Sustainability in Your Ear Transcript: Glacier CEO Rebecca Hu-Thrams On How To Build An Advanced Manufacturing-Based Circular Economy

Mitch Ratcliffe 0:00
Hello, good morning, good afternoon or good evening, wherever you are in this beautiful planet of ours. Welcome to Sustainability In Your Ear. This is the podcast conversation about accelerating the transition to a sustainable, carbon neutral society, and I'm your host, Mitch Ratcliffe. Thanks for joining the conversation.

Today, we're going to be talking about waste, or rather, turning waste into something useful.

Every week, millions of Americans dutifully sort their bottles, cans and cardboard into blue bins, trusting that those materials will be recycled, and the reality is disappointing. As much as 80% of what we put in those bins is not recycled. The materials end up in landfills, incinerators, and sometimes it's shipped overseas. The recycling system leaves as much as $200 billion worth of recoverable materials unrecycled each year. And the problem isn't our intentions. It's the dated technology at material recovery facilities. That's MRFs, that's where recycling actually happens.

These vast sorting operations are struggling with a labor crisis so severe that facilities often have to refill the same sorting job five times a year. The work is dangerous, with injury rates twice that of construction and the pay isn't great.

Our guest today, Rebecca Hu, is working to rebuild the recycling system from the inside out using robotics. As CEO and co-founder of Glacier, she's introduced AI-powered robotic sorters at facilities across the country. Glacier robots already processed recycling for one in 10 Americans. They use computer vision trained on more than 3 billion images of waste to identify and sort more than 70 different materials from the waste stream. They pick 45 items per minute, operating 24/7 in conditions that would exhaust or injure human workers. According to Glacier, each of its robots prevents over 10 million items per year from languishing in landfills.

Glacier is creating an intelligence layer for the circular economy, and that's a phrase we've heard in several recent interviews. These robots do more than sort—they also generate data about what's actually in the waste stream. They can report on which brands are creating recyclable packaging and where value is being lost because the packaging is not recyclable. That data is already reshaping how companies like Amazon think about packaging design, and they're moving from technically recyclable to provably recyclable. In fact, Amazon has invested in Glacier through its Climate Pledge venture fund.

With extended producer responsibility laws forcing brands to account for their packaging, robotics and AI can help provide accountability for communities and for brands. And you can learn more about Glacier and its robots at endwaste.io—endwaste is all one word, no space, no dash, endwaste.io.

So, can AI teach us to close the loop? Let's find out right after this quick commercial break.

[COMMERCIAL BREAK]

Welcome to the show, Rebecca. How you doing today?

Rebecca Hu-Thrams 3:14
I'm doing fantastically. Excited to be here.

Mitch Ratcliffe 3:18
Well, thank you for joining us. Glacier is doing a lot of interesting stuff, but your background is one of the reasons Glacier exists. And I'm curious, as a first generation American, you grew up with parents who told you you had to wash and reuse things like margarine tubs rather than throwing them away. How did that experience shape your relationship to the idea of waste?

Rebecca Hu-Thrams 3:37
Yeah, my upbringing definitely played a pivotal role in my relationship to waste. The way that my parents raised me was to really see the opportunity inherent in any given object, rather than considering it as just something to be easily disposed of. And so not only were we washing out margarine tubs to reuse as containers, you know, old T-shirts that developed holes would end up becoming dish rags and so on and so forth.

And so when I juxtapose that philosophy, which really was our household philosophy from day one, with a lot of the wasteful behaviors I ended up seeing in the world as I grew older, it created this immense cognitive dissonance that, at times can be distressing, but also really paved the way for me to see a huge opportunity to help all of society re-envision the things that it's currently discarding as waste into the value of the material that those things are made of, and to find second lives and new uses for all of these types of materials.

Mitch Ratcliffe 4:32
We talked about the fact that up to $200 billion in materials is just lost every year. How would that change with the advent of computer vision and robotics in the sorting process?

Rebecca Hu-Thrams 4:46
Yeah, the thesis behind Glacier is essentially that if you can find a way to financially align incentives, that's the best way to create change very rapidly in a capitalist society like ours. And recycling is a fantastic example of that. If you can find a way to generate value by recycling something and turning it into a new thing, rather than just tossing it into a landfill for the rest of time, you've done something fundamentally good, not only for the planet, but also for many people's wallets.

