Scholarlink

This paper presents a real-time system for object tracking and pose estimation of an industrial part—a small oil pan—using an xArm robotic manipulator. The system integrates an RGB-D sensor (Azure Kinect) within the Robot Operating System 2 (ROS2) Humble framework to acquire synchronized color and depth data. A You Only Look Once (YOLOv8) model performs semantic segmentation, enabling point cloud extraction and filtering. Coarse pose estimation is achieved using RANSAC, followed by refinement through Wunsch registration. The resulting transformation informs the robot’s inverse kinematics to align its end-effector with the target. Experimental results in controlled settings validate the system’s robustness and its potential for industrial automation.

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

Tracking a Small Oil Pan with xArm Using RGB-D Data and Deep Pose Estimation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rubén de Jesús Hilario Cruz,
  • Juan José Jaúregui Barba,
  • Gustavo De los Ríos Alatorre,
  • Luis Alberto Muñoz Ubando

摘要

This paper presents a real-time system for object tracking and pose estimation of an industrial part—a small oil pan—using an xArm robotic manipulator. The system integrates an RGB-D sensor (Azure Kinect) within the Robot Operating System 2 (ROS2) Humble framework to acquire synchronized color and depth data. A You Only Look Once (YOLOv8) model performs semantic segmentation, enabling point cloud extraction and filtering. Coarse pose estimation is achieved using RANSAC, followed by refinement through Wunsch registration. The resulting transformation informs the robot’s inverse kinematics to align its end-effector with the target. Experimental results in controlled settings validate the system’s robustness and its potential for industrial automation.