Transforming Waste: AI-Powered Recycling is Here

Welcome to the Ctrl+Listen Podcast, brought to you by Octopart! Join James and his guest, Carling Spelhaug, the Director of Communications at AMP, as they delve into the revolutionary impact of AI on the recycling industry.

Listen to the Episode:

Watch the Episode:

Episode Highlights:

• AMP's Mission: How AMP is maximizing waste value
• The role of AI in automating waste sorting and improving efficiency.
• Materials Recovery Facilities (MRFs)
• The importance of MRFs in the recycling process
• Circular Economy

Links:

Learn more about AMP here

Connect with Carling here

Transcript:

James Sweetlove: Hi, everyone, this is James, from the Ctrl+Listen Podcast, brought to you by Octopart. I'm here today with my guest, Carling Spelhaug. Welcome to the show, thank you so much for coming on.

Carling Spelhaug: Thanks, James. Appreciate you having me.

James: Anytime. And you're the Director of Communications for AMP Sortation?

Carling: Yes, actually, we just go by AMP these days. We were AMP Robotics until earlier this year, we dropped Robotics from our name, just to reflect kind of our increased focus on delivering what we're calling facility-scale solutions for the waste and recycling industry.

James: Okay. Oh, that makes total sense. You've expanded, so the name needs a broader scope. So just to get started -

Carling:  More than robots, you think of robots. Oh, I'm sorry.

James: Oh, sorry, there's a delay there. I apologize for that, go ahead.

Carling:  It's okay. No, I was gonna say, you know, I think we are best-known, given being in business for about a decade for our pick-and-place robots. But now that we're doing kind of full-scale facility systems, we think of these facilities as, you know, bigger, more comprehensive robots as well. But the name change, like I said, yeah, just reflects kind of that broader focus.

James: Okay, fascinating. So, I wanted to ask about the company. We'll start sort of from the beginning. What exactly is AMP's mission and story? What does the company do?

Carling: Yeah, so here at AMP, our mission is to maximize the value in waste, and transform it into an asset. We were founded about a decade ago by Matanya Horowitz, who remains our CEO. Matanya had always been interested in robotics and the origins of intelligence, and while studying for his PhD, he saw some of the major results in the subject that's now known as deep learning. You know, a series of breakthroughs led to machine learning, machines started to see, for the first time, roughly as well as a human, so, after graduating, he wanted to find places where the technology could really be most useful, and found that in the world of recycling. It seemed like the recycling industry really had the biggest need for this technology. When Matanya visited these facilities, he talked to people about the challenges they were facing, you know, issues like the quality of the materials being separated out during the recycling process, high rates of staff turnover, things like that. And so the convergence of robotics and machine learning offered really compelling opportunities to automate what had historically been tasks and processes that were high cost, labor-intensive, inconsistent, and really limiting. And so, at AMP, we installed our first robot in 2016 here in Denver, near where we're headquartered in Louisville. And fast forward to today, we have more than 400 systems installed globally, and three full-scale facilities that are powered by our AI technology.

James: Wow, that's some impressive expansion. And looking here, so what exactly is an MRF, and why are they so important?

Carling:  Yeah, this is an important acronym, I think, to clear up early in the conversation. So, an MRF, or a merf, is a materials recovery facility, or a centralized location, or a hub, where recyclables that are collected get processed. So, mixed material comes in, sorting happens, and then commodities and residuals come out. You know, here in the United States, there are thousands of municipal recycling programs that are collecting material generated by millions of people. And so between the collection and the actual recycling of the commodities, there's really only a small number of these MRFs, or recycling facilities, to cope with the tremendous volume of material. And a major challenge is that MRFs incur high costs when they're attempting to sort out the commodities, and these commodities are constantly evolving, you know, with complex form factors, packaging changes, material compositions. And so these legacy MRFs require a really tremendous amount of manual labor, and may incur high residue loss and continuous challenges to ensure bales that are of high quality. And this all amounts to, then, an erosion in the value of recyclables due to this high cost of sorting. So we're trying to change the economics of recycling, and, you know, bring that cost of sorting down. You know, we need MRFs to make recycling work, but we also need moderns MRFs, and more of them, if we want it to work better.

