Scholarlink

<p>Effective human–robot collaboration (HRC) requires real-time understanding of human pose, shape, and location within shared workspaces. This informs robot control strategies, enabling safe and efficient trajectory planning for collision prediction and avoidance. Existing methods for human motion tracking in collaborative environments often lack accurate 3D localization, human surface mesh reconstruction, or the practicality needed for real-world deployment. This paper introduces a method to digitally reconstruct an HRC work cell in 3D, enabling real-time monitoring of human motion, robot movement, and key spatial metrics, such as critical distances. Our proposed approach, TRIPL3, is a fully neural network-based approach that operates on input from a single RGB-D camera and achieves real-time deployment due to its low computational requirements. Extensive experiments and benchmarking validate our method’s capability for real-time 3D safety monitoring of human–robot interaction (HRI). Additionally, our approach outperforms state-of-the-art (SoTA) methods in critical distance monitoring.</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 vision-guided approach to 3D spatial safety monitoring in human–robot systems

  • 期刊论文

  • 发表日期：

    2026年3月28日

查看全文
  • Tarek Yahia,
  • Kody Haubeil,
  • Janet Dong,
  • Xiaodong Jia

摘要

Effective human–robot collaboration (HRC) requires real-time understanding of human pose, shape, and location within shared workspaces. This informs robot control strategies, enabling safe and efficient trajectory planning for collision prediction and avoidance. Existing methods for human motion tracking in collaborative environments often lack accurate 3D localization, human surface mesh reconstruction, or the practicality needed for real-world deployment. This paper introduces a method to digitally reconstruct an HRC work cell in 3D, enabling real-time monitoring of human motion, robot movement, and key spatial metrics, such as critical distances. Our proposed approach, TRIPL3, is a fully neural network-based approach that operates on input from a single RGB-D camera and achieves real-time deployment due to its low computational requirements. Extensive experiments and benchmarking validate our method’s capability for real-time 3D safety monitoring of human–robot interaction (HRI). Additionally, our approach outperforms state-of-the-art (SoTA) methods in critical distance monitoring.