Robotics Engineering Colloquium Speaking Series: Professor Peter Crowley-Dolen
12:00 p.m. to 1:00 p.m.
Multi-Agent Reinforcement Learning for Safe and Decentralized Swarm Coordination and Control
Abstract: Robot swarms are increasingly being leveraged across a variety of environments and domains such as farming, search and rescue, delivery, and defense. Effectively fielding such swarms requires autonomous coordination within the swarm, enabling a single operator to command complex, coordinated behaviors rather than individually controlling each robot. Reinforcement Learning (RL) has demonstrated success in producing high-performing multi-agent coordination. However, applying RL to robot swarm control requires a robust approach that enables decentralized, scalable, and safe decision-making. In this talk, we discuss Discrete Graph Control Barrier Function Proximal Policy Optimization (DGPPO), a graph neural network-based multi-agent RL algorithm that satisfies the requirements of safety, scalability, and decentralized execution. Additionally, we present a real-world deployment of DGPPO in a maritime asset defense experiment and discuss the key implementation details that were critical to its success.
Bio: Peter Crowley-Dolen is a Technical Staff member and an autonomous systems research scientist at MIT Lincoln Laboratory. He received his Ph.D. degree in Mechanical Engineering in 2026 from Boston University, where he studied multi-agent reinforcement learning in the BU Robotics Laboratory. He received his MS degree in Mechanical Engineering in 2025 from Boston University, and his BS in Mechanical Engineering from the University of Connecticut in 2021. His current research interests include intrinsically motivated reinforcement learning, and the application of decentralized multi-agent reinforcement learning to adversarial games and maritime robotics.