Upcoming Event: Center for Autonomy Seminar

Scalable Multi-Agent Decision-Making via Environment Abstractions and Mean-Field Approximations

Scott Guan, Postdoctoral Fellow, Guggenheim School of Aerospace Engineering, Georgia Tech

Abstract

Modern multi-agent systems face two fundamental scalability challenges: the complexity of the environment and the growing number of interacting agents. In this talk, I present approaches that address these challenges through environment abstractions and mean-field approximations. First, I introduce a performance-driven environment abstraction framework for efficient planning and learning in large state spaces. I then examine an abstraction concealment problem that arises in competitive settings, where agents strategically hide their internal abstraction to avoid exploitation. Second, I address scalability in large-population team-versus-team scenarios by introducing zero-sum mean-field team games, which capture mixed competitive–cooperative interactions among large teams of agents. Building on structural results on optimal policy classes, I present MF-MAPPO, one of the first MARL algorithms for competitive games that scales to hundreds or even thousands of agents.

Biography

Scott Guan received the B.S., M.S., and Ph.D. degrees in Aerospace Engineering from the Georgia Institute of Technology in 2018, 2020, and 2025, respectively. He is currently a Postdoctoral Fellow at the Guggenheim School of Aerospace Engineering at Georgia Tech. His research focuses on scalable decision-making and learning for large multi-agent systems, leveraging tools from game theory, optimization, control, and machine learning. His work develops principled and provable methods for coordination and competition in large-scale cyber-physical and robotic systems.