One of my favorite research projects involves the development of algorithms for socially aware robot navigation, the problem of enabling a robot to traverse crowded human environments while accounting for human safety and comfort. One of my most exciting moments has been the formalization of the coupling between human and robot motion in crowded environments using the formalism of topological braids from low-dimensional topology. Prior research in this space followed a conventional predict-then-act approach, treating pedestrians as dynamic obstacles that are not reactive to the robot; this resulted in practical issues like the "freezing robot problem" and the "reciprocal dance". I noticed that the trajectories of multiple agents navigating in close proximity entangle in a pattern that exhibits topological properties. Critical information like passing side and order can be formally captured into braid words, symbols with geometric and algebraic descriptions. This allows a robot to reason at a symbolic level about the way its own path gets entangled with the paths of other agents around it. Based on this representation, I developed a decision-making framework that treats the task of socially aware navigation as coordination (between the robot and surrounding humans) over a set of topological primitives. This computational idea has deep backing on studies from social sciences, suggesting that social order in pedestrian navigation is an artifact of negotiation, taking place via nonverbal signaling encoded in human motion and social cues. Empirically, I have showed that when the human-robot motion coupling is explicitly accounted for at a topological level in the robot’s decision making, no precise models for human motion prediction are required to empower robust robot navigation performance in crowded indoor environments. In a user study with more than 100 people, I demonstrated that this framework can empower a mobile robot to harmoniously operate in a shared workspace with a group of human users attending to their own tasks. My study design, comprising objective and subjective factors, and user distractors, has set the foundation for the development of modern benchmarking practices.
I started my research path at the National Technical University of Athens in Greece, where I received a Diploma in Mechanical Engineering (2013). My thesis focused on grasp planning algorithms for multi fingered robot hands. Then I moved to Cornell, where I got my PhD (2019), working on algorithms for social robot navigation in crowded environments. Then I moved to the University of Washington as a postdoc, growing in the areas of human-robot collaboration and multiagent systems. Since 2023 I have been an Assistant Professor of Robotics at Michigan where I run the Fluent Robotics Lab.