Equipment Focus: Robot-Assisted Sorting

July/August 2018

By Megan Quinn

It’s not exactly a robot uprising, but robotic sorting machines are starting to enter the recycling industry. These machines use artificial intelligence to make autonomous decisions about what materials to sort and robotic arms to remove selected items from the conveyor belt. Manufacturers of these futuristic sorting machines promise cleaner, higher-quality materials that will help recyclers respond to China’s tightened import restrictions and contamination standards. They provide safety and labor benefits as well: Robots don’t get tired and can eliminate sorting-line-related injuries or human fatigue, thus allowing recycling facilities to run longer with lower operating expenses, manufacturers say. And they’re reprogrammable: If a facility’s needs or streams change, an operator merely changes the settings instead of buying an entirely new piece of equipment.

Just a few U.S. recyclers have installed robotic sorting equipment in their facilities to date. It’s a significant investment in equipment that doesn’t yet have a proven track record. Recyclers must decide whether this new machinery will fit their budget, throughput, and current production needs, and whether they want to be a pioneer or wait until robots are more established in the market.

A Brief History of Recycling Robots

Robotic-assisted recycling has come a long way in a very short time. Just two years ago, a few companies were experimenting with prototypes that used artificial intelligence to identify and pluck target materials off sorting lines. One high-profile prototype was Clarke, a carton-sorting robot named after science fiction writer Arthur C. Clarke. In 2016, AMP Robotics (Denver, Colo.) designed and installed this Cortex-model robot at Alpine Waste & Recycling, a Denver-based materials recovery facility, with help from a grant from the Carton Council of North America (Denton, Texas). After some fine-tuning, the robot was able to pick cartons from the recycling stream almost twice as fast as a human sorter. AMP Robotics uploaded information Clarke “learned” from the pilot program to the cloud, where the company can download it into the “brains” of other AMP robots. The pilot program ended in 2017, but Alpine still uses the Cortex robot in its facility, Brent Hildebrand, Alpine’s vice president of recycling, said in a press release. AMP and the Carton Council installed a second carton-picking Cortex robot at Dem-Con Cos. (Shakopee, Minn.) in August 2017.

Clarke wasn’t the first robot created for recycling. ZenRobotics (Helsinki), which began working on its first prototype sorting robot in 2006, had a robot ready for the construction and demolition recycling equipment market in Europe by 2011. Today it offers several models of AI-assisted sorting machines; it installed its first sorting robot in the United States (in Austin, Texas) in 2016. Another major robotics company, Sadako Technologies (Barcelona, Spain), has made a name for itself by creating robots to do what it calls “dangerous, dirty, and dull” tasks. Sadako’s artificial intelligence is part of the Max-AI technology that equipment maker Bulk Handling Systems (Eugene, Ore.) co-developed with NRT (Nashville, Tenn.), a BHS subsidiary. The company debuted Max-AI in 2017.

North American recyclers using robotic sorters—including GreenWaste Recovery in San Jose, Calif.; Penn Waste in York, Penn.; and Recon Services in Del Valle, Texas—have only begun to install them in the last year or so. Equipment manufacturers are quickly rolling out new models, too. ZenRobotics introduced the Fast Picker, meant for lightweight materials such as packaging and dry mixed recyclables, in early 2018, around the same time Machinex (Plessisville, Quebec) debuted its first-ever AI-assisted sorter, the SamurAI. Lakeshore Recycling Systems (Morton Grove, Ill.) installed the first SamurAI in the United States at its Forest View, Ill., facility this spring. Manufacturers anticipate this trend toward robotic sorting will continue. “Robotics are in the early stages, and it can only improve from here,” says Chris Hawn, CEO of Machinex.

Better, Faster, Stronger

Recyclers typically install these robots as quality-control checks behind optical sorters or other sorting equipment, meaning they perform sorting jobs humans traditionally do by hand. Equipment manufacturers say they’ve designed the machines to offer faster, safer, longer-running, and more precise picking than what humans can do. They estimate humans can pick an average of 30 to 40 items off of a conveyor belt per minute—more when they’re at the beginning of their shift and less when they’re getting tired. Machinex says its SamurAI can perform up to 70 picks per minute, while AMP’s pilot Cortex robot achieved 60 carton picks per minute. ZenRobotics’ Heavy Picker, designed for heavier ferrous and nonferrous items as well as “meatballs,” or partially shredded small electric motors, does about 33 picks per minute, and its Light Picker, which sorts cartons and paper, does about 67 picks a minute. Max-AI can do about 65 picks a minute, these manufacturers say. “A robot has consistently high levels of performance,” says Peter Raschio, marketing manager for BHS.