So the amazing thing that Glacier is capitalizing on is that technology, for the first time in history, has gotten robust enough and dynamic enough that we can actually apply things like computer vision, AI and robotic automation to helping to better sort and clean up our recycling stream so that we can separate more materials, more cost effectively and at a higher purity level, so that they can be turned into more valuable materials again. And in doing so, we aim to fundamentally change the unit economics of recycling, make it really, really profitable for all of the operators and stakeholders involved. And in doing so, make that the sort of natural end of life for any piece of material that still has value.

Mitch Ratcliffe 6:04
It really would be something to be able to unleash that $200 billion in value, including for the new jobs and the roles in the community that that would support. But let's talk about those robots. We're not talking about humanoid robots. So can you explain the robot's role in the sorting system? How does it fit in?

Rebecca Hu-Thrams 6:22
So for those who are listening, who maybe have never been to a Materials Recovery Facility, also known as a MRF, what happens to your recycling after you toss it into your recycling bin is it gets taken to this MRF, essentially a large industrial manufacturing plant of sorts. It gets dumped onto a huge pile of commingled recycling, a lot of which is also trash, and then it gets shoveled into a series of sortation steps. So imagine large piles of commingled waste and recycling going through conveyor belts. And then the goal of this MRF is to sort that commingled pile into beautiful bales of plastic, aluminum, cardboard and so on, and all of those materials are then sold to end market manufacturers to be turned into new things.

Now, in practice, the sortation process is incredibly effortful and not always that effective. You can imagine the sorts of things that people are putting into their recycling bins are not the same pristine bottles and cans you buy at the grocery store. In fact, I often ask recycling operators when I visit them, what's the weirdest thing that you've seen coming through the line at your MRF? And I get crazy answers—oftentimes, not only things like surfboards and microwaves and engine blocks, but also really hazardous things like hand grenades and firearms and hypodermic needles and so on. And so that sortation process is extremely effortful, time consuming and so on.

The robots play an incredible role in the sorting process. First, we have our AI cameras, so they're taking constant footage of what's coming down that conveyor line at various parts of the sorting process, and that AI model can detect in real time—this is an aluminum can, that's a cardboard box and so on and so forth. We can do this both for good materials that you want to recover as well as trash that you want to pull out of the stream to decontaminate what you're sorting.

And not only is this data that we're collecting incredibly informative to a variety of operators in the circular economy, it also acts as the eyes and the brain of our robot. So that robot is sitting a little ways downstream on that conveyor belt, and based on the knowledge that it's being fed, it can actually move its arms to go pick up the valuable things or the contaminants and sort them into the right place.

At a high level, this sounds fairly straightforward in concept. In practice, I often call recycling the most demented form of manufacturing on the planet, because unlike every other form of manufacturing, there is so much volatility and heterogeneity in the types of materials that you're sorting that it wasn't until actually very recently, that AI and robots have gotten good enough and robust enough to actually sort out such a varied stream of materials effectively.

Mitch Ratcliffe 9:15
How do you describe what it sees? Because you specifically talk about the fact that you can identify specific brand packaging, for instance. Is it looking for those things, or is it ruling out most stuff and picking what it doesn't rule out? Can you explain how the AI makes those decisions?

Rebecca Hu-Thrams 9:35
A lot of what our AI does is actually still dependent on what we decide that it wants to do. And so I oftentimes talk about this idea of creating a full taxonomy of stuff. In other words, what is the right way to categorize the massive world of all of the items that could be coming through a recycling facility so that the AI knows what it should be looking for?

Typically, what we found is that while AI can detect various different types of items, it performs best when you give it a white list of sorts. In other words, you know, these are the categories that you should be on the lookout for, and that list of categories can be infinitely long. Today, you know, our core ability of our AI model is actually, we can detect, you know, over 70 different types of materials and form factors already—things as generic as an aluminum can or as specific as a toothpaste tube.

And as you mentioned, Mitch, we can actually now get really, really good at detection down to even the brand level, in a way that the human eye might not necessarily be able to pick up. So not only a toothpaste tube, but a Colgate toothpaste tube.

Mitch Ratcliffe 10:39
That's really interesting. I want to dig into that a little bit more. But first, let's talk a little about contamination. Can the robot also identify a material like, for instance, a cheese-contaminated, grease-soaked pizza box and rule it out of the picking process, or can it send it to the right place to deal with that kind of material?

Rebecca Hu-Thrams 10:59
In brief, yes. The way that the AI model works is not dissimilar to how you might teach a person to be on the lookout for something visually. So we often tell our users and our customers, if you could stand there at the conveyor belt and visually see that something is, you know, a container that has liquid in it, or a pizza box that has stains on it, you can also teach the AI model to detect that as well.

