A 6D Pose Estimation for Robotic Bin-Picking Using Point-Pair Features with Curvature (Cur-PPF).

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Accurate 6D pose estimation is crucial for robotic bin-picking.
• Real-world bin-picking faces challenges like noise, object overlap, and occlusion, hindering grasping success.
• Existing methods, such as the point-pair feature (PPF) descriptor, can be limited in complex scenarios.

Purpose of the Study:

• To propose a novel point-pair feature (PPF) descriptor for robust 6D pose estimation in robotic bin-picking.
• To enhance the accuracy and computational efficiency of pose estimation in cluttered and occluded environments.
• To improve the success rate of robot grasping in industrial applications.

Main Methods:

• Introduced a new PPF descriptor incorporating curvature information for stronger feature description.
• Developed an effective point cloud preprocessing technique to extract candidate targets efficiently.
• Implemented a weighted voting scheme based on curvature distribution to refine pose estimation.

Main Results:

• The proposed PPF descriptor significantly improved the point cloud matching rate.
• The method demonstrated higher accuracy compared to the existing PPF method in public datasets and real scenarios.
• The enhanced approach showed improved computational efficiency over the standard PPF method.

Conclusions:

• The novel PPF descriptor with curvature information offers superior performance for robotic bin-picking.
• The proposed method effectively addresses noise, overlap, and occlusion, leading to more reliable robot grasping.
• This technique is suitable for real-world industrial robotic bin-picking applications.