AAV Weapon Target Assignment via State Split Flow Deep Q-learning Network with Threat Degree Index

This article investigates the weapon target assignment (WTA) problem for the autonomous aerial vehicle (AAV) system in the safety transition task. Considering the heterogeneity of targets, a reinforcement learning environment is constructed for the problem. The aim of WTA is to perceive the threat of targets and then select proper ammunition to destroy hostile targets. To evaluate threat levels of heterogeneous hostile targets, the threat degree index is designed based on distances, speeds and heading cross angles. Acquiring the WTA policies depends on maximizing the worth reward of killing a high-threat target and minimizing the ammunition cost. Taking into account the temporal relationship of AAV actions, a state split flow deep Q-learning network algorithm is proposed to approximate the optimal policy and train the AAV with a multi-process approach. The simulation examples and comparisons are given to validate the effectiveness and advantages of the presented method.