Hierarchical Reinforcement Learning-Based Decision Generation for Beyond Visual Range Air Combat

UAVs are now platforms that can carry out activities on their own thanks to technological advancements. This research develops a hierarchical reinforcement learning-based decision-making framework to tackle the decision-making difficulties encountered by UAVs in air combat. An approach based on the LSTM-DQN model for UAV air combat decision-making is suggested. To overcome challenges like high-dimensional space, prolonged exploration, and scarce rewards in traditional reinforcement learning, the approach splits the air combat process into a decision layer and a control layer. The behavior tree governs how action sequences are carried out in the control layer, while the enhanced DQN model is in charge of task generation in the decision layer. This framework increases training efficiency, decreases the dimensionality of the state space, and shortens exploration time. Win-rate analysis in a simulated environment is used to verify the algorithm’s efficacy and the model’s viability.