Now this gets back to the beauty and the complexity of what do you teach the AI model in terms of what it should care about? And so step one was to just lay the foundation of the basic types of categories. Right? A pizza box might be classified as a cardboard box of sorts, but now that we've gotten really good at detecting those basic categories, you can start to layer on features like, what is the brand on the box, or does this PET water bottle have its lid on it? And so on and so forth.

And so one of the incredible things about the AI is as our understanding of the recycling and waste stream evolves, and as we hear about new use cases from our users, the AI can kind of continually keep building on itself, such that, you know, a robot that was deployed today in three years might be benefiting from a totally new and improved AI model that can detect far more things than it could even today.

Mitch Ratcliffe 12:18
Can it speed up over time? Right now, your system picks about 45 items per minute. And I'm curious, how does that compare to a human sorter, who, of course, also can look up and miss stuff? But can it evolve its capacity to do more in addition to recognizing more?

Rebecca Hu-Thrams 12:35
When we think about how to improve technology like Glacier's, there's actually two different technology stacks we're talking about. The first is our AI model, and to improve the AI includes things like adding more types of classes that we can detect at higher accuracy and more quickly in real time. The second piece is around the robotics. And so this is more around the hardware capabilities of how quickly our robot is picking, how good it's grabbing.

The second piece is around the robot, which is about how quickly it's able to pick, how effectively it can grab certain materials and so on. So on both fronts, Glacier is rapidly improving our capabilities. On the AI model front, we're actually seeing a lot of innovation, not only in what we're able to detect, but also how we can apply that growing data set to help MRF operators understand how their plants are running, recover more material and decontaminate their streams.

And then, on the robotics piece, we're seeing massive improvements in the ability of industrial automation. And so not only does that mean that we can get the arms to move faster through, you know, hardware improvements, but also software improvements, like, how do you get more juice so to speak out of your motors to increase the pick speed on the arms, but also things like, you know, what are the right end effectors, like the suction cup, for example, that can allow you to pick a wide variety of different materials. The type of cup or end effector you use to pick large pieces of cardboard might look very different than what you use to go after smaller items, like plastic bottles.

Mitch Ratcliffe 14:09
Yeah, the end effector, by the way, for our listeners, is the robot's hand for better comparison. So I also understand that AMP Robotics, which is one of your competitors, has a faster pick time. But how do you pitch yourself versus other robotic options?

Rebecca Hu-Thrams 14:24
It's important to understand that Glacier actually has a purpose-built recycling robot. And what I mean by that is, compared to a lot of our sort of competitive set, which is oftentimes bought off-the-shelf robotic sorting units from companies like ABB and Omron and FANUC—now these large manufacturers that make robots for warehouses and auto assembly plants and pharma manufacturing, all different use cases—in contrast to that, Glacier actually saw the need to essentially design a robot from scratch.

And we don't win any brownie points just for designing a purpose-built robot, but the reason we did it is because when we came to market, we interviewed dozens of recycling facility operators on their opinions about other robots. And what we found is that, you know, automation was something that they were eager to embrace, but that a lot of the solutions on the market were extremely expensive for the job they were doing. So we're talking about, you know, a five to seven year payback period for a robot that might only last five to seven years. Yeah, these robots are really physically large, so let's say about 10 to 12 feet in length, in a very space-constrained environment. And that size also adds to the price tag, because that means that you're spending hundreds of thousands of dollars on retrofits to get that thing into your facility in the first place.

In contrast, what Glacier has done is essentially built a robot of our own design specifically oriented around delivering really excellent payback period for these facilities by picking a wide variety of items at very high efficacy with very low cost. And so we end up with a robot that is only designed to take up about the same amount of space as a person. It's about three feet wide. We install it in less than a day, so there's no downtime, and we typically do it with minimal retrofits.

Mitch Ratcliffe 16:12
So, no downtime is the big—I was really impressed by that. I mean, less than a day in a market where a month of downtime to retrofit a waste picking line is standard is incredible. How did you do that?

Rebecca Hu-Thrams 16:27
Yeah, I mean, one observation that we made in the early days is that actually the complexity of a lot of the off-the-shelf robots is in some ways over-engineered for what you need in this use case. So what we did was we basically removed a lot of the moving parts on other off-the-shelf robots, and we ended up with a linear gantry system that essentially moves back and forth across the belt and up and down to pick up the item and deposit it in the right place. This gets you a robot that's significantly less expensive, so more affordable for these MRFs, and it allows you to focus on really doubling down on engineering the subsystems that need better reliability.

So this is something that's unique to recycling robots—in some ways you're not, you know, in a super clean environment handling uniform material, you have to deal with a wide variety of potentially really hazardous and damaging items, and build your system in a way to address that.

