Autonomous detection and sorting of litter using deep learning and soft robotic grippers

Area of Science:

• Robotics
• Environmental Science
• Computer Vision

Background:

• Roadside litter poses safety and environmental risks.
• Manual litter collection is labor-intensive, costly, and hazardous.
• Automation is needed to improve efficiency and safety.

Purpose of the Study:

• To develop an autonomous robotic system for roadside litter detection, localization, classification, and collection.
• To create a robust manipulation framework for handling diverse litter items.
• To validate the system's performance in various real-world scenarios.

Main Methods:

• Utilized a learning-based object detection and segmentation algorithm trained on the TACO dataset.
• Developed a modular manipulation framework with soft robotic grippers.
• Implemented a real-time visual-servoing strategy for dynamic object manipulation.

Main Results:

• Achieved over 80% success rate in classified picking and binning.
• Validated performance on static, cluttered, and dynamic litter configurations.
• Demonstrated accurate deployment of deep learning models in real-world applications.

Conclusions:

• The LitterBot system effectively automates roadside litter collection.
• The robotic system offers a cost-effective and safer alternative to manual methods.
• Deep learning models can be successfully integrated into robotic systems for environmental applications.