Robotic Recyclables Segregation System using TinyML

Abstract

The increasing volume of waste generated globally is a significant concern. Recycling plants worldwide attempt to reduce this waste, but they struggle to keep up. Technologically underdeveloped countries face further challenges, as they have not yet found cost-effective methods for recycling. Manual waste segregation worsens the issue, posing health risks to workers due to exposure to hazardous materials. To address these concerns, this paper introduces a robotic system for the segregation of recyclables, aimed at improving waste management efficiency. The review of existing literature highlights the prevalent use of supervised learning approaches, particularly CNN, SVM, Faster R-CNN, YOLO-V3, and RESNET architectures. The term “Tiny Machine Learning” (TinyML) describes the creation and use of machine learning models on devices with limited resources, allowing for effective inference with little memory and processing power consumption. The objective is to develop a lightweight system that identifies and segregates recyclables, reducing manual labour and improving the accuracy of material recovery facilities using TinyML. The proposed system utilises economical hardware such as ESP-32-Cam and Arduino Mega microcontrollers, 3D-printed components, and a CNN-based object detection model while maintaining similar accuracy and time efficiency. The system demonstrates high accuracy for single waste class type detection, achieving an F1 score of 96.3 % and low inferencing time. Merging machine learning, robotics, and sustainable practices heralds a cleaner, technology-driven future in recycling management.