How AI and Robotics Are Transforming Global Waste Sorting and Recycling
Four robotics and vision AI startups are deploying real-time automated sorting systems on conveyor belts, cutting recycling costs and diverting tens of thousands of tons of waste from landfills.
Globally, less than 7% of the materials entering the economy come from recycled or reused sources, while municipal solid waste exceeds 2 billion tons each year—a figure projected to nearly double by 2050. To address this mounting issue, recycling facilities worldwide are increasingly turning to vision AI and robotic automation to accurately identify and recover valuable materials from fast-moving waste streams.
Four startups participating in the NVIDIA Inception program are deploying computer vision and robotic pickers directly onto sorting lines, tackling bottlenecks in plastics, municipal recovery, and textile recycling.
AIRECO Automates High-Purity Plastic Sorting
In Taiwan, startup AIRECO is using vision AI to tackle the economics of plastic recycling. Recycled plastic often struggles to compete with virgin materials due to inconsistent scrap quality, high manual labor costs, and downcycling into low-grade goods.
AIRECO's automated approach focuses on fine sorting to support closed-loop recycling, where discarded plastic products can be processed back into the same grade of product rather than degraded alternatives. At a recycling plant in Tainan, the company's pilot line utilizes NVIDIA Jetson Orin NX modules to run inference and control robotic arms that pick target plastics off conveyor belts.
The test system achieves recognition accuracy above 95%. AIRECO plans to roll out a full production line next year, aiming to expand annual processing capacity from 18,000 tons to 48,000 tons while cutting operating costs by 85%.
EverestLabs Deploys Fast-Paced Sorting Robots
At material recovery facilities, items travel on high-speed conveyor belts where traditional sorting is physically demanding and hazardous. U.S.-based EverestLabs AI has addressed this by installing computer vision-guided robots capable of making rapid picking decisions in real time.
The company currently has approximately 150 robotic units deployed across recycling facilities in North America and Australia. Running edge inference on NVIDIA RTX GPUs and training models on data center hardware using NVIDIA CUDA, the systems classify materials, brands, and packaging types on the fly.
To date, EverestLabs estimates its systems have identified roughly 40 billion objects, recovered around 25,000 tons of recyclable material, and prevented approximately 30,000 metric tons of carbon dioxide emissions. The startup is also rolling out agentic AI software tools to deliver operational analytics directly to plant managers and enterprise executives.
Greyparrot Monitors High-Speed Waste Belts
While robotic arms physically handle items, facility managers also face an information gap: operators have historically monitored waste composition by manually sampling less than 1% of the belt. U.K. startup Greyparrot developed its Analyzer platform to monitor entire waste flows continuously.
Mounted directly above conveyor belts moving at up to 4 meters per second, the Analyzer tracks passing objects by material and brand despite dust and varying lighting. The system uses NVIDIA Jetson modules for edge processing, trained on billions of objects using NVIDIA GPUs.
With more than 300 Analyzer systems deployed across over 20 countries, the platform has surpassed one trillion bounding-box detections. The automated monitoring cuts the time needed to audit 15 tons of waste from 375 hours down to six. Facilities utilizing Greyparrot report overall efficiency gains between 10% and 30%. Global packaging producer Amcor used the technology to boost its yield of food-grade plastics by 12%, while consumer brands such as Coca-Cola, Colgate-Palmolive, and Kenvue use the data to track packaging recovery.
reverse.fashion Speeds Up Garment Classification
Textile recycling faces an even steeper sorting challenge. Europe generates nearly 7 million tons of used textiles annually, with roughly a quarter sorted manually into hundreds of categories—ranging from high-end vintage resale to cleaning rags and furniture stuffing.
Berlin-based reverse.fashion is automating this workflow using high-resolution computer vision pipelines running on NVIDIA Jetson edge modules optimized with NVIDIA TensorRT software. As garments move on belts at 3 meters per second, more than a dozen vision models evaluate small defects, fabric construction, size, brand, color, style, and season.
According to the startup, the automated system can sort in one hour what previously required a full week of manual grading. The technology is currently being deployed to processing facilities across Europe, providing a path toward scaling circular fashion and reducing textile waste.
