Stony Brook Applies AI to Improve Recycling Sorting

Researchers at Stony Brook University have built an experimental, AI-assisted system to analyze municipal solid waste and help identify contamination in recycling streams. According to Stony Brook University press materials and Business Insider reporting, the project combines high-resolution video capture, wearable cameras used by sorting crews, and machine learning models to identify, track, and count items as they move through real material recovery facilities (MRFs). Business Insider reports the research effort officially began in January 2025. Stony Brook also cites a broader, multi-year Waste Data and Analysis Center program supported by a $8 million investment from state-backed funding sources, including work done under a memorandum of understanding with New York State agencies (Stony Brook University; Business Insider).

Per Stony Brook publications, the team collects annotated video from multiple stages of the sorting process at local MRFs and pairs that data with sensor streams to train computer vision models that detect contaminants (Stony Brook University). Project communications show researchers using GoPro-style wearable cameras for worker-centered footage and fixed high-resolution cameras on sorting lines; the materials describe objectives to identify, track, and count objects as they traverse conveyors. The team quotes associate professor Ruwen Qin directly: "We're not just building tools in isolation, we're collecting data at multiple stages of the sorting process, engaging with recycling workers..." (Stony Brook University).

Editorial analysis - technical context: For practitioners, the described approach is a standard applied-vision pipeline: gather domain-specific, annotated video at scale; train detection and tracking models that contend with occlusion, motion blur, and deformation; and deploy inference at conveyor-line latencies. Industry-pattern observations: similar deployments often need continuous domain adaptation, high-quality bounding-box or instance segmentation labels, and robust augmentations for soiling and partial views. Worker-safety and ergonomic monitoring via wearable cameras is a practical dual use of the same data stream, but it raises additional privacy and consent requirements that teams must design for.

Public figures cited by Stony Brook and the EPA frame the need: the university materials cite Environmental Protection Agency estimates that 75% of U.S. waste is theoretically recyclable while the actual recycling rate is 35%, producing an estimated 68 million tons of recyclables sent to landfill; Stony Brook materials also cite a 25% contamination rate in recycling streams (Stony Brook University). Those numbers are used in the coverage to motivate automated contamination detection as a lever to reduce batch rejection at MRFs.

Observers should track whether the project publishes annotated video datasets or model benchmarks, releases code or pre-trained detectors, and documents field performance metrics such as per-class precision/recall, frame-rate at conveyor speeds, and false-reject rates on contaminated batches. Watch for partnerships with municipal MRF operators and any peer-reviewed publications that report evaluation on held-out, in-plant footage.

Editorial analysis: Adoption barriers for real-world MRF AI include dataset generalization across facility layouts, labeling cost for long-tail contamination classes, and the operational integration effort to present alerts to human sorters without increasing throughput delays. For practitioners, successful pilots usually follow a phased path: data collection across diverse sites, iterative model validation in shadow mode, then staged operator-facing deployment with clear human-in-the-loop workflows.

Multiple Stony Brook sources and Business Insider reporting document an active research program combining video, sensors, and machine learning to detect contamination at MRFs; the effort is framed around state-supported waste-data work and early-stage field data collection rather than a commercial product launch (Stony Brook University; Business Insider).