基于MIX-MAPPO算法的多无人机编队控制  

Multiple UAVs formation control based on MIX-MAPPO

作　　者：康家境 吴杰宏[1] Kang Jiajing;Wu Jiehong(Shenyang Aerospace University,Shenyang 110136,China)

机构地区：[1]沈阳航空航天大学,沈阳110136

出　　处：《计算机应用研究》2025年第1期231-235,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(62376165)。

摘　　要：单一无人机无法有效处理复杂的多任务场景,而无人机编队解决这方面问题具有显著优势。为了满足多任务场景和面向任务的编队聚集以及运行中编队队形保持的需求,设计了一种基于门格海绵分形的无人机编队模型。同时,采用多智能体近端策略优化算法(MAPPO)、近端策略优化算法(PPO)和注意力机制来训练编队控制策略,简化无人机编队模型的建立,根据无人机对其他无人机的关注为所有输入分配权重,增强其在动态环境中的适应性。针对MAPPO算法收敛速度慢和对多个智能体适应性有限的问题,提出了一种基于门格海绵分形结构的MIX-MAPPO算法。实验结果表明,与DDPG、PPO、MADDPG和MAPPO等算法相比,该方法不但成功地实现了稳定的编队,而且具有明显更快的收敛速度和更高的奖励值,说明了MIX-MAPPO算法在编队集群控制领域的优越性。Single unmanned aerial vehicles(UAVs)struggle to effectively handle complex multi-task scenarios,whereas UAV swarms exhibit significant advantages in addressing such challenges.This paper proposed a drone swarm model based on the Menger sponge fractal to meet the needs of multi-task scenarios and maintaining formation during swarm operation.The model employed multi-agent proximal policy optimization(MAPPO),proximal policy optimization(PPO),and attention mechanism to train the formation control strategy.The approach simplified the establishment of the drone swarm model by assigning weights to all inputs based on each drone’s attention to other drones,which enhanced adaptability in dynamic environments.To address the slow convergence and limited adaptability of the MAPPO algorithm with multiple agents,the paper introduced a Menger sponge fractal-based MIX-MAPPO algorithm.Experimental results demonstrate that this method not only achieves stable formations successfully,but also has significantly faster convergence speed and higher reward values compared to algorithms such as DDPG,PPO,MADDPG,and MAPPO.The MIX-MAPPO algorithm thus proves to be superior in the field of swarm control.

关 键 词：无人机编队 门格海绵结构 注意力机制 集群控制 编队保持 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]