Point Pair Feature-Based Pose Estimation with Multiple Edge Appearance Models (PPF-MEAM) for Robotic Bin Picking.

Area of Science:

• Robotics
• Computer Vision
• Industrial Automation

Background:

• Robotic bin picking requires accurate object pose estimation for successful manipulation.
• Existing methods struggle with industrial parts lacking detailed 3D point cloud data.

Purpose of the Study:

• To develop a novel pose estimation descriptor, Boundary-to-Boundary-using-Tangent-Line (B2B-TL), for robust robotic bin picking.
• To improve recognition rates and reduce computation time using Multiple Edge Appearance Models (MEAM).

Main Methods:

• Utilizing a novel point pair feature descriptor (B2B-TL) that combines 3D point cloud and 2D image data.
• Implementing Multiple Edge Appearance Models (MEAM) for enhanced object recognition.
• Developing an online computation pipeline for real-time application.

Main Results:

• The B2B-TL descriptor effectively compensates for missing details in 3D point clouds using 2D image data.
• The integrated B2B-TL and MEAM approach demonstrated high accuracy in pose estimation for industrial parts.
• The system achieved sufficient speed for real-time application in factory environments.

Conclusions:

• The proposed B2B-TL descriptor and MEAM method provide an accurate and efficient solution for robotic bin picking.
• This approach enables reliable grasping of industrial parts, even those with incomplete 3D data.
• The developed online pipeline is suitable for practical implementation in industrial settings.