Of the 8.3 billion tons of virgin plastic waste created each year, despite decades of efforts to reduce the amount that ends up in landfills, only about 9 percent gets recycled.
London-based computer vision startup Recycleye looks to give those recycling numbers a big boost with its AI-driven system for identifying waste materials.
By automating and speeding the movement of materials through sorting systems, and identifying them with more precision, Recycleye says it can significantly increase capacity for recycling companies while upping the overall recovery rate.
Pretty lofty promises from a two-year-old company that started in the most unglamorous of ways.
“We went out and collected trash from bins, took it back to my garage, and started building our first proof-of-concept,” said Peter Hedley, chief technology officer at Recycleye, which is a member of the NVIDIA Inception accelerator program for AI startups.
One Man’s Trash Is Another Man’s Treasure
Hedley and company co-founder and CEO Victor Dewulf started discussing the possibilities of their technology during their masters’ work at the Imperial College of London, where they worked on applying computer vision to waste streams.
Dewulf ultimately wrote a paper on the topic, got interest from academia and industry, and then left his job as an analyst at Goldman Sachs to start working on a Ph.D. so he could refine his idea. The next thing Hedley knew, Dewulf had him dumpster diving and developing the idea into a commercial product.
Buoyed by acceptance into Microsoft’s AI accelerator program, they found themselves on the receiving end of an £800,000 (about $1.1 million) seed investment, followed by another £400,000 in grants in 2020.
By the end of 2020, the startup had already proven itself with the top waste-management firms in the U.K. Last April, they partnered with energy leader TotalEnergies and recycling pioneer Valorplast on the OMNI project, one of seven winning projects selected by nonprofit French sustainability company Citeo.
An AI on Closing the Recycling Loop
To date, effectively recognizing and separating items that have contained food from other items has not been possible. However, a crucial step in improving recycling rates is to optimize the quality of recycled materials that are passed on to plastic manufacturers.
Recycleye’s partnership with Valorplast and TotalEnergies focuses on the application of AI to identify food-grade and nonfood-grade plastic packaging with the goal of increasing the circular recycling of these products. It could even help in the development of new applications, such as improved food packaging.
Recycleye has also partnered with universities to create WasteNet, an open-source database that is now the world’s largest waste dataset, with more than 2.5 million training images.
Recycleye is an NVIDIA Metropolis partner, providing offerings that integrate the full Metropolis stack for video analytics inference. The Recycleye team uses the NVIDIA Jetson platform and dug into accelerated deep learning tools such as pretrained models, the NVIDIA TAO Toolkit and the NVIDIA DeepStream SDK.
With the NVIDIA tools and huge training dataset behind it, Recycleye has slashed the time it takes to deploy its model from an unworkable two months down to just two hours, while achieving accuracies exceeding human vision in identifying items.
The company’s devices run on the network’s edge, doing all computations onsite, provided there’s sufficient internet connectivity. Models are trained in the cloud, on a GPU-enabled instance of Microsoft Azure, and then deployed to Recycleye’s devices at client sites.
Algorithms are automatically updated on client devices during software updates. The cloud system also processes data logs and provides the client with dashboard summaries of that data.
Coming: Global Expansion and More Robots
Recycleye also develops robots powered by Recycleye Vision, jointly with FANUC, one of the world’s largest robotics manufacturers. Having already installed Recycleye Vision and Recycleye Robotics in the U.K. and France earlier this year, Hedley expects more waste companies will follow suit to automate their manual sorting with robotics.
But Recycleye has much bigger aspirations now that it has a scalable system that can be deployed globally, and Hedley said that the more deployments the company does, the better its technology will get.
“When we have a device at both the end and the front of a line, we can see what happens if the quantity of material increases, but the quality decreases,” said Hedley. “We can start optimizing machinery and configuration, and start having machines making decisions.”