6D Object Localization in Car-Assembly Industrial Environment | JoVE Visualize

Area of Science:

Robotics
Computer Vision
Artificial Intelligence

Background:

Object pose estimation is critical for robotic grasping in industrial settings, particularly for human-robot collaboration.
Objects with challenging characteristics (e.g., weak texture, symmetry) and cluttered, poorly lit environments pose significant difficulties for current methods.
Mobile robotic platforms utilizing Robot Operating System (ROS) are increasingly deployed in manufacturing for complex tasks like car door assembly.

Purpose of the Study:

To develop and evaluate a visual object detection and localization workflow for accurate 6D pose estimation of challenging objects.
To integrate this workflow into a mobile robotic platform for real-world industrial applications, specifically car door assembly.
To assess the performance of a learning-based method trained on diverse datasets under realistic industrial conditions.

Main Methods:

A visual object detection and localization workflow was developed and integrated into a mobile robotic platform using ROS.
A learning-based method was trained to extract object pose from single image frames.
Two datasets were collected and annotated: one in a lab and one in an industrial environment. Models were trained individually and combined, then tested in the industrial setting.

Main Results:

The developed workflow demonstrated potential for 6D pose estimation of challenging objects in industrial environments.
Models trained on combined lab and industrial datasets showed promising performance in test sequences from the actual manufacturing setting.
Qualitative and quantitative results confirmed the method's effectiveness for relevant industrial applications.

Conclusions:

The presented robotic workflow offers a viable solution for 6D object pose estimation in challenging industrial scenarios.
The approach shows promise for enhancing human-robot collaboration in manufacturing tasks like car door assembly.
Further development and deployment of this system can improve automation and efficiency in industrial robotics.