SBIR Phase I: Autonomous waste sorting platform for decentralized pre-processing

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this Small Busisness Innovation Research (SBIR) Phase I project is in improving landfill diversion and recycling rates of non-industrial waste. This project targets high-traffic venues (HTVs) with controlled waste streams such as stadiums, universities, airports, and theme parks. Waste stream sorting to differentiate classes of materials for recycling and landfill is currently undertaken either using rudimentary, labor-intensive manual sorting, or expensive and technically complex robotic sorting, neither of which are feasible options for small- to mid-sized facilities (i.e., HTVs), which then shoulder heightened rates imposed by material recovery facilities due to contamination. These expenses have discouraged recycling, contributing to the mounting waste problem. This project seeks to develop an intelligent waste-sorting system that leverages computer vision and innovative hardware to enable on-site, decentralized sorting, facilitating the recapture of the 66 million tons and $200 billion worth of recyclable materials that currently go to waste each year. By mitigating waste accumulation in landfills and the greenhouse gas emissions associated with virgin resource mining, this technology supports United Nations sustainable development goals 11 (sustainable cities and communities) and 12 (responsible consumption and production) and aligns with the NSF’s mission of advancing national health, prosperity, and welfare. The proposed technology consists of a hardware-software solution that uses the latest in computer vision to perform automated singulation (i.e., arranging objects in a 1-by-1 stream) and classification in cluttered environments, allowing for increasingly complex (or diverse) structures (shapes, sizes, and materials) to be accurately identified and subsequently sorted. This approach produces an ordered stream of objects, which can then be sorted according to any diversion scheme for efficient recycling. The software uses semi-supervised learning to allow for domain adaptation from a centralized training set, enabling rapid implementation of optimized sorting schemes of site-specific waste streams, requiring significantly less human intervention than traditionally needed. Successful development would result in a simplified sorting platform that is cheaper, more robust, and less resource intensive than existing waste sorting operations, thus offering a novel turnkey solution that could be feasibly adopted on-site. Research objectives include: 1) Developing and evaluating hardware assemblies using electromechanical processes for waste stream singulation; 2) Developing software for efficient dataset generation and waste stream classification, particularly using semi-supervised learning and data augmentation approaches; and 3) Validating the developed system for high precision singulation and classification of recyclables under noisy like conditions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.