MechE PhD Prospectus Defense: Kamran Vakil

TITLE: Scalable and Interactive Planning for Multi-Agent Systems

ABSTRACT: In recent years, robots have been deployed for applications such as driving, package deliveries, exploration, and more. However, fully autonomous robots often remain out of scope for real world deployments due to the high compute requirement necessary for intelligent decision making. One major component of this decision making is the concept of interactivity, where a robot considers how the action of any agent in a multi-agent system impacts the actions of others, including itself. We explore interactivity through the field of game theory, allowing robots to make intelligent decisions based on their own personal needs or the needs of the system as a whole. This prospectus focuses on multi-agent systems on two levels: 1) low level path planning for self-interested interactive agents under uncertainty and 2) high level heterogeneous drone allocation for multi-drone multi-depot systems. We first present how self-interested agents can selectively negotiate with other agents when modeling, and how this selective negotiation lowers computation time while minimizing performance loss. Then, we present how agents can partially model uncertainty when planning paths, and how this partial modeling once again lowers computation time while minimizing performance loss. Next, we focus on high level heterogeneous drone allocation for multi-drone multi-depot systems, and derive a model and control algorithm based on concepts from coverage control to satisfy events over a long time horizon. Finally, we detail current and future work within these areas of research.

COMMITTEE: ADVISOR/CHAIR Professor Alyssa Pierson, ME/SE; Professor Roberto Tron, ME/SE; Professor Sean Andersson, ME/SE