MADDPG-Based Multi-UAV Autonomous Collaborative Attack in Confrontation Scenarios

In some military confrontation scenarios, to cope with the sudden intrusion of enemy attackable aircraft, our Unmanned Aerial Vehicles (UAVs) should have the autonomous interception capability. In particular, when the kinematics of the enemy aircraft is stronger than ours, our side need to send multiple UAVs to strike and intercept. Therefore, this paper proposes a novel MADDPG method, MADDPG with Experience Screening Mechanism (ESM-MADDPG), in the context of typical two-on-one Multi-UAV Autonomous Collaborative Attack (MUACA) mission. To realize MUACA, firstly, a new set of input setting scheme is proposed based on UAV kinematic characteristics and mission characteristics; secondly, in order to enhance the training effect, a new experience screening mechanism is proposed on the basis of MADDPG. To verify the effectiveness of the ESM-MADDPG, training experiments and testing experiments are conducted in a three-dimensional environment. The experimental results show that the ESM-MADDPG proposed in this paper can realize the MUACA mission and performs better compared with MADDPG, MADDPG with Destroyed Experience Discards (DED-MADDPG). Among them, DED-MADDPG is the method that partially include the ESM-MADDPG strategy.