强化学习
计算机科学
图形
人工神经网络
趋同(经济学)
一般化
过程(计算)
控制(管理)
人工智能
等边三角形
国家(计算机科学)
分布式计算
避障
深层神经网络
控制工程
分散系统
机器人
控制系统
遥控水下航行器
作者
Siqi Sun,Hualong Yu,Boyin Jin,Qi Wang
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-09-26
卷期号:: 252-255
被引量:1
标识
DOI:10.1109/aipmv67185.2025.11290043
摘要
Unmanned Aerial Vehicle (UAV) formation control is a critical challenge in cooperative multi-agent systems. Traditional methods often struggle with complex interactions among agents. This paper presents GAT-MAPPO, a novel framework that integrates Graph Neural Networks (GNNs) with Multi-Agent Proximal Policy Optimization (MAPPO) to address the formation control problem. In GAT-MAPPO, each UAV communicates with others via a graph structure, where GNNs process local observations and model interactions to generate coordinated actions. The centralized critic evaluates the global state to guide policy updates. We design a formation environment where three UAVs must form and maintain an equilateral triangle formation while following a moving center point. Experiment results demonstrate that the convergence speed of GAT-MAPPO is about 20% faster than that of MAPPO.The framework also exhibits strong generalization capabilities in varying scenarios. This work contributes to the application of graph-based multi-agent reinforcement learning in UAV formation control.
科研通智能强力驱动
Strongly Powered by AbleSci AI