Scholarlink

<p>This study proposes a method for adjusting LiDAR particle weights to enhance the localization stability of AMCL based on a multi-sensor model. Particle filter-based AMCL probabilistically estimates the robot’s position using weights derived from sensor models, and applying a multi-sensor framework allows effective fusion of multiple sensors. In particular, previous studies combining GNSS and LiDAR sensor models demonstrated that they can achieve accurate localization in outdoor environments and improve performance compared to conventional AMCL. However, when LiDAR sensor weight dominates GNSS, the algorithm behaves like conventional AMCL, which may lead to kidnapping and reduce localization stability. To address this, this study proposes a filtering method for LiDAR particle weights. In this process, particles in physically implausible positions are down-weighted, increasing the likelihood of selecting particles closer to the true position. This approach effectively mitigates kidnapping and enhances localization robustness.</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

Improvement of localization robustness through LiDAR weight adjustment in a multi-sensor particle filter

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Gyeongrok Jang,
  • Changwon Kim

摘要

This study proposes a method for adjusting LiDAR particle weights to enhance the localization stability of AMCL based on a multi-sensor model. Particle filter-based AMCL probabilistically estimates the robot’s position using weights derived from sensor models, and applying a multi-sensor framework allows effective fusion of multiple sensors. In particular, previous studies combining GNSS and LiDAR sensor models demonstrated that they can achieve accurate localization in outdoor environments and improve performance compared to conventional AMCL. However, when LiDAR sensor weight dominates GNSS, the algorithm behaves like conventional AMCL, which may lead to kidnapping and reduce localization stability. To address this, this study proposes a filtering method for LiDAR particle weights. In this process, particles in physically implausible positions are down-weighted, increasing the likelihood of selecting particles closer to the true position. This approach effectively mitigates kidnapping and enhances localization robustness.