One other nuance that a lot of our customers have started to understand as well is that metrics in isolation are not quite the right way to evaluate new technologies. So when you think about something like pick speed—intuitively, it makes sense that if something can pick at 80 picks per minute, that's better than 40 picks per minute. But the way that our users are starting to understand the technology is that it's not just increasing pick speed at all costs. It's actually much better to evaluate pick speed per dollar spent, or pick speed per square foot that the technology is occupying in your facility. And when you think about Glacier's comparison to a lot of our competitors, we're actually able to deliver really, really favorable picks per minute, per square foot, or per dollar spent, and that's really the key goal that we're optimizing for.

Mitch Ratcliffe 18:12
So the MRF can make more money, is the best way to put it. And if you do that, let me ask a question about the role of AI in terms of those workers that are displaced. Are they being reassigned to other work that improves the overall performance of the MRF, or are they simply being replaced by the AI?

Rebecca Hu-Thrams 18:28
It's important to contextualize this technology against the backdrop of what's happening to labor in the industry. These robots that we're installing are doing these frontline sorting jobs, which is a role that typically sees about 500% annual turnover. In other words, if you're operating one of these MRFs, you have to hire five different people to fill that role over the course of a year.

Mitch Ratcliffe 18:50
It is one of the worst jobs anybody can imagine.

Rebecca Hu-Thrams 18:54
Absolutely. It's really, really intense as you can imagine. It's certainly very dirty, and one thing people don't realize is that it's actually about twice the injury rate of construction work, not only due to things like repetitive stress injuries and dealing with industrial equipment, but also the types of hazards that come through the waste stream.

Now what we saw was a chance for automation to not only fill this void, because if you cannot fill that sorting void, then as a MRF, you go out of business because you live or die by your ability to effectively sort that material stream. But on top of that, we've seen a huge amount of appetite to upskill the existing workforce, such that the workers that are in these facilities have higher paying and more skilled jobs.

As Glacier, one thing we're really proud of is every time we deploy our technology at a new facility, we actually do a comprehensive training and development program with the staff that work on site. It's our belief that, while we're certainly there to help support our customers, we never want to get in the way of them being able to maintain the equipment with their own means. And so what we end up with is dozens of workers at this point that have become really, really skilled at interacting with, maintaining, troubleshooting and even directly repairing cutting-edge technology like ours.

Mitch Ratcliffe 20:11
So right now, about 21% of all recyclable materials is actually captured. With AI in the system, what do you think the recycling rate can be in 2030 or 2035? Where are we going?

Rebecca Hu-Thrams 20:23
So there's a lot of really exciting tailwinds for recycling on the horizon. I'd be remiss to not mention certainly some of the upcoming legislation in the US. There's something called extended producer responsibility, or EPR. And the idea of those laws is that, you know, all of the folks that are producing this packaging that needs to get disposed of, should also have some skin in the game to fund the system that recovers that material.

These types of legislation are already in effect in several other countries, like Europe and Canada, and I think we can look to the efficacy of that rollout for some sense of what it might do for recycling in the US. I think in certain areas of Europe, it's actually improved the recycling rate by, I believe, over 40 percentage points. And so it is possible to achieve massively improved recycling outcomes when you actually incentivize the creation of recyclable goods and fund a system to manage that well.

However, I think legislation on its own is not enough, and this is where Glacier comes in. We want to build the data layer that helps regulators, producers, recyclers, actually understand where all of their material is ending up and what is or isn't being recycled. We want to build the automation that can recover that material at much lower cost, which incentivizes the recovery of more material.

And then finally, when you think about recycling as a manufacturing exercise, one thing that all manufacturers care about is the quality of their incoming feedstock. In other words, how good is the stuff that you need to turn into new material? And in this case, that is the stuff we're throwing out. So what we're seeing is that the data we're collecting not only helps MRF operators run better facilities, we can actually help use those insights to instruct major brands and producers like Amazon, like Colgate-Palmolive, etc., on what designs are actually more or less recoverable, so they can help increase the quality of the recycling stream.

Mitch Ratcliffe 22:33
We've taken a deep dive into the trash so far, but I want to pick through a lot more when we come back. We're going to take a quick commercial break folks, stay tuned.

[COMMERCIAL BREAK]

Welcome back to Sustainability In Your Ear. We're talking with Rebecca Hu. She is co-founder and CEO of Glacier. It's an automated sorting system that's used in MRFs—that's material recovery facilities, folks—in San Francisco, Seattle, Chicago and other cities.

