I build personalized adaptive systems that enable robots to interact naturally and efficiently with users over weeks and months. I take inspiration from human theories of emotion and social relations to develop computational models that support the complex dynamics of social interactions between users and robots. These models rely on novel computational algorithms and machine learning techniques that allow robots to adapt their behavior to the particular context, user needs and preferences over time.

Socially Assistive Robotics


As part of the NSF expedition in Socially Assistive Robotics, I have been collaborating with the Yale Center for Emotional Intelligence to understand how robots can help children acquire social and emotional skills. In this context, I am investigating whether the advantages of one-to-one tutoring can also apply to one-to-many instruction, and what costs might be incurred when this shift happens.
back to top

Long-term Human-Robot Interaction

icat-chessMost of the existing social agents and robots are unable to keep users engaged over repeated interactions. One of my main research areas is to investigate what social and affective mechanisms are more effective for keeping users engaged with robots once the novelty effect fades away.
back to top

Multimodal Perception of Social Behavior

affect-detectionI study how multimodal perception systems can be used to classify naturalistic behaviors form users (mostly children) such as affect, turn-taking or engagement, using data collected in real-world environments. I have also been investigating how machine learning based models perform when they are tested in a group size different than the one they were trained on.
back to top

Computational Empathy

icat-2pDuring my PhD, I proposed, implemented and evaluated a computational model of empathy for robots that interact with users over repeated interactions. My thesis was the first in the literature to extensively evaluate how children interact with an autonomous robot over repeated interactions.
back to top

Adaptive Social Interactions

icat-RLThe question of how robots can adapt to different users in a natural way is becoming increasingly more relevant. I investigated a nonintrusive, self-regulating adaptive approach that enables robots to learn the best strategies for keeping the user in a positive affective state. Results from a longitudinal study suggest that affective expressions can serve as an informative feedback mechanism for personalizing interactions, especially for extrovert children. 
back to top


back to top