Advanced Plastic Waste Sorting System Using Robotics and YOLO Models

The demand for plastic becomes greater as the human population increases. Consequently, the over-accumulation of plastic waste has become a significant concern, leading to a major contamination issue. Waste sorting centres in Malaysia still rely mainly on humans, who have low sorting efficiency. Therefore, deep learning and robotics should be integrated into plastic waste sorting systems to reduce the accumulation of plastic waste. This paper presents a plastic waste sorting system utilizing the YOLO object detection algorithm. The study evaluates the performance of three different YOLO variants, YOLOv10m, YOLO11m, and YOLO12m, in detecting three types of plastic waste: PP, PET, and HDPE. A delta robot with a vacuum gripper was also integrated to sort waste efficiently from the conveyor. The YOLO models were evaluated using precision and computational speeds with test data. Although the performance differences between the models are insignificant, the best model is YOLO11m, with a precision score of 0.917 and a computational speed of 34.9 ms. YOLO11m also performs best in Precision-Confidence, Precision-Recall, and F1-Confidence curve analyses. This paper shows the importance of enhancing sorting capability, reducing human labour, and promoting a more sustainable waste management system through robotic and deep learning integration.