By Carol Brzozowski
For original article, click here.
“Looking at the current world situation, the amount of waste production is increasing with population growth,” notes Tanner Cook, CleanRobotics cofounder and senior vice president of engineering.
“Regular methods of waste collection are inefficient and costly. Smart waste management using AI is a more efficient and cost-effective way to manage trash, help with recycling, and reduce the negative impact of waste on the environment.”
Indeed, artificial intelligence (AI) is becoming more commonplace at solid waste facilities to achieve those goals. Options abound in doing so.
Glacier is building what is designed to be the highest-ROI, smallest- footprint robotic sorter and AI technology for MRFs, with a typical payback period of less than two years, notes Rebecca Hu, founder.
Glacier’s team comes from cutting-edge engineering and data science programs at Google, Facebook, MIT, and Stanford.
With Glacier’s AI technology, MRFs can detect real-time trends and aberrations in the material stream to make better business and operational decisions, says Hu.
Glacier’s AI data can help answer questions such as how much value is being lost to the landfill each minute; from where the leakage is coming; which lines need more staffing or equipment upgrades, and what inbound routes have the most contaminated loads.
Other questions it addresses include how does the month, day of week, or hour of day impact the composition of material being received; how can the facility react effectively to those fluctuations; the purity rate of each of the bales, and what bales need to be pulled for further quality control.
“Our robot supplements our AI technology by offering reliable, high-ROI sorting power on most material streams,” says Hu.
AMP Robotics provides a portfolio of recycling solutions powered by its neural network.
“AMP Cortex is our high-speed robotic sorting system guided by our AI technology,” notes Amanda Marrs, AMP’s senior director of product. “Our robots intelligently perform physical tasks of sorting, picking, and placing material to achieve up to 99 percent accuracy and 80 to 120 picks per minute.”
AMP Vision is a modular computer vision system designed to drop into key stages of recycling operations to better understand material flow from inbound processing to bale quality control to the end of the line.
“AMP Clarity is our material-characterization and robot-performance software solution that allows users to monitor real-time material composition and performance measurement throughout a facility,” says Marrs, adding recent Clarity features include mass estimation, robot pick assignments, alerts, status tracking, and expanded reporting capabilities.
AMP Vortex is an AI automation system for the recovery of film and flexible packaging designed for MRFs to “tackle the persistent challenge of film contamination and ultimately to target and recover film and flexible packaging for baling and selling,” says Marrs.
AMP technology has proven its effectiveness in installations such as Emmet County in the northern tip of Michigan’s Lower Peninsula.
The county’s Department of Public Works operates recycling, resource recovery, and solid waste transfer services—its MRF processing and marketing about 8,000 tons of material for recycling.
Emmet County had staffing challenges against the backdrop of seasonal issues in which tourists to the area led to a solid waste increase, forcing the facility to send material to a different MRF in summer 2020due to a processing backlog.
As part of the solution, managers installed three AMP Cortex highspeed intelligent robotics systems to sort PET, HDPE, cartons, aluminum, and mixed plastics on its container lines.
Among the many benefits, robots boosted labor efficiency by 60 percent and capture of recyclables by 11percent.
At a pace close to 90 picks per minute, robots are accomplishing in three hours what would normally take human sorters a full eight-hour day. The operation is not only saving $15,000 by not having to send materials to another MRF, but is accepting material from a nearby MRF that cannot process everything.
During COVID, with robots, the county was able to continue processing as normal with half of its previous staffing and now has brought on all of its long-term temps as full-time employees with benefits.
“It’s amazing how well and efficiently the surge hopper works. The more we run it, the less time we’re actually spending with it. It fills up, our people walk away, and we let the robots do the work. Whether it be minutes or seconds, we empty it sooner every week. There are more picks per minute, and it seems like it’s only getting faster,” notes Joshua Brubacher, Emmet County Operations Manager.
