Cooperative Encirclement Strategy for Multiple Drones Based on ATT-MADDPG

Aiming to address the challenges related to robustness, adaptability, and cooperative performance in the context of multi-drone cooperative encirclement of airborne evasive targets. We propose a cooperative encirclement strategy based on multi-agent reinforcement learning. Firstly, the environment and kinematic model for the encirclement problem are established, taking the context of drone cooperative attack as the background. The criteria for successful encirclement are provided. Secondly, this paper proposes a Markov Decision Process framework based on the Attention-based Multi-Agent Deep Deterministic Policy Gradient (ATT-MADDPG) algorithm. The state space, action space for encirclement distance, and a reward function combining capture and step rewards are designed according to the requirements of the encirclement task. Finally, the encirclement strategy is trained using a centralized training and distributed execution architecture, where the drone swarm shares the same policy and independently executes actions. Simulation experiments demonstrate the superior performance of this approach in achieving successful encirclement compared to the MADDPG algorithm. Furthermore, the proposed algorithm effectively navigates obstacles while efficiently capturing the target, thus highlighting its adaptability in complex environments.