America wastes $6+ billion worth of recyclables a year. Can robots and AI help?

Business Insider

TK TK

Transcript

00:00Every year, America throws away $6.5 billion worth of recyclable goods like paper, aluminum,

00:09and plastic.

00:11And the household recycling that gets mixed up in blue bins costs hundreds of millions

00:16of dollars to sort.

00:19That job can be dangerous.

00:21Things like dirty diapers, surfboards, televisions, even hand grenades come through fairly regularly.

00:29Now a growing number of companies have deployed AI and robotics to help make recycling safer

00:35and more efficient.

00:37Mattanya Horowitz, founder of the robotics company AMP, spent six years perfecting his

00:43trash handling robots.

00:45This little robot right over here, it's just a valve.

00:47It just uses a little piston and kind of flicks things off from the side.

00:51We went to AMP's biggest recycling plant in Cleveland, Ohio to see if AI can help fix

00:57The U.S. is one of the world's biggest waste producers, and only recycles about a third

01:06of its eligible trash.

01:08That's much lower than other countries like South Korea and Germany, which recycle more

01:12than half of their garbage.

01:15In America, waste management is a local issue.

01:18The country has over 9,000 different recycling programs, making it difficult to standardize

01:23at scale.

01:24As a result, U.S. recycling varies greatly by state.

01:28Today, California and Oregon have the highest residential recycling rates in the U.S. at

01:3437 percent, while Mississippi and Louisiana have the lowest, at 8 percent.

01:41For decades, the U.S. could rely on countries like China to buy mostly unprocessed trash.

01:47From 1992 to 2017, the U.S. exported $11 billion worth of recyclable waste to China, largely

01:54unsorted.

01:55But in 2017, China said it no longer wanted to import contaminated materials like plastics

02:02and paper products.

02:05As a result, U.S. facilities had fewer places to sell their recyclables, and many struggled

02:10to turn a profit.

02:12Facing steep costs, many sent their waste to landfills or incinerators.

02:18These challenges created an opportunity for robotics companies like AMP, which Matanya

02:23founded in 2014.

02:26There's value in the plastics, there's value in the aluminum, but the problem has been

02:30that you needed all of these manual sorters to pull that material out, and there's a lot

02:34of cost associated with that.

02:36And you're left with this kind of marginal business.

02:39At its start, AMP put its own trash-sorting robotics in other recycling facilities.

02:44In 2022, the company began operating its flagship facility in Cleveland, Ohio.

02:49This AMP-1 facility, as it's called, runs on AMP machinery and software.

02:54A big part of what we do is focused on making that automated, making it so people don't

02:58have to touch garbage, and also making it so that this technology can sort through dirtier

03:02and dirtier material streams, so you can require less of consumers and really have a solution

03:07that is pervasive.

03:10The facility gets its waste from plastic manufacturers and materials recovery facilities, or MRFs,

03:16that have earmarked these bales for landfill.

03:20For most of America's recyclables, this is their only chance to avoid being sent to the

03:24dump.

03:26Dozens of workers use specialized machines, like trommels and screens, to sort trash.

03:32MRFs can sort roughly 87% of the materials they receive.

03:36But workers can be exposed to sharp objects like used syringes and batteries or aerosol

03:41cans that can explode.

03:46This plant in Cleveland has 15 employees, and they rarely touch garbage.

03:51Instead, a fleet of cameras and air jets get the trash sorted.

03:57Kevin Papich's job is to make sure AMP's automated plants are running smoothly.

04:01We view this building as an entire robot.

04:05There's very little hands-on sortation that takes place in this facility.

04:09Obviously, we're maintaining the line, keeping the place orderly, and producing quality bales.

04:17Up to 10 trucks arrive daily, dropping off a total of about 200 tons of mixed recyclables.

04:23It all arrives in giant, 1,500-pound bales.

04:27Forklift operators drop them into a reducer, which breaks them apart with metal teeth.

04:31The material we get in this facility comes from the eastern half of the United States.

04:36We receive material as far away as Iowa and Alabama and Florida.

04:41Cliff, you copy?

04:43Cliff?

04:44Go for Cliff.

04:45Full line, start.

04:46Full line, start.

04:47Thank you, sir.

04:49After the bales are broken, they come through the sortation line.

04:53The first step is the ferrous offtake.

04:57A rotary magnet pulls iron-based metals off the line.

05:00We get baling wire, car parts, food cans, that sort of stuff.

05:05Next, air jets blow away paper scraps and flexible packaging, like the thin, clear plastic

05:11that's wrapped around food and toys.

05:14Amp's jets can sort thousands of recyclables a minute, more than 10 times the amount its

05:18robotic arms can.

05:21Any remaining materials travel along these belts, passing under cameras that ID and track

05:26them using Amp's custom-made, deep-learning neural network.

