemble the speech acts between children and adults or peers, and offer a unique opportunity to personalize social interactions to promote areas of development important for learning. When a child enters Kindergarten, she is a unique distribution of the various cognitive, visual, social and linguistic skills needed to be a successful reader. However, in at-risk communities, it is almost impossible for a Kindergarten teacher to offer a curriculum that addresses the diverse cognitive and pre-literacy starting points upon which children enter school. Young children would clearly benefit from personalized instruction that can measure and adapt to many intersecting domains of skills and abilities during the process of learning to read and storytell. That's where our research contribution lies.