A pizza box may feel like a recycled pop quiz. It’s cardboard, but there’s still some greasy stuff and cheese crumbs on the bottom. Should I put the box in the blue recycling bin or in the trash?
A wrong decision may seem like a harmless fling, but it can have serious consequences, and engineers hope artificial intelligence can solve it.
Recycling facilities, or material recovery facilities, sort and process recyclable materials such as plastic, glass, and paper, which are then sold to manufacturers to create new products.
However, if non-recyclable items like oil-soaked pizza boxes get mixed with other valuable materials, the entire batch can be rejected and sent to a landfill. Large-scale landfills threaten the environment and human health, and the United States is one of the world’s largest per capita waste producers.
At Stony Brook University, researchers are exploring AI as part of the solution by developing an AI-assisted system to analyze and characterize municipal solid waste much faster and at scale than traditional methods.
Stony Brook’s project reflects a broader national trend, as scientists and engineers across the country increasingly place AI at the center of efforts to streamline recycling programs and create more efficient and effective waste management and separation systems.
Train AI to sort trash smarter
The Stony Brook project officially began in January 2025. As part of preliminary work, the project’s principal investigator, Associate Professor Lewen Chin, visited material recovery facilities on Long Island and spoke with staff about the challenges they face and the solutions they are interested in. “Without cooperation from local facilities, it would be impossible to conduct this type of research, because that data is essential for the development of artificial intelligence algorithms,” she said.
During these site visits, Qin and her team used low-cost cameras such as GoPros to capture video and audio. Qin said this data was used to guide the development of the AI model.
The Stony Brook AI model was then trained to identify paper, plastic, food waste, and textiles and automatically estimate their amounts. This research was supported by a Stony Brook University AI Innovation Seed Grant. After receiving the grant, Hata was able to involve graduate students in his research. Mr. Hata has also worked closely with the university’s Waste Data Analysis Center throughout this effort.
“A very important job is to sample and separate the waste to determine what materials are in the waste stream and how much,” Qin told Business Insider. “Once an algorithm is trained, it can analyze large numbers of samples more efficiently than humans.”
This process of identifying, separating, and analyzing waste stream components is known as characterization. This is detailed work that takes time. But Hata said AI could ideally speed up the process. AI models like the one she’s developing can identify if non-recyclables are accidentally mixed with other recyclable products, preventing them from being rejected and sent to landfills.
Although the project is in its early stages, Qin said the short-term goal is to provide researchers with high-quality data that they can then use to develop more affordable and accessible open-source models.
Qin added that the team will continue to train the model so that it will eventually be able to “identify different waste products under all conditions.” She also hopes to secure additional funding to apply the technology to real-world applications such as material recovery facilities.
In the future, Qin said he is interested in combining AI and robotics. The algorithms can tell the robot what can and cannot be taken from the waste stream.
Expanding technology
AI recycling algorithms are starting to trickle into the waste management industry. For example, AMP Robotics in Colorado has developed an AI robot system for factory lines. Additionally, London-based startup Greyparrot uses its AI sorting system in more than 20 countries across North America, Europe, and Asia.
Aurora del Carmen Munguía López, an assistant professor who studies recycling solutions at the University at Buffalo, said there is still work to be done when it comes to developing AI sorting systems. As pilot projects move from various universities to factory facilities, Munguia-Lopez told Business Insider that part of the challenge is determining whether these algorithms can work at the scale needed in professional settings.
Munguia López said that while AI’s energy-intensive data centers create environmental risks, its overall impact could still be positive if the technology could increase recycling rates, reduce reliance on fossil fuel-based plastic production and reduce greenhouse gas emissions.
Given the technology’s potential to improve recycling and reduce emissions, Qin wants to ensure that Stony Brook’s AI models are intelligent products that anyone can use to their advantage. “We want to publish data, models and technology to benefit society,” she said.