Optimizing UAV-UGV System for Reconnaissance and Strike Based on Soft Actor Critic Algorithm

In this paper, reinforcement learning is used to solve the reconnaissance and strike tasks of UGVs and UAVs in the battlefield environment, and different types of agents are trained separately under the premise of environmental interaction. As a mature reinforcement learning algorithm, SAC algorithm has been widely used in the training of single agents. By assigning part of the state of the UGV to the UAV, the problem that the fully distributed structure does not converge is solved. For the actual battlefield environment, UGVs and UAVs need to undertake different tasks. The UAV provides accompanying vigilance for UGV and the main reconnaissance task of the target area. The UGV should go to different target areas according to task priority order to carry out auxiliary reconnaissance and strike tasks after reconnaissance.