Everest Labs, provider of RecycleOS, uses deep learning AI combined with robotics to make recycling economically viable by drastically increasing recovery at a MRF.
“Through close collaboration with recyclers, RecycleOS was built to be the most accurate and scalable AI system in real-world facilities, iterating on computer vision models and robotics designs that beat industry benchmarks,” says Jagadeesh “JD” Ambati, CEO of Everest Labs, adding the systems can be installed anywhere in a recycling line without retrofits and can guarantee successful picks over its life time for more than 100 classes and subclasses of materials.
RecycleOS AI models are designed to accurately identify nearly indistinguishable objects and guarantee successful picks at rates twice that of manual sorters, Ambati notes.
“MRFs realize profits within months of installing RecycleOS without huge upfront costs of retrofits,” he says. “RecycleOS also provides benefits of 24/7 monitoring of robot performance, minute-by-minute visibility of plant operations, and precise quantification of what goes in and out of a facility with the ability to aggregate sustainability data across plants, packaging brands and regions.”
RecycleOS not only provides MRFs a solution to run effectively and realize higher revenue from recovered materials, but it also benefits the global ecosystem, notes Ambati.
“Preventing materials from being landfilled saves energy used in— and GHG emissions from—mining and production of virgin materials,” he adds. “RecycleOS tracks all materials entering and exiting recycling facilities to provide granular analytics based on EPA-approved methodology on the environmental impact
of a recycling facility. Data from RecycleOS also can be used to validate EPR [Extended Producer Responsibility] goals of CPG brands and provide much needed end of line data to improve packaging designs.”
Everest Labs recently entered into a partnership with Sims Municipal Recycling (SMR) Sunset Park Materials Recovery Facility in Brooklyn, New York—North America’s largest commingled recycling facility—to install up to eight RecycleOS-powered robotics cells.
New York City operates the largest curbside recycling program in the U.S., with SMR processing morethan 300,000 tons of glass, metal, and plastic produced by NYC and several other New York and NewJersey municipalities.
Tom Ferretti, SMR general manager, notes “These installations allow us and our partners to stay committed to sustainability while also saving us on costs, and enable moving SMR’s key personnel into higher-priority positions across the plant. It is a win for New York City recycling (as we are recovering more), for our team members, and plant safety and efficiency.”
TrashBot, by CleanRobotics, is a smart recycling bin that sorts waste with an accuracy of 95 percent, usingAI, robotics, computer vision, and machine learning. TrashBot also is designed to generate high-qualitydata for on-demand waste audits. It triggers fullness alerts and features a large display for custom videoand educational content.
TrashBot is designed for areas such as airports, hospitals, stadiums, and other high-traffic facilities with the goal of improving diversion, leading to affordable recycling and cost-effective environmental programs.
Clean Robotics cites U.S. Environmental Protection Agency statistics that sporting event attendees can generate up to 39 million pounds of trash per year, with most of it being containers for food and drink consumed in stadiums and parking lots.
Waste management and recycling in stadiums presents challenges in that contamination of recyclables reduces the ability to divert waste at stadiums; educating the transient public is difficult because of varying recycling rules, and granular waste data is unreliable, expensive, seasonal, and changes often.
In stadiums with transient populations, this leads to large-scale contamination, low recycling yields, and poor diversion.
TrashBots were piloted for one year at a Pittsburgh multi-purpose indoor arena with a capacity of more than 18,000 spectators. In the pilot program, 30,000 items were sorted with 90 percent accuracy compared to conventional sorting with 30 percent accuracy.
Some 1,800 pounds of recyclables were collected compared to 650 pounds with conventional bins.
MSS Optical Sorters offers a full range of NIR, color, metal, and AI-based sensor sorters using either air jets or robotic arms for extraction technologies. Various combinations of sensor and extraction configurations are available depending on the specific application and particle size.
Why this is important
Better data and data-capture technology provide opportunities for consumer-packaged goods companies, retailers, and packaging manufacturers to understand the quality, flow, and recovery of their specific containers and packaging, says Marrs of the importance of AI in recycling facilities.
