This repository provides experimental videos for our waste sorting robot.
Accurate and efficient sorting of diverse mixed solid waste is essential for maximizing recycling rates and reducing environmental impact. We present a robotic waste sorter designed for rapid identification and sorting, employing an affordable CCD array camera and a cost-effective air jet device. After detecting waste items using deep learning models and tracking their positions, precise control of the air jet nozzles ensures the accurate ejection of waste items into assigned bins, considering both their positions and sizes. Physical tests demonstrate our robot's capability to function at a high belt speed of 4 m/s, with an air ejection accuracy of 100%. The waste detection model, trained on our self-collected dataset, excels in identifying 14 categories of solid waste, achieving an F1 score over 95% at a 0.7 confidence threshold in the validation set. The robot achieves an overall sorting success rate of 93.9% across five sorting tasks. Adding a material shaking device and an object state stabilization device at the sorting machine's front end can boost the sorting success rate to 98.6%. These findings highlight the efficiency and reliability of our robotic waste sorting system. Experimental videos are available at https://github.com/JiatongBao/FastVisualSorting.
In the MRF (as shown in the following linked video), our robotic sorter was used to eject plastic bottles with white, green and blue colors. https://www.youtube.com/shorts/z3-s1pcRQ-w
If you have any questions, please let me know: [Jiatong Bao] jtbao[at]yzu[dot]edu[dot]cn