(MARGOT) Monocular Camera-Based Robot Grasping Strategy for Metallic Objects.

Area of Science:

• Robotics
• Computer Vision
• Machine Learning

Background:

• Robotic object handling, especially in teleoperation, can be complex and stressful for operators.
• Existing methods struggle with complex shapes, reflections, and shadows in unstructured environments like nuclear facilities.

Purpose of the Study:

• To develop a novel grasping strategy for robotic object manipulation.
• To enhance precision and reduce operator workload in teleoperated tasks.
• To address challenges posed by complex geometries and difficult lighting conditions.

Main Methods:

• A groundbreaking geometrical analysis to extract diametrically opposite points for uniform grasping.
• Utilizing a monocular camera for target recognition, isolation, and spatial coordinate estimation.
• Employing machine learning and computer vision techniques, including a specialized dataset for improved object detection.

Main Results:

• The algorithm successfully identifies and isolates targets, estimating stable grasping points for various objects.
• It effectively handles reflections and shadows, crucial for unstructured industrial settings.
• Experimental results show high accuracy with error rates in the millimeters range.

Conclusions:

• The novel grasping strategy significantly improves robotic object handling capabilities.
• The approach is robust in challenging, low-contrast environments, demonstrating high repeatability and accuracy.
• This method offers a viable solution for reducing workload and enhancing safety in critical applications.