Stony Brook researchers launch AI-powered recycling project to reduce waste contamination and improve sustainability.
Stony Brook, NY, July 6, 2025 — A new research initiative funded by Stony Brook University’s AI Innovation Seed Grant is reimagining how we tackle one of the most persistent problems in recycling: contaminated waste streams. By combining video footage and cutting-edge artificial intelligence, researchers aim to automate the analysis of recycling materials, reduce contamination, and lay the groundwork for smarter, cost-effective, and sustainable waste management.
Ruwen Qin
According to the Environmental Protection Agency’s estimates, approximately 75% of the waste produced in the United States is recyclable; however, the actual recycling rate is only 35%, resulting in an estimated 68 million tons of recyclables being sent to landfills or incinerators. Meanwhile, the rate of recycling contamination (when non-recyclable items are mistakenly mixed in) is 25%, resulting in millions of tons of recyclables being rejected annually, and sent to landfills.
Material recovery facilities (MRFs) automate the sorting of recyclable materials by type, yet human intervention remains crucial for contaminant removal. This manual process, however, is time-consuming, labor-intensive, and poses significant safety risks. While robotic sorting technologies are emerging, their widespread adoption has not yet reached a scale that can fully replace human workers. Providing the research community with annotated, industry-grade big data will accelerate the application of AI within the solid waste management sector.
To address this challenge, a team of researchers at Stony Brook is building an AI-assisted design to support sustainable recycling. The project aims to use sensors and develop machine learning algorithms capable of identifying, tracking, and counting waste materials as they move through real recycling facilities.
Ruwen Qin, Associate Professor, Department of Civil Engineering, and the principal investigator of the project, said, “We're not just building tools in isolation, we're collecting data at multiple stages of the sorting process, engaging with recycling workers to understand the pain points, and using those insights to help them work faster, safer, and with greater insight.” Collaborating with municipalities and the Waste Data and Analysis Center housed within the Department of Technology, AI, and Society and funded by New York State Department of Environmental Conservation (NYSDEC), the team is collecting high-resolution video data from multiple stages of the sorting process at local MRFs in Long Island. Unlike previous datasets, which are often proprietary, lab-generated, or limited in scope, this collection captures the full complexity of recycling streams in motion. It will be one of the first public datasets of its kind.
Backed by the technical resources of AI^3, the team is implementing state-of-the-art computer vision tools such as object detection (YOLO), segmentation (SAM), and object tracking (DeepSort) to recognize and quantify recyclable materials. They are also identifying the limits of current models, with the goal of finding new methods to better handle challenges like cluttered objects, transparent materials, and fine-grained classification.
Left: Sample video frames. Right: Challenges in using existing AI models.
“This project has given us a rare opportunity to advance both AI research and environmental sustainability at the same time,” said Vismay Vora, a Graduate AI Consultant on the project. “Recycling is a surprisingly complex visual problem. Solving it means pushing the boundaries of what AI can do.” The long-term goal is to create a cost-effective, scalable AI tool that can be deployed across various recycling systems. The project also seeks to deepen collaborations between AI researchers, recycling professionals, and policy stakeholders, including NYSDEC and local municipalities.
For months, the team has been collecting industry-grade video data from multiple stages of the recycling process, working closely with professionals to capture the complexities of sorting through waste streams. Building on this dataset, they are currently focused on refining existing AI models to accurately identify and track recyclable materials, to minimize the need for manual intervention. Their research findings will be shared on GitHub and through conferences and publications.
By building both the technical foundation and the collaborative ecosystem needed to support AI-enabled recycling, the project positions Stony Brook’s AI Innovation Institute as a leader in AI-driven solutions for sustainability.
“We’re not just studying the problem. We’re building tools that can make a measurable difference,” said Qin. “And we’re doing it in partnership with the people and organizations that make recycling work every day.” The team’s work is paving the way for more efficient and accurate recycling processes, serving as a blueprint for how academic institutions can partner with industry and government to tackle pressing environmental challenges using artificial intelligence.