Title: Proactivity and Personalization in Longitudinal Robotic Assistance

Date: Wednesday, July 1st, 2025

Time: 10:00am - noon ET

Location: Klaus 1120A or Zoom

Maithili Patel

Robotics Ph.D. Student

School of Interactive Computing

Georgia Institute of Technology

Committee:

• Dr. Sonia Chernova (advisor) – School of Interactive Computing, Georgia Institute of Technology
• Dr. Matthew Gombolay – School of Interactive Computing, Georgia Institute of Technology
• Dr. Mark Riedl – School of Interactive Computing, Georgia Institute of Technology
• Dr. Roberto Martín-Martín – School of Computer Science, University of Texas, Austin
• Dr. Jacob Andreas – Electrical Engineering and Computer Science, Massachusetts Institute of Technology

Abstract:

Robots are becoming increasingly capable of helping people with everyday tasks, but they remain reactive, waiting to be told what to do. This dissertation asks what it would take for a robot to become a proactive assistant that notices what users need and helps without being asked, the way a good roommate or colleague might. Achieving this requires a robot to anticipate a person's routines, personalize how it helps to match their preferences, adapt to dynamic environments and diverse tasks, and interact with users to gather information and minimize errors. This work builds a foundation for proactivity in such unstructured settings, beginning with a unifying formalism that connects existing efforts and defines the problem concretely. From there, it takes on the two central challenges of proactive assistance: anticipating what a user will do and personalizing how the robot helps. For anticipation, it formulates user-behavior prediction over action sequences and contributes graph-based and latent autoregressive models that forecast a person's future actions from unobtrusive observation, and a dataset capturing realistic longitudinal human behavior. For personalization, it contributes methods that generalize a user's preferences from limited feedback through abstract concept representations, and that actively elicit unspecified preferences through dialog. Finally, it studies how people respond to a proactive robot during an open-ended creative task, surfacing design insights to guide future robot design. Together, these contributions establish a foundation for proactive robot assistance and map out the open computational problems that future research in the field must address. The dissertation charts a course toward robots that don't just follow instructions, but take initiative in fulfilling human needs, taking on not only the physical, but also the mental burdens of everyday life.