Self-Similar Formation Control of Multiple UAVs Based on G-MIXDDPG

Unmanned aerial vehicle (UAV) formations offer significant advantages in managing complex and multitask scenarios that exceed the capabilities of a single UAV. Despite the existence of various UAV formation strategies, current approaches struggle with slow convergence, limited scalability, and inefficiencies in dynamic environments. To address these challenges, we propose a UAV formation model based on Sierpinski's fractal, designed to meet the specific requirements of formation gathering and rapid separation. Our approach integrates heterogeneous information and a multi-agent deep deterministic policy gradient (MADDPG) framework, enhanced with a deep deterministic policy gradient (DDPG) and an attention mechanism, to optimize cluster control strategies. Moreover, we introduce a novel G-MIXDDPG algorithm based on Sierpinski fractals, which improves convergence speed and adapts to a larger number of agents. Experiments demonstrate that G-MIXDDPG achieves a 63.13%, 21.63%, and 19.71% higher aggregation success rate compared to DDPG, MADDPG, and MATD3, respectively, and reduces aggregation time by 8s and 25s compared to back-stepping and MPC methods. This work significantly advances UAV formation control, offering a practical and scalable solution for multitask scenarios in dynamic environments.