Rebecca, you raised the notion of EPR, extended producer responsibility, because the Glacier system can explicitly identify a brand. Is this potentially a part of how a company is going to provide accountability for its attempt to successfully recover its packaging and other materials from the waste stream? Is this the beginning of accountable EPR?

Rebecca Hu-Thrams 23:27
You know, it's still in the really early innings of EPR in the US. So I'd have to be sort of reading the tea leaves to give you my prediction there. It's certainly something that the technology is capable of doing. So if we have the ability to recognize plastic bottle, we can now recognize not only the brand of plastic bottle, but also even at the SKU level.

In some cases, what we're seeing with EPR right now is because we're in the early innings before we can get to the brand level requirements, there's even just an understanding of what types of form factors and what material resins are being recycled or not effectively recycled. So that's where we're seeing most of the interest from both regulators and the brands that we work with. It might be things like, you know, what happens to PET bottles versus PET thermoforms? Where are cartons ending up, and so on and so forth.

But as that starts to get cemented and we have this general understanding of what packaging formats are or are not widely recyclable, I think you will start to see specific producers and brands taking a vested interest in where their owned material is ending up.

Mitch Ratcliffe 24:33
Let's talk a little more about Amazon, because they have invested in Glacier, and you're running a pilot program with them tracking packaging recyclability. What does that partnership look like in practice? And what are you learning about the gap between recyclable and actually recycled?

Rebecca Hu-Thrams 24:49
So we have several projects going with Amazon, and I do commend them for being good stewards of the packaging stream that they're putting out into the world. One of the most interesting projects we've done with them is to actually help them understand the ability of AI to trace materials like bio-based plastics in the waste and recycling stream. And you know, this is pretty experimental stuff—the insights from this project may inform the nature of packaging designs many years out.

But what I loved about this project is that we worked in collaboration with an Amazon packaging design team as well as one of our MRF operator partners, so sort of this three-legged stool of sorts, to say, can we train Glacier's AI to detect these specific bio-based plastic materials as part of the regular recycling stream? And furthermore, can we teach our robot to effectively recover that material?

Now, what we're trying to validate is, if this type of material were to make it onto the mass market, what does the recycling pathway actually look like for that material? You know, it's one thing to say, Hooray, we've technically designed something that can be recycled. It's another to validate in the real world. Is that stuff actually going to be able to get recycled?

Mitch Ratcliffe 26:07
Well, a lot of that is a matter of getting it to the right processor with the right level of contamination for it to be viable, to do it, to do the process economically. Are you finding that the way that these organizations are thinking about their packaging is changing to reflect their ability to track it throughout its lifecycle? Is that changing the way they think about their material and potentially connecting them to the customer in a different way?

In fact, I recently talked with Tom Szaky, the CEO of TerraCycle, and he said packaging systems should be part of the customer service experience, not part of the waste management system. And does that mean, for instance, that Amazon's going to start collecting its boxes from the front porch, and the other material that they use to package stuff in the future—is that, can we get that circular?

Rebecca Hu-Thrams 26:55
Yeah. I mean, I'd say, you know, never say never, certainly. But one of the things that I've learned over my years of building Glacier, and one of the reasons I love the industry we work in, is because, unlike a lot of other sustainability or climate initiatives, I wouldn't necessarily call what we're doing extreme moonshot technology, right?

One of the things that's beautiful about recycling is in a landscape where we're all fighting the race against time to reverse the effects of climate change, or at least slow it down, the infrastructure for handling waste and recycling already exists. Is it, you know, perfectly effective? Maybe not. But I think that there's much more that we can do to piggyback off of the existing system.

And I would say that, you know, in a world where Amazon or another specific producer is responsible for handling their own waste, you ultimately end up with potentially an extremely fragmented waste experience, right? Because you're, as a household, you're buying dozens of different types of materials for dozens of different manufacturers.

Now what I have seen effective is essentially a cross-industry coalition approach, right? And so if you can get several producers or brands to sort of funnel their funds and their brain power into helping create a system in collaboration with a lot of the existing waste and recycling operators, I think that's where a lot of the magic really happens.

Mitch Ratcliffe 28:16
Well, they've always treated that as an externality, something someone else takes care of. Do you see this being integrated into their entire supply chain and product lifecycle planning?

Rebecca Hu-Thrams 28:27
Absolutely. One of the biggest points of optimism for me on the horizon is that even in the six years that we've been building Glacier, we've seen a lot of the attitudes change from viewing end of life of packaging as an externality to understanding that brands and producers can actually take this first-of-a-kind data on recycling from Glacier and turn it into insights that help drive how they design their packaging to be managed, and what they can do to educate consumers on how to dispose of that material.

