Distributed UAV Swarm Confrontation Decision-Making Based on Reinforcement Learning

With the rapid development of unmanned aerial vehicles (UAV) technology, the decision-making process in UAV swarm confrontation has become a critical research focus both domestically and internationally. To address the computational challenges posed by centralized decision-making methods, this paper introduces a decentralized framework for strategic decision-making in UAV swarm conflicts. First, this paper constructs a confrontation scenario consisting of multiple homogeneous and equal-numbered UAVs, and allocates a specific strike target to each UAV through a target allocation algorithm, transforming the multi-UAV combat into single UAV combat tasks. Then, the Deep Deterministic Policy Gradient (DDPG) algorithm is employed to train the decision-making model for the solitary UAV engagement. To markedly enhance the model's convergence rate, a strategy integrating reward shaping and curriculum learning is implemented. Moreover, the Artificial Potential Field (APF) method is employed to address the issue of collision prevention in multi-UAV operations. Ultimately, numerical simulation validates the effectiveness and scalability of the proposed approach.