“Our technology can help produce initiatives to increase recycling rates and create new value streams for recyclables, ultimately aiding their pursuit of recycled content goals,” she says. “As Extended Producer Responsibility (EPR) schemes emerge and mature, sensors growing in the fleet of MRFs can help satisfy the demand for reporting recovery rates.
“Data collection, measurement, and material characterization for recycling also create a mechanism to support federal, state, and local government programs focused on landfill diversion goals and recycled content standards to advance a more circular economy.”
Ambati concurs, noting not only does AI strengthen the circular economy, but it also ensures more valuable recyclables such as widely used metals like aluminum, that are facing deep shortages, can be recovered and reused.
It also plays a key role in helping reduce GHG emissions and decarbonizing the planet for a more sustainable future, Ambati adds.
MRFs are faced with increasing pressure to produce more and higher quality bales, while dealing with labor shortages and commodity price volatility, notes Hu, adding AI is a new tool to combat the mounting challenges of running MRFs and enable more data-driven decision making to ensure their 24/7 peak operation.
“Differentiating between food-contact and non-food-contact materials is important in order to preventhealth risks using recyclates that do not meet quality demands,” says Husam Taha, Deep LearningSolution Manager at TOMRA Recycling Sorting, adding that the U.S. Food and Drug Administration andthe European Food and Safety Authority provides clear guidance and regulations for the safetyassessment and the determination of the quality levels of recyclates.
“With the possibility to recover more material at higher purity levels, recycling rates can be increased, and materials previously thought hard or impossible to recycle be recycled,” he adds. “Thus, we save primary resources due to high quality recyclates on the market to meet demands—including legislation standards—and even off er new business opportunities in new material streams.”
On the paper and plastics packaging side, new materials, combination of materials and structures are coming to market at an increasingly faster rate, notes Felix Hottenstein, MSS Optical Sorters sales director.
“Conventional as well as AI-based optical sorting technologies need to be able to adapt quickly and without hardware upgrades,” he adds. “A combination of sensor technologies and extraction methods may be required to optimize certain sorting functions.”
AI can serve to augment human labor along the solid waste collection stream, addressing such issues as health and safety.
The AI technology is able to identify material stream trends that allow for smarter staffing decisions, says Hu.
“Everyone in the waste and recycling industry is working with a limited labor pool, so it’s important to staff workers where they can add the most unique value,” she says. “Furthermore, real-time tracking allows AI technology to flag issues to operators that may present a health or safety issue to workers such as detecting propane tanks or batteries on the line to reduce fi re hazard.
“By installing Glacier’s robot in lower-value, more arduous, or higher risk sorting positions, MRFs are able to staff their workers to locations where humans are uniquely capable of adding value.”
“Robotics are perfect for replacing dirty, dull, and dangerous labor,” says Cook. “Offl oading tasks to robots has already become commonplace in our industry with noticeably positive results.”
TrashBot, for instance, is designed to do the work of sorting waste at the source, eliminating the need for the painstaking process of human sorting, Cook adds.
“TrashBot also delivers on-demand waste audits which are otherwise done by humans manually going through the trash,” says Cook.
Automation in recycling drives consistency, as robots can work 24/7, Marrs points out.
“They don’t tire, nor do they need breaks,” she says. “Plus, they can work on faster-moving belts than humans. Their consistency also results in higher quality of recovered commodities.
“Robots are flexible. Our systems can be adjusted to reflect material stream changes, commodity prices, and more. Robots can work in areas and on materials where volumes don’t warrant a human, because they can multitask and target a variety of materials in lieu of just one or two.”
For many end-users, the addition of robots has allowed them to shift staff to higher-skilled positions withthe facility, such as roles in maintenance, as an equipment operator, or as a route driver, she adds.
