利用強化學習於無人機自主巡檢之路徑規劃;Path Planning for Autonomous UAV Inspection Using Reinforcement Learning

NCUIR > School of Management at National Central University > Graduate Institute of Industrial Management > Electronic Thesis & Dissertation > Item 987654321/97345

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/97345

Title:	利用強化學習於無人機自主巡檢之路徑規劃;Path Planning for Autonomous UAV Inspection Using Reinforcement Learning
Authors:	黃彥憬;Huang, Yan-Jing
Contributors:	工業管理研究所
Keywords:	覆蓋路徑規劃;強化學習;無人機巡檢;近端策略最佳化;工業監測;Coverage Path Planning, Reinforcement Learning;Reinforcement Learning;UAV Inspection;Proximal Policy Optimization;Industrial Monitoring
Date:	2025-07-28
Issue Date:	2025-10-17 11:09:58 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著智慧製造與工業4.0的快速發展，廠房巡檢的自動化需求日益提升。本研究提出一套結合人工智慧與強化學習技術的無人機自動巡檢系統，旨在提升巡檢效率，並降低人力成本與人力巡檢時的潛在風險。系統設計以自訂的 GridWorld 環境模擬廠房空間，並透過 Proximal Policy Optimization（PPO）演算法訓練無人機學習最佳巡檢策略。為強化探索能力與實現策略泛化，本研究引入多種隨機生成的地圖場景、障礙物配置與視野限制，首次訪問獎勵與回原點的獎勵設計。此外，輔以指引資訊以提升學習效率。實驗結果顯示，訓練後的模型具備良好的環境適應能力，不僅能有效完成全區覆蓋式巡檢，亦能成功應用於未曾訓練過的新地圖上，展現出良好的泛化能力。本研究展示了無人機結合深度強化學習於智慧工廠巡檢的應用潛力，並為後續拓展至實體部署奠定基礎。;With the rapid advancement of smart manufacturing and Industry 4.0, the demand for automated inspection in factory environments has been steadily increasing. This study proposes an autonomous UAV inspection system that integrates artificial intelligence and reinforcement learning techniques, aiming to enhance inspection efficiency while reducing labor costs and the potential risks associated with manual inspections. The system is designed using a customized GridWorld environment to simulate the factory layout, and employs the Proximal Policy Optimization (PPO) algorithm to train the UAV to learn an optimal inspection strategy. To improve exploration capabilities and achieve policy generalization, the study incorporates various randomly generated map scenarios, obstacle configurations, and visibility constraints, along with reward mechanisms such as first-visit bonuses and return-to-origin incentives. Additionally, guiding information is provided to further accelerate the learning process. Experimental results demonstrate that the trained model exhibits strong adaptability to diverse environments, effectively completing full-area coverage inspections and successfully generalizing to previously unseen maps. This research highlights the potential of integrating UAVs with deep reinforcement learning for smart factory inspections, laying the groundwork for future deployment in real-world applications.
Appears in Collections:	[Graduate Institute of Industrial Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	22	View/Open

社群 sharing

Loading...