James: That makes sense. So, for someone like myself, who knows very little about the recycling side of things and the sorting side of things, roughly how many categories are there to sort into, as far as recycling goes?

Carling: Our AI can categorize more than, I think, about 50 categories of material. You know, most of the material will fall into your primary paper, plastic, metals, and kind of the big categories you may be familiar with. But the platform's also capability of, capable, excuse me, of tracing materials down to the brand owner. So, it basically learns to recognize, you know, any visual property that a human would be able to identify. So, you know, it looks at things like a Coke bottle, you know, it's an empty plastic bottle, it's clear, it has a consistent shape that looks like a container we're all familiar with. And while it doesn't actually read labels, you know, it comes to associate things like red, and maybe a type of label or script on the packaging, as, you know, a property that it can observe, and recognize with pattern recognition. And so, you know, AI sees packaging at all angles, including all levels of dirt, grime, and damage, and it can pick an object even if it can't see the entire thing. And so the more of these identification parameters you have, the more identification accuracy you get. You know, we often get questions like, "What does the AI have a hard time identifying?" And it would be anything that a human would have trouble identifying, just because it is all based on computer vision, or seeing the visually observable properties of an object.

James: Interesting. And then there's a term that gets thrown around a lot lately, I think some people don't know what it means, people like to just say it, but what exactly is a circular economy, and how does recycling play into that?

Carling: Yeah, I think the best way to think about a circular economy is just kind moving from a linear way of using and managing resources, you know, take, make, consume, dispose, to kind of closing the loop, or, you know, using things over and over again at their highest possible use. So, you know, where recycling and AI comes in, our AI technology provides this low cost sensing platform designed to recover value from waste streams. We're trying to create a world without waste by making waste itself more valuable. By reducing the cost of sorting material, our technology is helping to grow the market of recycled materials by making it profitable to recover material, where maybe it currently isn't. And then by providing more control over material chemistry, or being able to recognize more of those visual attributes, and sort accordingly, kind of like the way I just mentioned, we're helping to create the opportunity to sort new combinations, unlocking higher revenue from the same material. So then this, in turn, enables businesses to convert those chemistries back into useful products to generate more value. So, all these levers create an economic incentive to reuse waste, which, you know, leads to things like less land filling, lower pollution, and reduced emissions from lower carbon material production. So, altogether, these dynamics can kinda unlock the total value in waste to help, you know, transfer or transform it from a hazardous byproduct of society into something that's a really a valuable local resource.

James: That's awesome. So, back to MRFs that we were discussing before. You actually offer quite a wide range, I should say suite, of MRFs. Do you wanna just run us through some of the the ones available?

Carling: Sure, so, here at AMP, we provide AI-powered sortation, and we offer it really in two models. So, there's the facility-scale solution called AMP ONE, and then upgrades to existing materials recovery facilities, or MRFs. So, as I mentioned, you know, AI provides this universal sensing platform for recovering value from waste streams, and it's the common engine that powers all of our solutions. So, we offer multiple versions of pick-and-place robots that perform upwards of 80 picks per minute on 300 feet per meter belts. You know, that's compared with a good manual sorter, can probably do 4 to 50 picks per minute, and, you know, not consistently. We also offer jet systems that perform thousands of picks per minute on 600 feet per meter belts, that's more of our version of an optical sorter for those that may be familiar with other types of equipment used in the industry. And then full-scale smart facility solutions that are capable of processing anywhere from, you know, 10,000 to over 100,000 tons of material per year.

James: Wow.

Carling: So, combined, our technology suite can tackle the majority of non-automated sorting stations in a MRF to make the business steadier and more reliable, and they all drop in without a significant retrofit to existing infrastructure. And then we've also integrated, like I said, these technologies into this facility solution that is, you know, like I mentioned at the outset, kind of our version of like one big system-wide robot that's helping lower processing costs, enable more profitable sorting, and ultimately expanding recycling access and infrastructure.