AI and robotics technologies address labor challenges by helping MRFs reduce time and money spent on high-turnover manual sorting positions, says Marrs.
“Automation isn’t eliminating jobs, but helping facilities run fully staff ed and run additional shifts,” she says. “AI and robotics also help improve facility safety by reducing contact with harmful, hazardous materials in the stream and lowering training overhead.
“MRFs weren’t designed with social distancing in mind and robots create natural barriers between humans—a feature that took on greater importance in the early days of the COVID- 19 pandemic.”
Ambati notes that Everest Labs spent more than three years working with MRFs of different sizes to iterate on a design that works with the way MRFs are currently designed to vastly improve the operations.
“The design thinking behind our systems is that our robotics can go anywhere people can, but more importantly go in places where it’s too difficult for other robotic systems to be placed or where it’s too dangerous for people,” he adds.
“Most MRFs are a complex maze of conveyor belts, heavy equipment, walkways, and forklift paths. Installing a new piece of sorting equipment can take weeks of work and costly downtime due to the required retro fitting and reorganizing.”
The same is true for most other robotic systems, because they utilize large, belt-encompassing structures, Ambati says, adding “human sorters fi t in more places, but with moving equipment and belts overhead, safety is a limiting factor.”
Ambati notes that EverestLabs has recognized that if the objective is to pick bottles, papers, and boxes one-byone, “then our solution does not need a huge robot or piece of equipment.
“The robotic cell, end effector, suction and grabbers, and mountings were all designed with the end goal of easily placing a robotic system anywhere in the MRF. These robots can ultimately work in cells to coordinate on a task or can work individually to augment humans or other sorters.”
Additionally, vision and AI is modular and can go anywhere on the waste stream inside of a MRF, informing and optimizing processes, Ambati says, adding AI can produce data that can evaluate material input to the facility to predict the outcome for the facility, but also be used to optimize collection schemes.
Hand picking is not a preferred job at the plant, and it can pose safety challenges, notes Taha.
“The use of deep learning technology will not entirely replace manual sorting. It does create other types of jobs to make use of this technology.”
“The newest overall facility designs by CP Group, as well as the latest generation of MSS optical sorters,off ers a chance to completely automate the container and paper lines, with some manual sorting laboronly required for oversized items. This approach increases the health and safety rating of a recyclingfacility by default.”
Return on Investment
An AI investment leads to an ROI in multiple areas for MRFs, notes Ambati.
“Operationally, it is more cost-effective to run robots in areas where manual sorting is difficult or impossible due to labor shortages, difficult working conditions, and impractical locations,” he adds. “Robots can also do this job more accurately and efficiently around the clock.
“Additionally, AI-driven automation prevents safety and line down situations which are extremely costly. If small, dangerous items are identified early on, it can prevent expensive damage.”
Avanti notes that on the revenue side, AI and robotics can improve the volume of valuable recyclables that would otherwise have been landfilled, as well as the quality of the yield, which fetches higher prices.
Case in point: in one month, EverestLabs’ robots deployed in end users’ facilities successfully picked more than 4.2 million objects.
“Of those, more than 1,400,000 were recyclable objects of value sorted by only three robotic cells, totaling around $106,000,” Ambati says.
AI data yields strong ROI in many ways, including commodity revenues, reducing landfill fees, or strategically investing in better staffing and equipment upgrades to increase throughput where it’s most needed, notes Hu.
“As an example, losing even a few dozen containers to landfill per minute is very costly,” she notes. “Over the course of a year, this leakage adds up to nearly $1 million in lost commodity revenue and almost$100,000 in additional landfill fees. If AI data can help a MRF owner make decisions to recoup even a small fraction of that cost, the AI data pays for itself in only a few months.”
For AMP robots, while it depends on the facility and how many shifts it runs, costs can drop by as much as 50 to 70 percent by replacing hard to- secure human labor in various sorting roles, notes Marrs, adding AI-driven automation also produces a higher-quality end product, increasing the facility’s revenue opportunities.