I can give you an example of this. So when we are working with someone like Amazon on their bio-based packaging detection, one thing I loved about this project is that, unlike many of our projects where there's sort of a success metric you're trying to hit—right, maybe we want our AI model to be X percent accurate at detecting those materials—we and Amazon actually turned that on its head and said, We're going to test a variety of different types of packaging, and we want to treat the AI model training exercise as an experiment to show us what design features make it easier or harder for AI to detect this in the waste stream.

And so off the backs of those learnings, we've actually presented at a variety of industry conferences with Amazon about how other packaging producers can weave these design factors into their packaging to create stuff that is ultimately more detectable and therefore more recoverable.

Mitch Ratcliffe 29:58
When you're talking with a MRF operator now, and they're considering your robots, do they see this as a business extension, not simply a way of changing the economics of what they currently do? Are there new revenue opportunities in being able to output better-sorted and cleaner material?

Rebecca Hu-Thrams 30:13
Absolutely. There's many different ways that a MRF operator can benefit, you know, financially and operationally, from installing Glacier's equipment. You know, the robot itself offers direct benefits in the form of the ability to kind of supplement your existing and dwindling labor force, recover more materials so that you're making more money by selling those commodities, as well as decreasing how much you're paying to send your waste output to landfill, and also selling higher value commodities, because you're producing higher value goods.

The other pillar that is new and extremely exciting is helping MRF operators understand what they can be doing with the data that we're returning to them about their operations. So I'll give you one example. We actually have worked for a long time with a MRF in Detroit, in Michigan, and we showed them using one camera on their residue line—that's the line that's taking all of the alleged trash material out of the facility, it's already been picked over—we actually showed them that they were losing a massive amount of PET bottles on that residue stream.

And this was something that they had maybe intuited but hadn't really quantified before, because it's incredibly difficult to just stare at a running conveyor line and be able to understand what you're losing out the back end. From that data, this MRF was actually able to quickly pencil out that it was certainly going to pay for itself to add another sorter just to collect PET bottles upstream.

And what's really neat is that as soon as they put that person on the line, the data immediately showed a two-thirds drop in the amount of PET being sent to landfill. And this equated to, of course, not only an improved recovery rate, but also a direct $138,000 in annual revenue for that MRF.

Mitch Ratcliffe 31:59
This points to a completely different economic model for recycling. Currently, a lot of municipalities treated as a cost center, right? We collect it, we put it somewhere, and particularly because landfills are licensed to operate for a limited period of time, you have to fill them and then monitor them for another 30 years afterwards, we're oriented around filling those landfills. Should we be separating the MRF process from the landfill explicitly, and saying it's a failure when that material goes to the landfill?

Rebecca Hu-Thrams 32:31
Yeah, I mean, I think in a lot of ways, when Glacier thinks about turning the unit economics of recycling on its head, it is in the landscape of, how do we think about all of these different business lines that waste companies and recycling companies are considering? And so I think it would be a huge win for, you know, this industry and society, if we said, hey, you can extend the life of your landfills pretty significantly by sending less stuff in that direction, while also making more money off of the commodities that you're baling and able to sell. This is something that Glacier definitely hopes to enable for the industry.

Mitch Ratcliffe 33:04
So when you consider that potential future, what do you think the biggest misconceptions about the MRFs' operations that we hold as articles of faith today need to be revisited?

Rebecca Hu-Thrams 33:19
As a layperson consumer, I think we oftentimes have this wishful image of recycling as this perfect kind of Willy Wonka factory that takes anything you throw into your bin and figures out what to do with it. And in reality, this behavior of, you know, what I've learned since is called wish-cycling, is incredibly difficult for MRFs to deal with.

You know, MRFs today are put in a really difficult position in which they are more or less entirely reactive. I was speaking with a customer recently who summed it up when he said, you know, on any given day I show up at work, I turn on the lights to my facility, and I hold my breath and wait to see what new, crazy things I have to deal with coming across our conveyor lines.

What I hope is true for recycling in the coming years, and what Glacier's technology intends to help enable, is to actually make it such that producers are making things that are designed to be really easy to be recycled and sorted well and high value when they're sorted to the right place. And if that's the case, then you can actually help recycling facility operators take control of what is in their recycling stream and build a more consistent and reliable business out of it, versus today, being kind of at the whims of whatever new packaging happens to hit the market, even though it may not necessarily have an end use case.