James: You kind of bled into the next topic we're gonna discuss, which is scalability. So, it seems like this is a pretty scalable product, you can go right from the small scale local to large, full facility size. Is there any plan in the future to sort of branch out from just recycling and then apply this technology to other areas, the sorting technology?

Carling: Yeah, so I'll start with, kinda your question on scalability, it's definitely one of the key benefits that we tout. Each of our products, you know, uses the same core vision system to sense and sort, so that makes the solutions really scalable and highly modular. And with, you know, these flexible configurations meaning customers have the ability to easily add volume or additional commodities, kind of depending on what their needs are. One important point in this regard is that our technology works with existing waste assets. So, due to the small footprint of most of our solutions, we can leverage existing infrastructure, like transfer stations, where sortation solutions likely don't currently exist and material would be bound for landfill. And then kind of bridging back to what you asked for, or asked about at the end of your question, you know, most of our solutions are processing single stream recycling, but we have recently branched out into municipal solid waste. So, we have a pilot project in Portsmouth, Virginia, where we are using our technology to extract the recyclables and the organic material from bagged municipal solid waste. And so, you know, that is material that absolutely would've been going to landfill without this kind of technology. And so, that's been a really successful project, and one that we're looking forward to expanding to more locations to, you know, increase landfill diversion wherever we can.

James: That's fantastic. I mean, this is one of those things where recycling has been around as a process for so long, but I feel like only now are we really getting to a point where we have technology to actually make a huge impact on it, because it's been so manual up to this point.

Carling: Absolutely. And I mean, I've worked for AMP for four and a half years, and, you know, I think I kind of came into the role really just with an idea that recycling is just this, you know, environmental good, something that we do because we care about the planet. But it really is a business, and it operates with many of the same dynamics. And commodity prices, you know, can be very volatile and that plays a huge role in kind of what makes these businesses successful or not. But, you know, as I mentioned, the goal with the technology that we are bringing to the industry is to try to create that level of consistency and reliability so that, you know, some of these other factors don't play as big of a role in determining whether recycling works in a municipality or community or not.

James: Right, right. And as you said before, recycling is a business, it's a massive international business. I think that a lot of people probably don't realize how much recycling from developed nations is actually exported to other countries to be processed and sorted.

Carling: Yeah so, what you're referring to there is, you know, China had a National Sword policy that I believe went into effect in 2018, and that really raised the bar on, you know, what was accepted in terms of imports for recyclables. So, instead of outsourcing this problem to other countries, where they may not have the infrastructure to deal with it, you know, that really forced us to take a look at, how do we deal with this material and manage it more sustainably, you know, within our own borders. And as I mentioned, AMP started about a decade ago, so we were kind of getting started right as this was starting to, you know, the industry was kind of coming to reckoning with it. So, you know, I think we're playing a role with our technology, of increasing quality through domestic sorting, and, you know, just making recycling work a little bit better here in the U.S., and other developed countries.

James: Right. Yeah, definitely. So, looking back at this, we've got... Data is what I wanted to ask about.

Carling: Yeah.

James: So, data wise, you don't just sort, you actually collate a lot of data from the sortation, so people can actually see what's happening, you can form patterns. How is this data analytics processed?