“Moreover, AI significantly decreases the cost of measuring what’s happening in a facility,” she adds. “Each AI sensor can identify nearly all the different material types that are of interest. With its software-focused approach, the cost to do waste characterization within a facility drops from thousands of dollars per ton to only several dollars (or in some applications less than a dollar) per ton. This is a several-orders f- magnitude reduction in cost of understanding what’s really happening to material flows in the industry.”
“While economic conditions vary, generally facilities save money by recycling more,” notes Zak Wehman, Clean Robotics associate director of business development. “Instead of paying for landfill collections, they can earn recycling rebates.
“The value is not only in identifying and sorting waste properly before it goes to recycling facilities, but since computer vision is used to identify various objects within the waste stream, it generates a considerable amount of useful data. With this data, facilities can reach their goals faster, eliminating overspending.”
Forms of AI have been used in the recycling industry for decades, with investments in AI/deep learning as a powerful subset of AI can lead to higher recycled product quality, which can allow the recycler to receive more money for the material, says Taha.
In one example, while not using AI’s subset of deep learning, the Santa Barbara County’s Resource Center at the Tajiguas Landfill site in California is using a series of 10 TOMRA AUTOSORT units with its form of AI to help recover recyclable materials from both MSW and single-stream feed material.
The facility is the single largest reducer of GHG in the county, reducing 117,000 metric tons of CO2annually, notes Ty Rhoad, regional director Americas for TOMRA, adding the units are recovering valuable recyclable commodities like plastics, plastic film, paper, and wood from the MSW feed material.
“AUTOSORT’s SHARP EYE technology increases light efficiency, while maintaining the same energy consumption to advance sorting sharpness and improve separation of difficult-to-target fractions,” he adds.
Air jet sorters using AI sensing technology will have similar ROIs as conventional NIR sorters, in the two-to three-year range, notes Hottenstein.
“Depending on whether a robot is used to sort out contaminants or valuable commodities, payback could be as short as three years, but we’ve seen ROIs up to eight years due to the inefficiency of the suction-based effectors and slow pick rate,” he adds.
If the AI is used only for monitoring purposes, the ROI will be highly variable depending on what the application is, Hottenstein says.
“For example, if it’s used on a residue line to monitor the loss of recyclables into the waste stream, the available statistical data allow the AI to provide the plant operator real time feedback if a certain recyclable is not being captured by the upstream sorting equipment,” he says.
“If the AI is used to monitor the quality of a certain commodity output such as mixed paper, then the AIsensor will make sure the adequate quality is achieved by the upstream sorting equipment for the best market price.”
Predictions for the Future?
Looking at what may be possible for the future, AI becomes more useful as it becomes more prevalent, says Hu.
“With one AI scanner, a MRF may be able to quantify the inbound contamination rate, or the value lost to the landfill,” she says. “With multiple scanners, MRFs can understand the entire flow of material through their facility and improve the system as a whole.”
For example, a MRF can track how much inbound PET ends up baled, contaminating the fiber line or other container bunkers, or going to landfill.
“AI can also enable more ‘smart MRF’ technology,” says Hu. “For example, imagine a MRF where the
AI data on infeed volumes and composition can automatically adjust the metering rate to ensure consistent burden depth for downstream sorters and equipment.”
AI can learn to identify nearly anything a person can be taught to identify, notes Marrs.
“It can identify brands, form factors, certain types of damage. This provides an entirely new dimension to sorting capabilities – it can identify aluminum foil versus aluminum cans, or food-grade versus non-food grade polypropylene. Our AI platform then digitizes an image of each item it sees on a conveyor belt.”
The digitization of scrap objects in MRFs opens up many potential applications, notes Marrs.
