Robo-Sort: Real-Time Object-Based Detection for Automated Solid Waste Collection and Segregation Using YOLOv8

Efficient waste segregation remains a significant challenge in urban areas, necessitating innovative solutions to enhance sustainability and automate waste management. This study explores the integration of artificial intelligence (AI) and robotics for real-time waste classification using YOLOv8 object detection models. Specifically, it compares the performance of YOLOv8n (nano) and YOLOv8x (extralarge) in identifying waste items categorized as recyclable, biodegradable, and non-biodegradable. A dataset of 548 waste items was used to evaluate precision, recall, F1-score, and training time. Results show that YOLOv8x outperforms YOLOv8n in accuracy, achieving higher precision (0.96 vs. 0.959) and recall (0.941 vs. 0.929), effectively minimizing misclassifications. However, YOLOv8x requires significantly longer training time (0.437 vs. 0.111-time units), making it less suitable for real-time deployment on resource-constrained devices. In contrast, YOLOv8n offers faster processing with slightly lower accuracy, making it ideal for applications prioritizing speed and efficiency. The study concludes that selecting between YOLOv8n and YOLOv8x depends on the required balance between accuracy and computational efficiency.