Carling: Yeah so, our vision-based AI software, you know, like I said, identifies and characterizes each object in the waste stream in real time. So, it digitizes each item that passes by. You know, these objects are really being captured as a new form of data, so it's object counts, it's packaging descriptions, and more over time. You know, we've deployed hundreds of robots and sensors that are processing billions upon billions of objects on conveyor belts in recycling facilities. And this effectively enables automated, continuous characterization of the material. You know, as more systems get deployed, the industry is able to leverage the networked intelligence of every system. So the more of these AI-based sensors that we deploy into production, there's more of a network effect that's created that increases the collective sorting intelligence. So, if a challenging packaging type, or a new material, emerges, we can, you know, take pictures, or capture imagery, and train the AI to identify that object, and then deploy that knowledge across all systems. You know, whether a facility is currently sorting that material or not. So, kind of with this context, at the user level, operators can use the data that this machine learning technology produces for analysis to support inbound, outbound, and product quality questions. So with data and tangible metrics, operators can get ahead of, you know, mechanical or configuration-based issues, and communicate with business partners, or staff in their facility. Use cases like, you know, how much contamination is entering the facility, and from what sources? How much valuable material am I losing at the point of residue? Could I drive a higher price for bales with data that shows the quantity and quality of, you know, what's in the bale? And then when you build AI capabilities into a facility from the outset, like we're focused on now with our facility-scale solutions, you can really optimize its design around the technology. So, a retrofit environment, like where we're putting, you know, a robot into an existing environment is limited by, you know, existing hardware, space constraints, things like the speed or the width of conveyor belts. You know, but this kind of AI from the ground up approach supports, really, order of magnitudes changes in how these next generation facilities can perform. So, it's not just that you need fewer sorters, you could operate without manual sorting altogether, and add shifts more easily, given the automation and reduction in variable costs. And the deployment of vision systems throughout a facility provides operators with the ability to monitor every line continuously and in real time. So, when you design a facility with AI-enabled data collection in mind, like we're doing now, you can make the lines wider, you can make them higher speed than they otherwise would be, to, you know, get the best spread on the material, and thus the best data. So, in this way, AI can really help replace entire classes of equipment that create maintenance and downtime events. And, you know, tying back to what we talked about earlier, that's really why we're able to operate with smaller footprints, and install these solutions a little bit more rapidly.

James: Do you see this sort of AI integration as being the new standard going forward with factories, sorting facilities, and logistics hubs?

Carling: Yeah, I think so. And you know, obviously AI is getting a lot of attention lately with some of the breakthroughs that are going on in the generative AI space. But you know, to be clear, our technology is different, it's a different form of AI. And I feel like we've been talking about it kind of before it was a buzzword. So, I think we're kind of over the hurdle where, you know, the technology is intimidating, or, you know, too new to be reliable for recyclers. So, I think we're now at the point where we're trying to just get the technology to take hold a little bit more broadly, like I said, as opposed to maybe a couple of sorting devices, like how do you kind of make your entire facility operate like a smart factory, and, you know, reap the benefits of all of this rich data that these systems are generating?

James: Is there any sort of friction between, say you have a facility that's semi-automated and semi-traditional, is there any sort of friction that space, or is it operated quite clearly and smoothly between the two?

Carling:  That's a good question. I guess I don't have any like specific anecdotes, but I do think, like I said, you know, I guess maybe one thing to talk about here would be kind of the broader conversation when you think about like AI taking jobs, or, you know, disrupting operations that maybe were smooth prior to the onset of this technology. But we touched on earlier just how challenging that labor consistency is in the recycling industry. You know, that was before COVID, you know, got more challenging with COVID. You know, recycling is really, manual sorting and recycling, I should say, is, you know, really like a dull, dangerous job that automation is ripe for. And what we found, and to answer your question, where I don't think there is as much friction is, you know, there's always gonna be a need for humans, and I think this technology does help to create higher skilled jobs within these facilities. You know, maintenance technicians, you know, people that are servicing or helping to maintain the technology once it's installed. And then, of course, you're always gonna need things like production operators, plant managers, forklift operators, truck drivers. So, the role of humans is not gonna go away, but we want want them to work with this technology, learn from it, and ultimately, again, make recycling work better.

James: Right, and I assume you need data analysts as well to actually monitor those results and implement them.

Carling:  Well, yeah. You know, I think the data that is being used, like by operators or others in the facility, is presented in a fairly intuitive way where it can be learned and understood pretty easily, you know, by people with a bit of training. But, you know, certainly here at AMP, we have quite a few data team members that are helping to oversee the annotation and the ongoing learning of our AI platform. You know, driving the development of new categories for AI, and that kind of thing. But I think most of those, like, you know, AI data science type roles, are more here at AMP, and less maybe on the customer side, or the end user side.

James: Well, that's great. That takes the pressure off the actual person doing, the actual owner of the machinery who's purchased it, and you had that whole network of operators.

