Will recycling ever be profitable? AMP Robotics uses artificial intelligence to achieve this

The problem is that developed countries, especially the United States, have historically been unable to carry out this task. The average American produces Nearly five pounds of waste Every day, EPA data comes out. Relatively little of it is recycled: about 43% of aluminum cans, 31% glassAnd only 5% plastic.

These numbers have remained relatively stagnant over the past decade or more, and they are numbers that Matanja Horowitz set out to change. Today, Horowitz’s startup, AMP, based in Louisville, Colorado, has developed technology that could move the needle — using artificial intelligence to snatch recyclables from the overwhelming tide of community trash more quickly and efficiently.

“Recyclables have real value,” Horwitz says. “Somehow, the recycling process was only touching a small portion of the waste stream.” The drawback, ultimately, was the cost of sorting it from regular trash and separating it to reintroduce it into the supply chain. Recycling sorting machines were large, bulky, and expensive, and required human supervision, including, often, forcing humans to wade through piles of trash to handle items the machine missed. (Sorting trash is also something you want people to avoid doing. “You don’t want people looking through needles and hypodermic diapers,” Horowitz says.)

“All that cost means that if you want to produce $1 worth of aluminum, you’re going to have to invest $1,” Horowitz says. “It was kind of a sideline business.”

Horowitz, who taught robotics at the University of Colorado at Boulder and then became a graduate fellow at Caltech, believed that artificial intelligence and robotics could help improve those margins. He launched AMP Robotics in September 2014 (it renamed the company last year to AMP). The premise was simple: “You can unlock the value of these recyclables by reducing the cost of hauling them away,” he says.

Progress was slow at first. It took a year or two for the company’s first robot, installed at its first customer location in Denver, to successfully sort the metaphorical wheat from the chaff. At one point, “I picked up, like, two cardboard boxes or something in a row, and it was like, ‘Oh my God, yeah!’” Horowitz recalls.

Courtesy of AMP Robotics

More installations followed, including AMP’s first “full facilities” — standalone recycling plants — in 2020. The company also raised a $16 million Series A in 2019, followed by Series B round in 2021. In between those rounds, in 2020, AMP occurred Long term agreement With Waste Connections, a US-based waste collection company Reserved $8.9 billion In revenues last year. Initial agreement has been made to deploy 24 automated sorting systems across Waste Connections’ owned container, fiber and waste lines Expanded to include 50 facilities Two years later. (AMP, which remains privately owned, declined to share revenue or profitability numbers.)

“The technology they’re putting out is really amazing…and what they’re doing with AI now,” says Mark Cerizza, division vice president at Waste Connections. “As more and more equipment is available to them online, AI is generally getting better at knowing what is a good target and what is a bad target.”

Smaller but more powerful sorters

The company’s modular recycling systems, powered by artificial intelligence, are smaller and smarter than pre-existing alternatives, which either sorted items poorly by comparison, or required humans to then delve in and sort recyclables by hand. The key innovation is the use of computer vision and robotics to sort materials, especially from raw trash — a feat Horowitz calls “the Holy Grail of the industry.”

What Horowitz refers to as a “robot” appears to an observer more like a complex factory assembly line, with multiple conveyor belts running through a complex network of machines. AMP systems use cameras to scan waste as it moves along processing lines, tracking output through its artificial intelligence system. order It was trained on 200 billion pieces of data collected from hundreds of millions of image examples. Once the material is identified as recyclable, it releases a stream of air onto the object, pushing it away from the belt to be processed.

Courtesy of AMP Robotics

These innovations allow AMP tools to operate at a lower cost and in smaller spaces than traditional alternatives. Horowitz says his systems consistently recover more than 90% of reusable materials successfully, “sometimes that percentage is very close to 100%.” He adds that all of this is done at a cost that is 30% to 50% lower than a traditional recycling facility, which can cost $100 to $120 per ton of material sorted. At the same time, the spatial footprint of the recycling system can be reduced by up to 75%.

As AMP systems improve, they are increasingly able to handle trash streams that traditional recycling facilities refuse to touch. The company’s technology can sort food waste, and it can identify mixed material products, where some parts can be recycled and others cannot. “It’s diapers, dog bags, a lot of food waste,” he says. “It’s brake pads and pads and all kinds of things.” It is then up to the customer to decide what to do with these items, but technology is at least increasing the possibilities of recycling more mixed materials.

“You just have to break the robots.”

In a sense, AMP has grown alongside the AI ​​revolution. In 2014, GPT, the model upon which ChatGPT is based, did not exist. Generative Adversarial Networks, or GANs, have been powering AI image generators It was just invented that yearTheir outputs were small, indistinct, and usually in black and white. Today, artificial intelligence has developed significantly. “Every incremental gain in performance gets easier, faster, and cheaper,” Horowitz says.

Advances in AI prowess — in particular, improved neural network algorithms that learn more quickly and more accurately — have made AMP’s computer vision-based systems stronger, and helped improve the quality of the materials it recycles. But it would be incorrect to say that AMP’s sorting capabilities have improved only because the technology has improved on the curve, Horowitz says. “You have to spend all night in these facilities,” he says. “You just have to smash the robots.”

Perhaps the most significant achievement is that the combination of technological advances, plus long nights spent tinkering with machines in factories, meant that the company went from operating only in stand-alone recycling facilities – where much of the material had been pre-sorted by consumers – to sorting straight through piles of rubbish. This means eliminating more intermediate steps and further reducing the cost of recycling efforts. “We kind of feel like we’ve finally opened it up in the last couple of years, (especially) direct trash sorting,” Horwitz says.

This benefits AMP – and its 100-plus customers, who use more than 400 of the company’s AI systems spread across North America, Asia and Europe. AMP will equip and operate the upcoming Waste Connections facility in Commerce City, Colorado, which is scheduled to open in 2026. The plant will be able to process 62 thousand tons of recycling annuallyAccording to waste communications.

AMP doesn’t have this field to itself, of course: Competitors using robotics to attack the recycling problem include Glacier, ZenRobotics, and EverestLabs. But AMP is rapidly scaling up, and has other, larger facilities to announce soon, Horowitz says. He declines to share details, but he has big visions for the company: “We want to put our stuff in every landfill, and that’s totally achievable,” he says.

Horowitz also believes the technology could be used in other areas, including reverse logistics: handling the returns of clothing or other items. “We backed off because there was a big enough opportunity in recycling,” he says. “But I think about it all the time.”

Clarification: AMP opened its first independent recycling plants in 2020. An earlier version of this story indicated a different date.