Postdoctoral Researcher, University of Cambridge
Hi! I'm Michael, a postdoctoral researcher affiliated with the Prorok Lab and Trinity College.
I develop algorithms for large-scale multi-agent and multi-robot systems. Currently, I am focused on achieving scalable, safe, and efficient coordination by blending machine learning with rigorous mathematical guarantees. My previous work during my PhD explored multi-agent systems operating on graphs (see my PhD thesis).
Explore many more publications on my Google Scholar profile.
ReCoDe: Reinforcement Learning-based Dynamic Constraint Design for Multi-Agent Coordination
We introduced ReCoDe, a novel RL framework that learns to dynamically generate constraints, rather than actions, to improve team coordination in real-time. This allows agent teams to benefit from expert controllers that determine their next action subject to these constraints.
Time, Travel, and Energy in the Uniform Dispersion Problem
This work explores the fundamental efficiency limits of robot swarms. We prove when it's possible (or impossible!) to simultaneously optimize for time, distance, and energy during dispersion, and introduce FCDFS, an ant-robotics algorithm that achieves these bounds using only 5 bits of memory and zero communication.
When Is Diversity Rewarded in Cooperative Multi-Agent Learning?
Our latest preprint investigates a critical but often misunderstood aspect of MARL: behavioral diversity. We identify reward structures where promoting diverse agent behaviors leads to superior team performance--and ones where it doesn't.