Multi-UAV Distributed Expulsion Based on Deep Reinforcement Learning

Aiming at the problem of collaborative expulsion in multiple unmanned aerial vehicles (UAVs), a method for maneuver decision-making based on deep reinforcement learning is proposed. Firstly, a reward function is designed based on the proximity of the single UAV to the enemy UAV, the constraints of task boundary, and the effectiveness of group expulsion. A framework for training multi-UAV collaborative expulsion is built on such a basis. Meanwhile, the latest deep reinforcement learning algorithm is enhanced based on a training mechanism of the shared feature encoding network. The simulation results show the rationality of the collaborative expulsion reward function design, the effectiveness of the maneuvering decision-making method, and the superiority of the training mechanism of the shared feature encoding network.