05:32This network uses multiple layers of nodes, like neurons in the human brain.

05:38These nodes work together to analyze patterns like textures and shapes to ID specific material.

05:45Amp scans more than 50 billion objects a year.

05:49The more it scans, the more it learns.

05:52We're focused on a system that you're not really programming so much as teaching.

05:56You can teach them how to identify number one plastics, even though they're smashed

05:59and folded and dirty and inconsistent, and with that, you essentially have a garbage

06:03sensor.

06:04And that garbage sensor is what the industry's been missing.

06:07Amp's system can also target specific materials and colors.

06:11We've had somebody come to us and say, hey, I didn't know you could sort color, can you

06:15give me just clear and natural polypropylene?

06:19And we were able to do that by just adjusting the neural net and giving them that fail.

06:25Because everything is tracked, the plant can detect problems on the line.

06:29So pretty much every corner of the facility is being monitored for quality, for contingencies,

06:33like is a jam forming?

06:35Is there material where it's not supposed to be?

06:37Eventually, Amp's trash gets to the recirculating line, a belt that moves missed material back

06:42into the sortation line.

06:44We're currently capturing 85 to 90 percent of all the product we send through the line.

06:50Once the materials are sorted, they fall into their own bunkers.

06:53From there, those bunkers are emptied and then moved to the baler operation, where we

06:58can actually bale each unique commodity and then it goes into finished goods for sale.

07:03Bales of recyclable plastics get more valuable the better they're sorted.

07:07The higher the purity, the higher the price.

07:09We're always geared up for 90 plus percent purity inside of our bales, and I think we

07:14hit that 100 percent of the time.

07:17Amp sells those bales to other recycling mills, which use them to make plastic bottles, aluminum

07:22cans, and car parts.

07:24The company also sells its tech to other recyclers.

07:28Today, its systems support nearly 100 facilities around the world.

07:34But Amp's machines can't ID every piece of trash.

07:37This is where AI still needs an assist from humans.

07:43Images of missed material are sent to a team of human annotators in India, who identify

07:48and label them in the neural net to train the AI.

07:52When annotators are going through that data, looking for mistakes, we are also always getting

07:57more specific about what we identify.

08:00So at first it was enough to say that this was a number one plastic, now we say, oh,

08:04we also want to know what color it is, what form factor it is, what brand it is.

08:10The company says its AI-trained machines and leaner workforce help make Amp Ones up to

08:1550 percent cheaper to run than traditional MRFs.

08:19It also runs an R&D lab in Denver, where the team studies information gathered by its

08:24hundreds of sorting devices around the world.

08:26They're all uploading data to this cloud infrastructure we built.

08:30We then have people go through that data set and say, here's what's number one plastics,

08:35here's aluminum cans, and everything else.

08:37Joe Castagnieri leads Amp's software development.

08:40The video feed is coming in, it's received in the computer, and we pass it through the

08:45brain of Amp's system, which is the neural network, which is looking at those images,

08:50trying to figure out what those things are.

08:53These green outlines are for natural high-density polyethylene.

08:57There's also an aluminum can under there.

09:02But making sure these computers work under real-life conditions is a challenge.

09:06You have to get it to work in a hot, dusty, trash-filled environment that's 80 degrees

09:12or 40 degrees.

09:14The company says it has buyers for all the waste that flows through the AmpOne facility

09:19in Cleveland, but it has trouble selling waste at some of its other facilities.

09:23The real challenge is, there are a lot of materials we can recycle where there isn't

09:29what we call an end market.

09:30There isn't somebody who will buy it and use it.

09:33Number one plastics that's colored, almost nobody wants.

09:36We can sort it out all day, nobody will buy it, and it means that we sort it out, and

09:39it just goes right to the landfill anyway.

09:42Part of the problem is that making new plastic is often cheaper than recycling it, especially

09:46when oil prices are low.

09:49Some U.S. states have passed extended producer responsibility laws, or EPRs, that hold manufacturers

09:55accountable for the waste they create.

09:58Right now, five U.S. states have passed EPR laws targeting packaging waste.

10:02Broadly, I think these rules are important and they're good, but I worry that people

10:07are going to think that incremental improvements are good enough.

10:12People should be aiming for doubling the diversion rates that they have now, and they should

10:15be aiming for it to do that in the next several years.

10:18And it's possible.

10:19It's possible with technology like ours, it's possible with other technology.

10:22For now, he's focused on growing Amp's facilities and getting its tech in more recycling centers

10:27around the world.

10:28If we can make recycling and waste management a more profitable business, we can make waste

10:33management infrastructure spread across the world more quickly, and we can stop having

10:37plastics in the ocean and things like that.

10:40So when you look at these huge global problems and you see how recycling can become a solution

10:44to them, it's very exciting and very motivating.

10:53The hardest part of the job, it is not the robots.

Category

Transcript

Recommended