A transformer-based reinforcement learning approach for scalable weapon target assignment

Abstract

We develop a novel reinforcement learning model, the attention-based reinforcement learning framework for assignment (ARFA), designed to optimize the weapon target assignment problem. Our ARFA model outperforms selected exact and heuristic algorithms in large problems. The ARFA model also demonstrates scalability and transferability, as it was trained on small configurations and performed well when applied to larger problems. This flexibility is achieved through the use of transformer architectures that adapt to varying problem dimensions while maintaining steady performance. Our model enhances computational efficiency and accelerates solution delivery by integrating parallel computation techniques with advanced reinforcement learning algorithms. It has been shown that ARFA can accelerate operational tempo in large-scale, high-tempo military operations.