I develop robots that shape human interactions in complex and dynamic environments throughout day-to-day tasks. I do this by building robots that create user models and give personalized feedback for long-term skill acquisition.
Computational Models of User's Skills
During my Ph.D. I have created algorithms that continuously model a user’s skills during complex tasks. I have applied my algorithms to electronic circuit building tasks. I integrated user skill modeling algorithms and optimal action selection into a model that selects the best task to hand a user to maximize learning or to maximize novel information about the user’s skill state. By continuously modeling the user throughout task progression, I enabled the robot to personalize its help actions to the user’s skill level. The algorithm I designed significantly increased the user’s knowledge of electronic circuits from before to after the interaction.
Peer Robots for Tutoring
I investigate different robot roles when tutoring adults. My research had compared a peer robot to a traditional teacher robot by switching the pronouns the robot used. Participants viewed the peer robot as more intelligent, friendly, and respectful than its tutor counterpart. Furthermore, people who had low prior knowledge in the domain learned significantly more with the peer robot than the tutor robot
My current research expands on my previous work by enabling the peer robot to not only tutor individuals but also to teach a group of people. Many workplaces are collaborative, and having a robot that can help a group and enhance each member's capabilities could greatly benefit the team. For a robot to be an effective peer, it will need to create individual models of each user's skills while they collaborate on tasks. I construct robot systems that can detect the strength and weaknesses of each team member and create a collaborative environment where the robot encourages team members to share their knowledge with each other.
Robots That Positively Influence People
I have investigated different ways that robots influence people, with a focus on robot group conformity. My research shows that a group of robots can indirectly influence people using conformity. Furthermore, my experiments show that robots cause both normative and informational conformity.
We investigated how to use robot conformity to positively influence people. Our research shows that a group of robots that show disagreement (in the form of sadness) with a robot being mistreated significantly increase the number of people who intervene to stop the mistreatment. These studies leave many open research directions about how to further create robots that can positively influence society.
Shaping Human Behavior In Long-term In-Home Studies
During my research I have built two in-home long-term systems:
(1) We have developed a robotic system that motivated people to exercise daily in their homes. A machine learning algorithm classified each person's exercise movements as either correct or a pre-defined mistake, such that the robot could provide personalized feedback. The robotic system successfully motivated people to exercise daily, decreased the amount of errors people made over time, and many participants planned to continue exercising post-study.
(2) We built a novel and completely autonomous in-home long-term system where a socially assistive robot tutored children with ASD (Autism Spectrum Disorder). Children and their caregivers played daily games with the robot that trained social, perspective-taking, and sequencing skills. The robot encouraged engagement, personalized the difficulty of the activities, and modeled positive social skills. It was the first study that showed that robots could significantly improve the social skills of children with ASD using clinically verified measures. Furthermore, parents reported seeing significant improvements in their children's social skills due to the system's training.