Mitch Ratcliffe 34:42
So we can be very intentional about tracking materials through the entire process, from production through end of use. And we've had GS1 on and they have a new barcode that could do it at the unit level. So we know it's not just a milk carton, it's that specific milk carton, and it was sold here, and it ended up in this MRF here. We can rethink our entire economy with that kind of line of sight into what's going on with materials.

How do you think, say, 10 years from now, success will change the shape of the recycling system and Glacier's business, ultimately, the American relationship with stuff, all the materials they consume. What's your vision?

Rebecca Hu-Thrams 35:27
Glacier's vision is to make the circular economy a reality by helping to transform recycling into advanced manufacturing. And to do so, we believe, requires kind of three things to be true.

The first is that you actually need a really reliable and accurate data layer to help trace what's happening in the recycling stream. I believe that a lot of the stagnation in circularity is because it's really hard to point to a single source of truth, because a MRF may do one bale audit, a producer may do another bale audit, and suddenly everyone's saying different things about what's actually happening in the waste stream. So our AI is meant to help unlock those insights consistently and at scale for everyone in the circular economy.

The second pillar of advanced manufacturing, of course, is reliable automation, and so the ability to know that you can consistently hit your spec, your diversion rates, your uptime, your yield with technology that is reliable and cost effective, and that's where our robots come in.

And then finally, it all goes back to the quality of that feedstock that you're working with as a manufacturer. So our technology is designed to help enable the waste stream to be a cleaner output and something that is designed to be easily recovered, and this involves a lot of our work with brands and producers.

We believe that if we can get these three pillars to ladder up and perhaps aided by some of the tailwinds of legislation that are coming our way, we can essentially transform the conception of recycling as a cost center or a nice-to-have activity into essentially a foundational manufacturing exercise, and one that is built using the cutting-edge technologies that we assume to be par for the course for all other forms of manufacturing.

Mitch Ratcliffe 37:19
You're describing a 50 or 100 year process. Where are we in terms of percentage towards that goal today, and what do you think the most important things to focus on for the next few years are?

Rebecca Hu-Thrams 37:33
Yeah, I'd say that, you know, the percentage of the way towards that vision—it's a little bit of a loaded question, because while I would say we're nowhere near close to the finish line, the other really inspiring thing is that the acceleration of technology and of consumer and, you know, stakeholder attitudes is also not linear, right? And so what we may end up with is the ability to close that gap in a much shorter time than the past 60 years or so of recycling gaining its footing in the country.

Mitch Ratcliffe 38:07
Are we further along than you thought you would be at this point?

Rebecca Hu-Thrams 38:12
In some respects, yes. In other respects, no. I think that one thing that gives me a lot of hope is even in the six fairly short years that we've been building Glacier, we've seen pretty close to a 180 on what this data can do for the industry.

I remember, you know, back when we were starting Glacier, we always have said we don't consider ourselves a robotics company. We consider ourselves a company that uses technology to find ways to help end waste. But when I spoke with MRF operator, and tried to explain to him, Hey, we can put this vision system on your line, and it'll help you learn in real time how your composition is looking, what your throughput looks like, how you're recovering materials. And he told me, point blank, I would never pay you to tell me what I already know about my waste.

And that moment, I was like, Okay, I'm not going to convince this person in the course of one conversation that he doesn't know everything about his waste already. But over the past few years, we have seen, I think, a huge groundswell of support for this type of data and this type of visibility as operators understand what it could really do to benefit their businesses. And so a lot of our expected impact on recycling is not just our ability to put out reliable, impactful technology, but it really needs to be accompanied by supportive attitudes and embrace of this innovation from the folks that need to use it.

Mitch Ratcliffe 39:36
There's a lot more for this story coming in the future. I'd love to keep in touch. But how can folks track what Glacier is doing along the way?

Rebecca Hu-Thrams 39:44
The best way to keep in touch with us is certainly to follow us on LinkedIn, so you can follow Glacier directly, or you can also follow me at Rebecca Hu-Thrams. That's H-U, hyphen, T-H-R-A-M-S. And then if you go to our website, it's endwaste.io, and you can also sign up for our mailing list there.

Mitch Ratcliffe 40:01
Rebecca, thanks so much for your time today. It's been a pleasure.

Rebecca Hu-Thrams 40:05
Thank you, Mitch.

[COMMERCIAL BREAK]

Mitch Ratcliffe 40:12
Welcome back to Sustainability In Your Ear. You've been listening to my conversation with Rebecca Hu-Thrams. She's co-founder and CEO of Glacier, whose AI-powered robotic sorters are already processing recycling for one in 10 Americans at material recovery facilities in San Francisco, Seattle, Chicago and other cities around the country. You can learn more about Glacier at endwaste.io—endwaste is all one word—endwaste.io.

