A grasp point generation algorithm for waste handling based on a generative reasoning network.

Area of Science:

• Robotics and Automation
• Computer Vision
• Environmental Science

Background:

• Robotic waste sorting faces challenges with grasping diverse and dirty objects like bottles.
• Current methods struggle with accuracy and efficiency in complex urban kitchen waste scenarios.

Purpose of the Study:

• To develop an advanced garbage grasping framework for improved robotic waste sorting.
• To enhance the accuracy and speed of identifying and grasping recyclable bottles.

Main Methods:

• Proposed a garbage grasping framework utilizing the Channel Exchange Generative Residual Inference Network (CE-GR-NET).
• Integrated hierarchical visual features from RGB and depth images for optimal grasping trajectory synthesis.
• Employed an object detection network for identifying recyclable bottles within solid waste.

Main Results:

• Achieved good inference performance and fast inference speed on the Cornell Grasping Dataset for single-object scenarios.
• On self-constructed multi-source urban kitchen waste images, achieved 96.03% image-based model accuracy and 94.40% object-based model accuracy.
• Demonstrated a grasping object classification accuracy of 97.87%.

Conclusions:

• The CE-GR-NET framework effectively generates grasping boxes and classifies recyclable bottles in complex waste streams.
• The proposed method offers a significant improvement for robotic systems in urban kitchen waste sorting applications.
• The framework shows high accuracy and efficiency in both controlled datasets and real-world waste scenarios.