Scholarlink

The automation of wire harness handling and installation in the automotive industry presents a challenge due to the inherent flexibility of cables, the high variance in wire harnesses and plug combinations, and the intricate spatial configurations required for accurate installation. Addressing this challenge requires the integration of sensors for accurate pose estimation with high-dexterity robotic systems. This work introduces a novel approach to automate the process of grasping of wiring harnesses for autonomous installation using a robotic arm. The methodology encompasses several stages. Initially, a multi-stereo camera setup creates a high-accuracy representation of the working area. Next, a deep learning model predicts a spline representing the segment with the biggest connector attached to it for 6D grasp pose estimation. The final stage uses a skill-based robot program to perform the grasping of the wiring harness, which is evaluated using 50 random configurations inside a bin. As a result, the proposed solution achieves an accuracy of 82% of successful wiring harness bin picking grasps, where success is defined when the result is that a specific connector on the wiring harness is in a predefined spot after grasping. Future work will use another robot to grasp the connector from the first robot to install it in an automotive demo door using reinforcement learning.

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

Robotic Wiring Harness Bin Picking Solution Using a Deep-Learning-Based Spline Prediction and a Multi-stereo Camera Setup

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年

查看全文
  • Manuel Zürn,
  • Carsten Schmerbeck,
  • Andreas Kernbach,
  • Mara I. Kläb,
  • Alper Yaman,
  • Daniel Bragmann,
  • Michael Heizmann,
  • Marco Huber,
  • Werner Kraus,
  • Armin Lechler,
  • Alexander Verl

摘要

The automation of wire harness handling and installation in the automotive industry presents a challenge due to the inherent flexibility of cables, the high variance in wire harnesses and plug combinations, and the intricate spatial configurations required for accurate installation. Addressing this challenge requires the integration of sensors for accurate pose estimation with high-dexterity robotic systems. This work introduces a novel approach to automate the process of grasping of wiring harnesses for autonomous installation using a robotic arm. The methodology encompasses several stages. Initially, a multi-stereo camera setup creates a high-accuracy representation of the working area. Next, a deep learning model predicts a spline representing the segment with the biggest connector attached to it for 6D grasp pose estimation. The final stage uses a skill-based robot program to perform the grasping of the wiring harness, which is evaluated using 50 random configurations inside a bin. As a result, the proposed solution achieves an accuracy of 82% of successful wiring harness bin picking grasps, where success is defined when the result is that a specific connector on the wiring harness is in a predefined spot after grasping. Future work will use another robot to grasp the connector from the first robot to install it in an automotive demo door using reinforcement learning.