Cooperative Maneuvering Decision-Making of Multi-UAVs Based on MADRL-VD

In order to make UAVs more effective in accomplishing complex combat missions, multi-UAV cooperative combat will become the most effective means. In this paper, we extract the basic elements of UAV combat and under certain assumptions, different flight states as well as action sets are constructed based on the principle of kinematics, and an air combat confrontation simulation platform is established. Under the changing confrontation environment, the Qatten algorithm is designed using the value decomposition method, and the UAVs are trained to take appropriate decision-making actions based on their own local observation situation, so as to enhance the capability of the UAVs’ autonomous cooperative confrontation. Our experiment takes the UAV 2 to 1 as the research scenario, and the results show that in the above scenario our algorithm can present a good fit.