基于强化学习的三维无人机路径规划综述

A Review of 3D UAV Path Planning Based on Reinforcement Learning

摘要: 传统无人机路径规划算法依赖精确环境模型，在复杂动态三维环境中存在适应性差、实时性差等明显局限性。文章首先阐述了传统路径规划算法的不足，引入深度强化学习作为解决该问题的全新技术路径。其次全面梳理值函数、策略梯度及值–策略混合三大类典型强化学习算法，深入探讨各类算法的核心原理，从单机与多机两个维度，系统总结了强化学习在三维无人机路径规划中的改进成果与应用进展。最后，聚焦无人机实际飞行场景的独特挑战，明确强化学习的应用瓶颈并总结未来发展方向，为该领域的理论研究与工程实践提供系统性参考。

Abstract: Traditional unmanned aerial vehicle (UAV) path planning algorithms rely heavily on accurate environmental models and exhibit significant limitations, including poor adaptability and inadequate real-time performance in complex dynamic three-dimensional (3D) environments. This thesis first elaborates on the inherent limitations of traditional path planning algorithms and introduces deep reinforcement learning as an innovative approach to address these challenges. Subsequently, it comprehensively reviews three major categories of reinforcement learning algorithms—namely value-based, policy gradient-based, and value-policy hybrid methods, delving into their core principles and systematically synthesizing their advancements and application outcomes in 3D UAV path planning from both single-UAV and multi-UAV perspectives. Finally, focusing on the unique challenges of actual UAV flight scenarios, the thesis clarifies the application bottlenecks of reinforcement learning and summarizes the future development directions, providing systematic references for theoretical research and engineering practice in this field.