Autonomous Capture of Submerged Plastic Bags

Abstract

Plastic bags are a major contributor to oceanic macroplastic pollution, presenting serious threats to marine life and ecosystems. As these bags break down into microplastics, the hazards worsen, affecting both marine organisms and human health. This thesis introduces a complete pipeline designed to capture drifting plastic bags using an underwater manipulator. This pipeline could be used for future deployment on autonomous underwater vehicles (AUVs). The system integrates object detection, tracking, and intercept planning to capture drifting underwater objects. Chapter 1 introduces FindingTrash, a vision pipeline combining YOLO and DeepSORT, trained on a curated dataset to detect and track plastic bags. This system is paired with a robotic manipulator to enable reactive grasping. Chapter 2 builds on this by integrating a reachability-based planning strategy that compares predicted drift times with precomputed arm reach times, allowing the robot to select timely interception points. Bag drift was simplified to one degree of freedom to reduce complexity. A simulation of our system tracked and intercepted plastic bags through our test tank, offering a promising solution for automated underwater plastic bag removal.