Rebecca's vision that is to make the circular economy a reality by helping to transform recycling into advanced manufacturing, goes beyond papering over the issues with today's recycling, to tackle how we can reframe what recycling can become.

Right now, as she pointed out, MRF managers show up to work, turn on the lights and hold their breath and wait to see what new, crazy things come down their conveyor lines. That's reactive. They're at the mercy of whatever packaging designs happen to hit the market, regardless of whether those materials have any viable end use. And that, my friends, is the description of an almost 19th century industry in need of an upgrade.

What Glacier is building—these robots and the systems that run them—will help change the recycling equation. No longer will it be a collection and processing business in search of end markets. It'll be a planned input to the next generation of products and packaging that we use.

Rebecca described three pillars: a reliable data layer to trace what's actually happening in the waste stream, automation that consistently sorts materials into higher value streams to produce higher quality feedstocks, and a robust supply chain to deliver the feedstock to end users in the form of new products.

And when I asked about the $200 billion in materials that we lose each year, she cut right to the economics, saying, if you can find a way to generate value by recycling something and turning it into a new thing, rather than just tossing it into a landfill for the rest of time, you've done something fundamentally good, not only for the planet, but also for many people's wallets.

So how much are we talking about? If we can capture that $200 billion worth of used materials for reuse, the United States could generate up to $2 trillion in new economic value. Even now, EPA data shows that every 1,000 tons of recyclables processed generates 1.7 jobs, $65,000 in wages and over $9,000 in tax revenue. And that's just the direct effects, even in today's relatively antiquated recycling system. And folks that work in the recycling industry, I'm not knocking you, I'm pointing out that we have a lot of opportunity for progress.

Even in that environment, a single recycling job supports eight or nine additional jobs in the supply chain, while reducing waste and the need to extract raw materials from the planet. World Economic Forum research projects that the circular economy could unlock $4.5 trillion in additional economic output globally by 2030. We're not talking about the value of a recycled aluminum can here. We're talking about the entire manufacturing ecosystem that that can feeds once it's recovered.

So Rebecca's example from the Detroit MRF makes this tangible. One AI camera on a resin line identified the MRF was missing the chance to recycle massive numbers of PET bottles. Material that they intuited was slipping through, but had never been quantified. By adding just one sorter based on that data, they produced a two-thirds drop in the volume of PET sent to landfills and earned $138,000 in additional revenue. That's what happens when visibility meets action—and in just one of hundreds of waste streams that flow through the nation's MRFs.

Perhaps most compelling is where this leads with brands. Glacier's work with Amazon to discover what design features make it easier or harder for AI to detect valuable discarded packaging points to an economy of designed-for-recycling products, packaging and shipping materials. And with that feedback loop from end of life back to design, we move from technically recyclable to provably recyclable.

As extended producer responsibility laws spread—and Rebecca notes that they've improved recycling rates by over 40 percentage points in parts of Europe—this kind of brand-level accountability will become table stakes. You'll actually be able to expect a company to take their stuff back and make sure it gets used again, and that is an important change.

So we're in the early innings of a game that can rival the promises AI is making in terms of economic value while reducing our environmental impact. The gap between where we are and a truly circular economy may close faster than past 60 years of recycling's slow build would suggest. Glacier's robots, along with other components of an intelligence layer that makes the recycling system legible, manageable and ultimately profitable, are the scaffolding for a profitable future in which what we call waste will become value.

We'll know that this circular economy is here when failing to keep something out of a landfill is recognized as a loss, not just sanitary disposal. So stay tuned, folks. We'll keep track of the innovators turning waste into wealth.

And I hope you take a moment to check out any of the more than 500 episodes of Sustainability In Your Ear that we have produced in the archive. A rating and review on your favorite podcast platform, and sharing one with your friends will help your neighbors find us. Folks, you're the amplifiers that can spread more ideas to create less waste. So please tell your friends, family and co-workers that they can find Sustainability In Your Ear on Apple Podcasts, Spotify, iHeartRadio, Audible, or whatever purveyor of podcast goodness they prefer.

Thank you for your support. I'm Mitch Ratcliffe. This is Sustainability In Your Ear, and we will be back with another innovator interview soon. In the meantime, folks, take care of yourself, take care of one another, and let's all take care of this beautiful planet of ours. Have a Green Day.

You may need to click through the site to see the video.

Walking along Elk Creek above the old dam site in late Autumn, I found these leaves trapped against a rock under the flowing stream.