Application of Reinforcement Learning to UAV Decision Making and Path Planning

Abstract

The aim of this study is to systematically explore the framework and potential of reinforcement learning for UAV decision-making and path planning. The article firstly analyses the core advantages of reinforcement learning over traditional planning methods. its ability to cope with environmental uncertainty, achieve multi-objective optimisation, and complete end-to-end decision-making. Then, by focusing on three typical application scenarios, namely autonomous navigation and obstacle avoidance, multi-UAV collaboration, and resource-constrained planning, the paper analyses the technical path and effectiveness of reinforcement learning in solving key problems in specific domains. Despite the challenges of sample efficiency, security, and interpretability, cutting-edge developments such as simulation-to-reality migration and constrained reinforcement learning are continuing to drive the technology towards practicality. The systematic discussion in this study aims to provide theoretical references for cross-disciplinary research on reinforcement learning and robotics, and to lay an important technical foundation for building the next-generation intelligent, robust and adaptive unmanned aerial systems.