Carling: Yeah, it hopefully makes it, again, an easier learning curve, and something that's not too intimidating to adopt.

James: Definitely. So I kind of wanted to ask something about, back to adoption, what is the sort of timeframe for fitting this equipment, or installing this equipment in a facility?

Carling:  Sure. So, most of our retrofit solutions can be installed over the course of a weekend. So, you know, two or three days without interruption to existing operations. And then when it comes to, you know, our AMP ONE facility scale solutions, they can be installed relatively quickly as well, usually with about nine months of lead time for custom built plants. And you know, we're not necessarily going in and building the buildings, of course, but we're kind of handling everything that's like, within the walls. So we'll go in and, you know, set up the whole system, maybe within a existing facility or something, that maybe was being used for something else other than recycling before.

James: Wow. And how many facilities, like custom facilities, have you actually got operating right now?

Carling: We have three operating right now. Like I mentioned, we have Portsmouth, Virginia, which is a facility that's owned by RDS of Virginia, as is a facility in Pitt County, North Carolina, also operated by RDS. And then we have our own facility, which is a secondary sortation facility. Secondary sortation means that we are sorting material that has already been run through a primary recycling facility, and, you know, they've sorted what they've wanted, or think they can sell out of that material, and then we're able to process it again, and extract even more value, you know, before the residue goes to landfill. And that facility is outside the Cleveland, Ohio area. I don't know if I actually mentioned that.

James: So, talking bigger picture, at the moment, there's a big focus on clean energy, the green economy, all that sort of thing. How important is recycling in that big picture fight against those issues?

Carling: I would say it's quite important, it plays a pretty key role. You know, I spoke a bit earlier about how reducing the cost of sorting helps to grow the market of recycled materials, and how AI allows for greater specificity and control over material chemistry in sorting. You know, these levers create an incentive, an economic incentive, to reuse waste materials. Again, less landfill, lower pollution, reduced emissions, and, you know, making waste more valuable. You know, I mentioned that we recently made inroads into sorting bagged municipal solid waste, taking the recyclables and the organic material out of the garbage that was headed straight for a landfill. You know, very few innovations can drive such a significant impact to greenhouse gas emissions, and fewer still can progress as quickly within the next decade. And I think we can all agree, we don't have a lot of time to wait or to waste. And so this is really the potential of recycling. If we unlock the value in waste, the world gains this economic imperative to capture this material and reduce the environmental impact. The billions of dollars in waste can create a valuable business. But, you know, just as importantly, it gives the world a really strong incentive to quickly adapt to a new, really more sustainable way, of managing our resources.

James: Right. And AI in that, in this context, how important is AI in this fight against these issues?

Carling:  Well, I think we're seeing, you know, AI is this enabling technology that's allowing us to do kind of all of these different things in the recycling industry that weren't possible before, so I'd say it plays a pretty fundamental role.

James: I definitely agree. I think it unlocks a lot of potential, and removes a lot of limitations. I think you can do a lot more with a lot less if you have that automation operating effectively.

Carling: That's absolutely right.

James: So, is there anything exciting or interesting coming up for the company that we should be keeping an eye out for?

Carling:  Yeah, absolutely. We have a couple of projects on the horizon that we should be able to share more about soon. But like I said, this kind of foray into sorting municipal solid waste we see as really a game changer for the industry, and something that we think is gonna help to expand our impact quite rapidly. So, if you'll be able to join again in the future and share a little bit more.

James: Yeah, that'd be fantastic. And then I just, before we wrap up, if anyone wants to keep up with the company, look into your offerings, that sort of thing, is your website the best place to do so?

Carling: Yep, you can learn more about AMP at ampsortation.com. Follow us on LinkedIn, Instagram, YouTube, and X.

James: Fantastic. Well, thank you so much for coming on the show, and it's been fascinating talking about your technology, it's really innovative and super groundbreaking stuff.

Carling: Thank you so much for having me, we appreciate it.

James: Anytime. And for everyone listening and watching, come back next week, we'll have another guest.