This page focuses on my PhD work at the University of Southern California. For an overview of past research projects I have been involved in at Yale, USC (as an undergraduate student), Heidelberg University, and Penn State, visit my other research page.

I am currently studying as a PhD student with Professor Maja Matarić at the Interaction Lab at the University of Southern California. My research is in the general area of socially assistive robotics, in which robots serve as therapeutic or assistive tools through their use of social interaction. Within this area, I am specifically interested in equipping robots with the social tools necessary to influence human behavior, both in terms of short-term interaction steering for more natural social interaction and improved functionality and in terms of achieving longer-term persuasive goals such as compliance with treatment or therapy. I am particularly interested in applying these socially assistive robotics approaches to one-on-one child-robot interaction.


Socially Assistive Robotics for Teaching Children About Nutrition Through Play

This ongoing project is part of the Socially Assistive Robotics NSF Expeditions grant and centers around an engaging one-on-one interaction between the MIT Personal Robotics Group's DragonBot robot and first-grade children. During the 6-session, 3-week interaction, the child and robot character interact in a hands-on nutrition learning task.

Machine Learning for Modeling the Interaction Between a Robot and a Child with Autism

This project focuses on data coding and analysis of an interaction between robots and children with autism, using data collected by David Feil-Seifer. The goal of this work is to begin to identify patterns in child-robot interaction that can be used to predict child social behaviors and identify robot behaviors that lead to child behaviors of interest (in the context of DIR/Floortime therapy, which we use as a basis for this experiment, these would be all social behaviors, regardless of appropriateness), as well as the context in which those behaviors should be used in order to achieve the desired outcomes.

In our initial work, we compare two machine learning techniques, one dynamic and one static, for use in predicting child vocalizations using unimodal and multimodal features, on a subset of the data. We find that using multimodal features and a dynamic classifier can achieve good (over 80%) classification accuracy for this task with leave-one-out testing, suggesting that such a classifier can achieve acceptable generalization across participants.

Current work focuses on predicting more general social behaviors and distinguishing between productive (leading to child social behaviors) and unproductive robot behaviors.

This project resulted in the following publication:

Short E, Feil-Seifer D, Matarić M. A Comparison of Machine Learning Techniques for Modeling Human-Robot Interaction with Children with Autism. The 6th ACM/IEEE International Conference on Human-Robot Interaction (2010):251-252. Poster paper.

Robot Coach for Post-Stroke Rehabilitation

In this project, we use the bandit upper-torso humanoid robot in a stroke rehabilitation setting. Using a button-board implementation of a reaching task, we look at the effect of different kinds of feedback on participants actual and perceived performance, as well as their perceptions of the robot and task.

Dragonbot Robot for Improving Nutritional Choices in Younger Children

This project takes the Dragonbot robot to schools in the Los Angeles area to interact one-on-one with first-graders over the course of several weeks. Following a story in which the robot (Chili) is training for a big upcoming race, the children have to choose foods that will make Chili stronger and faster.

Robot Behaviors for Influence

Inspired by results in psychology, we take behaviors that have been shown to be used by humans in social interactions in order to influence each other, and implement them on a robot. We plan to use the results to develop a library of parameterized robot behaviors that can influence human behavior.