Computational and Applied Mathematics seminar

Abstrakt finns enbart på engelska: Small and inexpensive drones are increasingly used in surveillance, reconnaissance, and attack operations. Future drones are expected to operate autonomously in large swarms, posing challenges for traditional Surface-Based Air Defense (SBAD) systems, which rely on radar sensors and effectors like missiles or jamming. Swarms can overwhelm SBAD systems due to their sheer numbers and economic asymmetry, as drones are significantly cheaper than defense systems. A potential countermeasure is deploying defensive swarms of small drones with sensors and effectors, necessitating optimized swarm behavior and capability evaluation. The proposed project focuses on decentralized control of such swarms using game theory, addressing challenges in operational goal representation, situational awareness through sensing and communication, and managing uncertainty about enemy capabilities. These challenges are modeled by Partially Observable Stochastic Games. Approximation techniques like Multi-Agent Reinforcement Learning are explored, leveraging algorithms like MADDPG and MAAC for mixed competitive and collaborative swarm-versus-swarm scenarios. While related works address components of the problem, they fall short of addressing the full complexity, particularly in handling unknown drone numbers, controlled sensing, communication, and intelligent adversaries in a competitive game-theoretic framework.