“The first two that are deployed into MRFs today are robotic sorting and the descriptive and diagnostic analytics provided by standalone sensors,” she says. “As the sensors become distributed throughout aMRF, we’re able to help the MRF become a more data-driven facility to reduce costs and increase revenue. Currently, the MRF is a centralized material hub, but the proliferation of these sensors begins to transform MRFs into information hubs.”
Data capture in MRFs also is influencing the design of new facilities, Marrs points out.
“For example, AMP’s application of AI for material identification and advanced automation has matured to the point where it’s become viable to develop high-diversion secondary sortation facilities that are economical to deploy and sustain nationally.
“Through our secondary sortation model, we recover mixed paper, metals, and a portfolio of #1 to #7plastics in a variety of form factors and attributes with high precision and purity with a special focus on plastic blends uniquely enabled by AI.”
The commodities—including bespoke chemical and polymer blends needed by processors and manufacturers— are re-sold to end-market buyers.
“This secondary sortation model is helping to address the millions of tons of recyclables and billions of dollars’ worth of material feedstock lost to landfill despite the demand for high-quality recycled content from consumer packaged goods companies and brand owners,” says Marrs.
“Increasing regulations, global climate change initiatives, the looming economic downturn, and the labor shortage will put the waste and recycling industry under increased pressure to innovate,” says Ambati.
“We foresee the industry taking advantage of the decreased cost of compute and robotics automation to undergo rapid digitalization. AI-powered automation for not only the robotics but the entire system of industrial machinery inside a MRF will be necessary to keep up with increased demand for recycled feedstock.”
Noting optical sensors in circuits have employed AI algorithms to sort material for more than 30 years, Taha says the combination of massive amounts of data and significantly improved computing capabilities opens the opportunity for solving much more complex sorting problems.
“Deep learning is a class of machine learning algorithms that analyze multiple layers to progressivelyisolate higher-level features from raw input,” he says.
“Thanks to the possibilities of deep learning—a powerful subset to AI— we can contribute to closing the gaps that are important to address when aiming to realize a circular economy and maximize material recovery and recycling.”
Deep learning technologies used in the sorting process allows for the recovery and accurate sorting of materials that cannot be detected and sorted by conventional sorting technology, Taha adds.
Unlike those technologies, deep learning detects materials on the belt not only by color and material type, but also the form of the product scanned, its texture, and the product application, he says.
“For example, in addition to detecting the type of plastic, it can tell if it’s beverage bottle or a detergent container,” notes Taha. “Differentiating between these two products is key to ensuring the quality requirements as stated by the FDA are met,” Taha says.
“Deep learning allows for the differentiation at a sorting stage, therefore recovering more recyclables and achieving higher purity sorting results that lead to higher quality recyclates,” he adds. “With more information obtained when the material is scanned, a clearer sorting decision can be made (e.g.—that is PET packaging and has been used as a tomato tray), and more material fractions be recovered. Consequently, more can be recycled and new revenue streams [can] be opened.”
Another example is the separation of wood chips by type—non-processed wood versus processed wood, notes Taha.
“Conventional technology can’t differentiate between the different wood types and recover MDF from a wood chip fraction,” he says. “Thanks to deep learning, plant operators can now decide which mono-fraction they’d like to recover from wood chips, process and recycle them.
“The purest fractions of non-processed wood can then be used to produce high-quality particle boards, support the industry in meeting their recycled content targets, and give them access to material that in its primary form is hard to access and expensive.”
The power of deep learning is rooted in the use of artificial neural networks that are trained with thousands of extensively labeled images, says Taha.
“When scanning materials on the conveyor belt, the technology draws connections between what it sees and the pictures it has been trained with and thereby learns how to differentiate recyclable materials from nonrecyclables,” he adds.
“Higher resolution and faster sensors of all kinds as well as continually increasing computing power will allow AI-assisted sorting systems [to] make more accurate and granular classification and sorting decisions and provide real-time feedback to the plant operators for maximum uptime and overall system sorting performance,” notes Hottenstein.
Carol Brzozowski specializes in topics related to resource management and technology.
