Research on Maneuver Decision Algorithm of Multi-UAV Based on Action Intention Reinforcement Learning

In multi-UAV air combat, traditional reinforcement learning (RL) struggles with high computational demands and policy learning inefficiencies arising from complex mission tasks and intensive inter-agent interactions. To address these limitations, this study proposes an intention-based RL framework that leverages prior knowledge to enhance learning efficiency and improve state-action mapping. Within this framework, a multi-UAV maneuver decision-making algorithm is developed, defining four action intentions - attack, surveillance, support, and evasion - to represent tactical objectives, each mapped to feasible maneuver actions based on situational data. The algorithm integrates a situational assessment model, cooperative target allocation, and a deep Q-network (DQN) to guide decision-making. Simulation results confirm the approach's effectiveness in enhancing multi-UAV maneuver coordination and decision performance.