Scholarlink

<p>This paper addresses the complex path-planning problem for automated guided vehicles (AGVs) in semiconductor fabrication facilities, which are characterized by dense layouts, narrow corridors, and dynamic obstacles. To address these challenges, we propose the hierarchical state feature-driven deep reinforcement learning (HSF-DRL) framework, which leverages the options framework in hierarchical reinforcement learning (HRL) to decompose navigation into a two-tier decision-making process. Specifically, a rule-based high-level meta-controller selects temporally extended options (e.g., global navigation, dynamic avoidance, and precise docking) by integrating online heuristic search with context-specific features; meanwhile, a low-level executor, implemented with a Deep Q-Network (DQN), generates primitive actions. A key contribution is the option-conditioned dynamic feature-fusion mechanism, where the weights of environmental, procedural, and heuristic features are conditioned on the active high-level option. This enables context-aware perception and mitigates the fixed-input limitations of conventional DRL. Evaluations conducted exclusively within a 2D grid-based semiconductor-fab simulation demonstrate that HSF-DRL outperforms traditional DQN and Dyna-Q baselines in path optimality, convergence speed, and stability under highly dynamic scenarios. Overall, this work provides a structured architectural approach for AGV navigation, establishing a foundation for future physical deployments in complex industrial settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hierarchical State Feature-Driven Deep Reinforcement Learning Framework for Semiconductor Fabrication AGV Path Planning

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • XinLin Yang,
  • BoYang Zhang,
  • YanTing Ni

摘要

This paper addresses the complex path-planning problem for automated guided vehicles (AGVs) in semiconductor fabrication facilities, which are characterized by dense layouts, narrow corridors, and dynamic obstacles. To address these challenges, we propose the hierarchical state feature-driven deep reinforcement learning (HSF-DRL) framework, which leverages the options framework in hierarchical reinforcement learning (HRL) to decompose navigation into a two-tier decision-making process. Specifically, a rule-based high-level meta-controller selects temporally extended options (e.g., global navigation, dynamic avoidance, and precise docking) by integrating online heuristic search with context-specific features; meanwhile, a low-level executor, implemented with a Deep Q-Network (DQN), generates primitive actions. A key contribution is the option-conditioned dynamic feature-fusion mechanism, where the weights of environmental, procedural, and heuristic features are conditioned on the active high-level option. This enables context-aware perception and mitigates the fixed-input limitations of conventional DRL. Evaluations conducted exclusively within a 2D grid-based semiconductor-fab simulation demonstrate that HSF-DRL outperforms traditional DQN and Dyna-Q baselines in path optimality, convergence speed, and stability under highly dynamic scenarios. Overall, this work provides a structured architectural approach for AGV navigation, establishing a foundation for future physical deployments in complex industrial settings.