A Survey on UAV Control with Multi-Agent Reinforcement Learning

摘要

Unmanned Aerial Vehicles (UAVs) have become increasingly prevalent in both governmental and civilian applications, offering significant reductions in operational costs by minimizing human involvement. There is a growing demand for autonomous, scalable, and intelligent coordination strategies in complex aerial missions involving multiple Unmanned Aerial Vehicles (UAVs). Traditional control techniques often fall short in dynamic, uncertain, or large-scale environments where decentralized decision-making and inter-agent cooperation are crucial. A potentially effective technique used for UAV fleet operation is Multi-Agent Reinforcement Learning (MARL). MARL offers a powerful framework for addressing these challenges by enabling UAVs to learn optimal behaviors through interaction with the environment and each other. Despite significant progress, the field remains fragmented, with a wide variety of algorithms, architectures, and evaluation metrics spread across domains. This survey aims to systematically review and categorize state-of-the-art MARL approaches applied to UAV control, identify prevailing trends and research gaps, and provide a structured foundation for future advancements in cooperative aerial robotics. The advantages and limitations of these techniques are discussed along with suggestions for further research to improve the effectiveness of MARL application to UAV fleet management.