Implicit and Explicit Communication for Human Oversight and Supervisory Control of Multi-robot Systems

Congratulations to Maria Stull on passing her thesis proposal! Teams of robots overseen by a human supervisor can operate at larger scales than single robots while still drawing on human judgment, but these teams scale only as far as the supervisor's cognition allows. Maria's thesis measures and relieves that constraint at three critical points in the supervisory loop: setting expectations for individual teammates, giving instructions the robots can act on, and handing a task back to the human operator.

Building on her prior work showing that identity cues help a supervisor tell robots apart and hold distinct expectations for each one, Maria proposes two new studies. The first develops a rubric of communication skills for scoring free-form instructions given to AI agents, and tests whether those scores predict task success, first in human–LLM teaming and then in robot supervision tasks. The second uses 3D reconstructions to speed task handover in multirobot supervision: in a simulated city-maintenance scenario, operators either classify hazards from 3D reconstructions instead of 2D photos, or explore an interactive 3D model of the area before taking control of a drone.

Committee: Bradley Hayes, Christoffer Heckman, Alessandro Roncone, Nikolaus Correll, and Daniel Szafir.