Robots can also sort and pick multiple items at once and operate for longer periods of time, Raschio says. Max-AI can look for up to six different items at a time and can run “virtually 24/7,” for example, while Alpine’s Cortex sorter ran for about 16 hours a day during the pilot, Hildebrand says. ZenRobotics designed its Heavy Picker to operate in conjunction with a bunker-fed conveyor belt. A human employee fills the bunker before going home for the night, allowing the robot to continue sorting long after the last shift leaves, says Will Hancock, vice president of operations for Plexus Recycling Technologies (Denver), the North American distributor for ZenRobotics. These robots can also help employees avoid the dangers the sorting line presents, he says. The work done by the Heavy Picker, for example, he calls a dangerous job. “The material is heavy and sharp and it has the potential to cause injuries.”

Hancock says that the Heavy Picker can replace six to eight manual workers. The idea of replacing human workers with machines is controversial, and some recyclers say they’re not buying robotic equipment to cut their workforce. It can be a challenge to hire and retain employees to work on the picking line, they say. Robots give recycling facilities more freedom to move existing employees to areas of the facility that need more manpower instead of always looking for workers for the picking line, Hawn says. Others tout the economic benefits of using robots instead of humans. Hiring and training new people can get expensive, Hancock says. Labor costs were one of the reasons Recon Services gave for having Plexus install a ZenRobotics sorter at its construction and demolition facility in 2016. It is the first ZenRobotics equipment installed in the United States. “Business owners are getting tired of dealing with the headaches of employment, labor costs, and finding good help,” Hancock says. “If two guys don’t show up on your picking line, where does that leave you?”

How They Work

Manufacturers say they have designed robotic sorters meant to work alongside traditional sorting equipment instead of replacing it. Traditional sorters are programmed to do one task with human operator guidance, whereas robots analyze data that helps them make autonomous decisions. “Optical sorters and robotic sorters are not substitute products, but complementary,” Raschio says. “Optical sorters are excellent at creating purity levels above 90 percent at high volumes. Robotic sorters complement optical sorters with the quality control in which many products are identified” and several types of products can be sorted at once, he says.

Many of these robots come with a suite of sensors and cameras, plus software that “learns” from sensor input and autonomously adjusts its settings for better sorting. The Heavy Picker, for example, has near-infrared sensors, a 3-D laser mapping sensor system, a high-resolution RGB camera, and an imaging metal detector. These sensors feed information to patented software—the robot’s “brain”—that quickly analyzes the data to help it make decisions about what to pick up, Hancock says. Max-AI learns in a similar way, Raschio says. It “is trained with millions of images with different materials already identified.”

Machinex’s SamurAI, which uses a Cortex-model robot with hardware designed for recycling applications, relies on one camera to identify materials, but it makes decisions based on data all other Cortex robots in operation have gathered. When one Cortex robot “sees” a new item, such as a limited-edition soda can that looks different from others, it stores that information for use when that can appears in a MRF a thousand miles away, Hawn says. “Packaging changes daily, and this [machine] learns and shares that information,” Hawn says.

Recognition is the first step. Physically removing the specified materials from the conveyor belt is the second. These machines have the ability to leave or remove whatever the operator wants. The operator can tell the robots to make “negative” picks, meaning the robot leaves a certain type of item on the belt, or “positive” picks, meaning it removes that type of item from the belt. Operators also can quickly change those assignments to prioritize one type of item over another. The SamurAI, for example, uses AI to identify materials according to a predetermined task hierarchy—for example, it can prioritize capturing PET over HDPE depending on which commodity is more valuable or has more reliable buyers at the moment, Hawn says.

Once the AI computes the data and makes a decision about what to pick up, the robot uses a mechanical arm to pluck the item off the belt. Some models use suction cups for removal, Hawn says, while others use paddles that grip the item after calculating how far down the arm must go and how firmly to hold the item.

Running the Numbers

Robots are cutting-edge technology designed to save recyclers money or increase profits in the long run. The purchasing decision “is all based on [your] challenges—what’s the inbound material and what are [your] end-product goals?” Raschio says.

Before you go down this road, manufacturers say, check your existing equipment to see if small tweaks can improve quality without such a big investment. Since most robots currently serve as a quality-control check at the end of the line, “you should ask, are you losing valuable commodities to residue? If you are, there might be modifications [you can make] within the plant to reduce that over time,” Hawn says. Throughput is another factor. Robots will do the job more efficiently, but it might not be worth the investment for smaller MRFs, he says. “If there are not enough picks per minute—say your robot can do 70 picks a minute, but it’s only doing three to four, and the robot is going to be static the rest of the time—it might not be the best fit.”

