Scholarlink

<p>Efficient autonomous exploration strategies in unstructured environments are essential core capabilities for embodied intelligent robots to perform independent operations. To address the challenges faced by conventional search-driven inspection robots—such as low target search efficiency and prolonged ineffective paths—in the occluded environment of a hydroelectric generator’s interior, this paper proposes an autonomous exploration strategy that integrates Large Language Models (LLMs) with Next Best View (NBV) planning. The proposed strategy leverages LLMs to parse semantic information from the environment, generate high-potential interest regions, and dynamically adjust task relevance factors based on these regions. Additionally, it incorporates a target-biased sampling mechanism to optimize NBV planning and improve exploration efficiency. A realistic experimental environment was constructed based on the actual interior structure of a hydroelectric generator, and autonomous inspection trials were conducted. Experimental results demonstrate that, under the same inspection tasks, the proposed method reduces the inspection path length by 49.8% compared to traditional strategies, achieving rapid navigation to target components under occluded conditions. These results validate the effectiveness of the proposed strategy in autonomous exploration tasks and offer a novel approach to enhancing robotic autonomous navigation capabilities in industrial inspection scenarios.</p> Graphical abstract <p></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

Research on an autonomous exploration strategy for hydroelectric generator inspection robots driven by large language models

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Yuanfa Dong,
  • Yongfei Ji,
  • Haoyang Geng,
  • Jin Yin,
  • Bin Zhou,
  • Wei Peng,
  • Youjun An

摘要

Efficient autonomous exploration strategies in unstructured environments are essential core capabilities for embodied intelligent robots to perform independent operations. To address the challenges faced by conventional search-driven inspection robots—such as low target search efficiency and prolonged ineffective paths—in the occluded environment of a hydroelectric generator’s interior, this paper proposes an autonomous exploration strategy that integrates Large Language Models (LLMs) with Next Best View (NBV) planning. The proposed strategy leverages LLMs to parse semantic information from the environment, generate high-potential interest regions, and dynamically adjust task relevance factors based on these regions. Additionally, it incorporates a target-biased sampling mechanism to optimize NBV planning and improve exploration efficiency. A realistic experimental environment was constructed based on the actual interior structure of a hydroelectric generator, and autonomous inspection trials were conducted. Experimental results demonstrate that, under the same inspection tasks, the proposed method reduces the inspection path length by 49.8% compared to traditional strategies, achieving rapid navigation to target components under occluded conditions. These results validate the effectiveness of the proposed strategy in autonomous exploration tasks and offer a novel approach to enhancing robotic autonomous navigation capabilities in industrial inspection scenarios.

Graphical abstract