Short-range air combat maneuver decision of UAV swarm based on multi-agent Transformer introducing virtual objects

With the development of Unmanned Aerial Vehicle (UAV) swarm technology, there has been a growing interest in using Artificial Intelligence (AI) to drive UAV swarms for short-range air combat. However, due to the complexity of situation information in UAV swarm air combat, making accurate air combat decisions based on air combat situation information has become a challenge. In this paper, we propose the multi-agent Transformer introducing virtual objects (MTVO) to address this issue. First, the proposed approach designs a multi-agent Transformer network structure by exploiting the homogeneity feature of the UAV state information in the swarm. This structure enables the structured processing of complex situation information. Specifically, the local situation information of each UAV is calculated by self-attention, which reduces the size of the swarm situation information while retaining the information of key UAVs. This approach reduces the difficulty of processing UAV swarm situation information. Moreover, we add a virtual object to the UAV swarm information to assist in calculating the weight distribution of the local situation. The weighted fusion of local situations allows us to obtain a more effective representation of the global situation, which serves as the basis for more accurate air combat maneuver decisions. We demonstrate the performance of the proposed method through air combat simulation results using Reinforcement Learning (RL) methods, and validate the applicability and effectiveness of the MTVO network for the short-range air combat problem of UAV swarms.