Robotics and Autonomous Systems Center (RASC) Seminar

Speaker: Prof. Momotaz Begum, University of New Hampshire (UNH)

Talk Title: Imitation Learning for Home Robots: Promise, Truth, and Tension

Abstract: Imitation learning algorithms pass the ultimate test when a grandma in a rural town in America can seamlessly teach her robot how to clean up her kitchen or make her tea — no sugar, two spoons of milk. Getting there requires solving a chain of hard problems that span, among many others, data quality, policy robustness, and run-time safety. In this talk, I discuss these challenges through the lens of long-term, real-world robot deployment — a perspective that is rare in the field and that has shaped every research decision in my lab.Specifically, I will present three lines of work that together address a core bottleneck: making imitation learning robust and safe enough to survive contact with real users in real homes. The first examines how to learn reliable policies from demonstrations collected by lay users — non-experts whose data is noisy, inconsistent, and is not available on demand. The second introduces safety watchdogs built on invariant sets and formal logic that enforce run-time safety guarantees for open-loop BC policies, without retraining. The third presents a one-shot automated demonstration synthesis framework that reduces the data collection burden while preserving policy quality. Together, they chart a path from laboratory imitation learning toward systems that are genuinely field-deployable.

Biography: Momotaz Begum is an Associate Professor of Computer Science at the University of New Hampshire (UNH) and directs the Cognitive Assistive Robotics Lab. Her research sits at the intersection of robust machine learning and assistive robotics, with a focus on building embodied AI systems that hold up outside the lab — in homes, care facilities, and the hands of everyday users. This commitment to real-world deployment over more than a decade distinguishes her work and grounds her perspective on what imitation learning must actually deliver. Her research is supported by the National Science Foundation and the National Institutes of Health. Momotaz received her PhD in Cognitive Robotics from the University of Waterloo, Canada, and completed postdoctoral fellowships in Human-Robot Interaction at the Georgia Institute of Technology and the University of Toronto, before joining UNH in 2016.

Host: Erdem Biyik

More Info: URL

Webcast: URL