Multi-Agent Game Strategies for Kill Chain Optimization in Networked Aerial Combat Systems

Abstract

In modern information warfare, single-agent and multi-agent systems (MAS) play a critical role in achieving integrated combat capabilities. This study examines agent-based game strategies in airborne networked systems, focusing specifically on the air-fleet kill chain—the central framework that unifies fighters, early-warning aircraft, and missiles into a cohesive, intelligent system. We analyse how MAS mitigate the complexity, uncertainty, and adversarial dynamics of aerial combat by optimising detection, decision-making, and strike efficiency through networked collaboration. Two representative scenarios are presented: (1) an AI-driven fighter (ALPHA AI) using genetic-fuzzy tree algorithms to surpass human pilots, and (2) adversarial multi-agent reinforcement learning (MADDPG) in OpenAI's simulation suite. We then propose a systematic MAS-based kill-chain optimisation design that integrates deep reinforcement learning, Bayesian inference, and tactical decision frameworks. Simulation results demonstrate enhanced coordination, real-time adaptability, and optimised damage probabilities in both single- and multi-agent confrontations. Our findings establish a theoretical foundation for transitioning from rule-based to AI-driven system-of-systems warfare in next-generation aerial combat.