Robots, cameras, and lots of data about garbage: Inside the recycling industry's new bet

Americans are really, really good at throwing things away. The country produces nearly 300 million tons of trash a year, and billions of dollars in reusable materials end up in landfills, even after passing through recycling bins.

The problem has always been sorting it all — pulling the wheat from the chaff or in this case, the aluminum can from the dirty diaper. To do that, the industry has historically relied on either shredding everything and trying to separate it mechanically, or paying people to stand over conveyor belts and pick things out by hand. Neither approach scales well. Shredding produces contaminated, low-value material. Manual sorting is slow, expensive, and increasingly hard to staff. Either way, pulling valuable materials out of waste costs about as much as the materials are worth.

Like just about every other industry right now, recycling is betting that AI can change the calculus. A growing number of companies are deploying computer vision, robotic pickers, and massive training datasets of waste to identify and separate individual items on a conveyor belt. It's not just startups: Waste Management $WM, the country's largest trash hauler, is spending more than $1.4 billion automating its recycling facilities. AMP, a Colorado-based company, is taking it a step further, building entire facilities that run on AI from the start. 

"Pretty much anything you and I can identify, it can learn to identify," said Matanya Horowitz, AMP's founder and chief technology officer.

And being able to identify things automatically means the whole business model of waste can change. Instead of charging cities to recycle and hoping commodity prices stay high, the company processes unsorted garbage and makes money in multiple ways, including selling sorted recyclables, converting organic waste into biochar that generates carbon credits, and saving cities a fortune on trucking by cutting landfill volume in half. 

Only 21% of residential recyclables actually get recycled in the U.S., according to The Recycling Partnership, and the number has been declining. Traditional programs depend on people participating, and mostly reach single-family homes. Apartments, commercial buildings, and entire communities without curbside service get left out. Processing trash instead of recycling means everyone's waste gets sorted whether they bother to recycle or not.

"Five years ago, you could do this, but now all of those compounded benefits have made it a no-brainer," Horowitz said. "You're starting to see entire metro areas where sorting through the garbage is just definitively a better thing to do."

From shredder to scalpel

Household trash is one problem. Electronics are another. For all the sophistication that goes into building a TV or a smartphone, taking one apart at the end of its life still mostly involves a screwdriver and a hammer.

Robots could speed this up considerably, according to Matt Travers, a roboticist at Carnegie Mellon University who co-founded roboLoop, a recycling startup. The company focuses on flat-panel displays, which are piling up faster than almost any other type of e-waste. 

"You're never going to run out of them, and nobody really knows what to do with them," Travers said. 

At his facility in State College, Pennsylvania, robots use computer vision to locate screws on the backs of TVs and punch them out with a kinetic bit inspired, Travers said, by the cattle prod turned murder weapon in No Country for Old Men. Humans then pull out components, separating materials, steps that need more dexterity.

The whole operation is organized around one thing: getting to the logic board. It contains gold, copper, and palladium, and is worth roughly 100 times more per pound than the steel and aluminum around it. With manual labor alone, the facility processes about 25 TVs an hour and loses money. Travers said the robotic line could hit 120.

Travers sees flat screens as a starting point. Electronics are all built roughly the same way — layers held together with screws — and the same robotic capabilities could apply to cell phones, tablets, or anything else still being recycled by hand. He's calling the goal "generalized disassembly intelligence": train a robot to understand how things are put together, and it can learn to take anything apart. 

"It's a vice and a hammer," Travers said of the tools currently used across e-waste facilities. "I can automate that."

Textiles face a version of the same challenge. The fashion industry produces enormous volumes of waste, but most of it is blended fabrics, which makes it extremely difficult to recycle into usable fiber. At the Hong Kong Research Institute of Textiles and Apparel, researchers have built an AI sorting system that classifies incoming garments by material, condition, and whether they're suitable for resale or recycling.

Items that can be resold get routed one way. Blends that can't get sent to a chemical separation process that breaks them back down into reusable fiber.

The technology is operational and has been licensed by manufacturers in the U.S. and Indonesia. But Jake Koh, the institute's chief executive, said the familiar barriers remain. "Virgin materials are often cheaper," he said, and collection infrastructure is too fragmented to deliver consistent supply. Without regulatory pressure, brands have little incentive to buy recycled fiber. 

Europe's upcoming extended producer responsibility mandates could shift that, but for now, profitable operations still depend on policy support and brand partnerships rather than market demand alone.

Waiting for buyers

The technology is getting there. The market for what it produces is a different story. AMP's systems can identify and sort roughly 90% of materials in the waste stream, but buyers exist for only 50 to 60% of it.

Part of the problem, said Callie Babbitt, a professor at the Rochester Institute of Technology who researches sustainability, is that recycling isn't like warehousing or logistics, where a few massive companies can invest billions in automation and scale it across hundreds of facilities. It's a fragmented, largely mom-and-pop industry constrained by municipal budgets and razor-thin margins. 

Even the biggest players operate facilities that look completely different from one region to the next, because local infrastructure and waste streams vary so much. The development costs for a robotic system run into the hundreds of thousands of dollars, and it can take years before anyone knows if it works. For a single-facility operator, that's a hard bet to make when the return on what you're sorting might be a few dollars a pound.

Scaling is complicated further by the fact that there's no federal policy for most of this. E-waste is governed by 25 different state laws, all slightly different. Organic waste requirements vary state to state. If you're a company trying to build a national business around recycling technology, every market is a different regulatory puzzle. 

"We're not necessarily designing products at the front end that are well suited for being processed at the back end," Babbitt said.

She thinks the more interesting opportunity for AI might actually be further upstream — before things become waste at all. Smart bins that tell commercial kitchens what they're throwing away. Phone software that nudges you to recycle when your battery health drops and points you to a facility nearby. Apps that help consumers find quality refurbished electronics instead of buying new. 

For all the progress in sorting trash, Babbitt said, the real win would be making less of it.

"From a sustainability perspective, it's far better to prevent waste than it is to recycle it," Babbitt said.