Cooperative Reinforcement Learning for Military Drones over Large-Scale Battlefields

Abstract

The development of autonomous systems in military drone operations signifies a paradigm shift in modern warfare strategies, and modern warfare emphasizes the crucial role of military drones to the extent that it is called drone war. Despite the importance of these military drones, the efficiency of such operations hinges on the swarm military drones' ability to operate cooperatively in complex battlefield environments where the target changes dynamically. The proposed algorithm for robust and reliable military drones needs to consider a total of three objectives, i.e., (i) cooperative swarm flight maintenance, (ii) target bombing zone maximization, and (iii) energy efficiency optimization. For this objective, multi-agent deep reinforcement learning (MARL) is utilized for the cooperation of military drones to achieve these objectives optimally. Lastly, this paper considers 'Shahed-136' which is one of well-known military drones for realistic battlefield evaluation. Additionally, the military drones employing the proposed algorithm are evaluated in a large-scale battlefield environment spanning 4×104 km3 . This paper assesses the efficacy of the proposed approach through simulations that evaluate the swarm flight, success rate of bombing, and energy efficiency of drones, particularly under dynamic target conditions