Autonomous Decision-Making Algorithm for Multi-agent Beyond-Visual-Range Air Combat

Beyond-visual-range air combat is the mainstream form of air combat in modern air battlefields. The decision-making ability of pilots will determine the result of the war. With the development of artificial intelligence, the research on autonomous decision-making methods for beyond-visual-range (BVR) air combat has become a focus of intelligent air combat. The complex battlefield environment and tactical actions in BVR air combat make this issue full of challenges. This paper proposes a novel multi-agent hierarchical decision-making network based on self game theory to decompose complex air combat tasks, effectively reducing the ambiguity of tactical actions. In addition, the algorithm adopts self-play to reduce the meaningless exploration of agents due to the large battlefield environment. Compared with other multi-agent reinforcement learning algorithms, it has been proven that the algorithm can not only help agents learn basic flight tactical actions and advanced combat tactical actions, but also outperforms the state-of-the-art multi-agent BVR air combat algorithms in terms of both defense and offense ability.