SHIFT-T: A Low-Cost Autonomous Mobile Robot for Trash Collection and Sorting for Effective Waste Management

Abstract

This paper presents SHIFT-T, a mobile robot designed to efficiently identify and categorize various types of waste for proper disposal. The robot classifies waste into three primary types: metal, plastic, and polythene. The system features an innovative trash classification and localization system that integrates within a mobile manipulator framework, enabling it to accurately grasp and sort items into appropriate recycling bins. This includes deep learning techniques, specifically employing YOLOv5 image segmentation neural network fine-tuned with a custom dataset for trash detection. An ultrasonic sonar sensor detects objects along the robot's path, while a camera module captures images and transmits them to a Raspberry Pi for classification and coordinate determination. This process allows for precise identification of an optimal grasp pose for retrieving trash from the ground. Finally, the identified items are placed at predetermined locations according to their waste categories as initially classified by the neural network. The model has achieved an accuracy rate of approximately 81 0/0, demonstrating its effectiveness in identifying and classifying various waste items. The low-cost prototype demonstrates high accuracy in detecting various categories of trash and utilizes 6 degrees of freedom arm for manipulation.