Effective waste detection and management are crucial for promoting sustainable growth in metropolitan areas. This project proposes an AI-based system that combines deep learning techniques with optimized navigation algorithms to detect and classify waste in real time while planning autonomous collection routes. The system employs YOLO (You Only Look Once) object detection models, namely YOLOv5 and YOLOv8, to detect and classify waste into recyclable, organic, and hazardous categories from real-time video streams. The detected objects are positioned within the frame and categorised into established classifications. A customised Grey Wolf Optimiser (GWO) algorithm determines the ideal navigation route from the robot’s present location to the garbage site to enhance collecting efficiency. The technology superimposes this route with navigational cues and distance indicators, directing autonomous robots for effective waste collection. It can perform continuous live video analysis and is compatible with automated garbage collection systems. Experimental findings indicate robust performance, with YOLOv5 attaining a detection accuracy of 92.6% and YOLOv8 achieving 94.3%. The incorporation of GWO enhances efficient and adaptive path planning in real-time situations. The suggested approach demonstrates considerable potential for use in smart city settings by enhancing waste detection precision, decreasing response time, and facilitating intelligent automation. This approach emphasises the potential of integrating computer vision and bio-inspired optimisation for advanced urban trash management solutions.