On the other hand, facilities with material streams that change over time might consider whether a robot can be more responsive to those changes than human employees or tweaks to existing equipment because you can program it with many “recipes” to pick out items that vary from source to source, Hancock says.

Finally, assess whether human sorters are the weak link in your operation. “Are your high-value commodities—say, PET—contaminated because your sorters aren’t doing the job you need and you can’t meet spec because of it?” Hawn asks. Calculate labor costs and the cost to hire and train employees who typically do hand-sorting jobs, then compare that with the cost of a robot, he says. Depending on the model and job you want it to perform, an AI-assisted sorter could cost from $200,000 to $900,000, with some also requiring thousands of dollars a year in software leases, these manufacturers say.

As they would with any other piece of sorting equipment, recyclers will need to consult with their equipment suppliers to determine where robots can best integrate into their facility, manufacturers say. Most models are designed to be modular, but they may require certain conveyors and feeders for optimal performance. For example, ZenRobotics’ Fast Picker and BHS Max-AI say they can work with different conveyor widths and fit most picking stations without additional modifications, but a company representative should visit the facility before making a recommendation. Machinex says its SamurAI fits conveyors 42 to 48 inches wide.

Recycling facilities considering robotic separation also need the ability to properly feed material onto the conveyor. Most of these robots can control the speed of the conveyor automatically, but they can only identify materials that are evenly distributed on the belt, manufacturers say.

To get the best fit, robotics companies will send a representative to your facility for a consultation and recommendation. After you purchase one of their machines, they then send a team to install it and make adjustments while it learns how to perform the needed tasks. The time it takes from installation to optimal sorting can range from a few days to a few weeks, these companies say.

Better Data Analysis is Ahead

These first-generation robots represent a new era of innovation, but manufacturers say they’re already working on ways to get even more out of the technology. In the future, robotic sorters will have the ability to provide in-depth data that can be useful for day-to-day operations, they predict.

Some models can already create daily reports that show what the robot picked, how much of the material it could recover, and the percentage of the stream the recovered material represents, Hancock says. “You can control your inventory much more easily with data in hand,” he says. In the future, “AI will be able to give us data for the full operation of the plant,” including real-time data about the makeup of the stream, Hawn says. “It can give you trend data, but it will use it to self-diagnose and fix issues without operator intervention. That’s down the road,” he predicts.

For now, manufacturers are working to gain the trust of recyclers who are skeptical of robotics’ as yet unproven reliability. “Back when optical sorters first hit the market, recyclers were wary that they could only reliably work for PET,” Hawn says. Then the technology improved to be used on everything from PET to various grades of fiber. As more recyclers integrate robotics into their facilities, more choices and more models will enter the market, manufacturers predict. That’s good and bad, Hancock says. “There will be more robot companies in the market. That will mean more choice, but not all of them will have quality products,” he warns.

There’s a lot at stake for equipment manufacturers with new AI equipment, Hawn says. “It’s promising, but this is also a critical time right now. [We] as equipment manufacturers need to be transparent about equipment limitations and not oversell. The second that happens, the industry is going to say ‘the robot is bad,’” he says. Yet robots are an exciting new frontier that could become a ubiquitous tool in the near future, Hancock says. “We have just scratched the surface of what robots can do. We don’t know everything they’re capable of yet.” 

Megan Quinn is reporter/writer for Scrap.

(Sidebar Content)

Apple Applies Robotics to Electronics Recycling

Apple’s new robot, Daisy, can take apart iPhones to recover valuable materials inside. Daisy can disassemble up to 200 devices an hour and can break down and sort components from nine different versions of the iPhone, it says. The robot is a successor to Liam, another recycling robot Apple created in 2016. Daisy was made out of some of Liam’s old parts.

Apple engineers created Liam in 2016 to better reclaim materials used to make its devices. A recycling robot that could sort components from Apple devices into various streams could provide recycled materials at close, if not identical, specifications to those it requires of virgin materials and new components, according to Apple’s Liam-An Innovation Story white paper.

It also designed the robot—first Liam, now Daisy—to extract rare earth metals that would be tough to recover from more traditional recycling processes. Specialized parts of the iPhone, such as its acoustic modules, use magnets made of neodymium, praseodymium, and dysprosium. In traditional recycling methods, “these highly magnetic materials disintegrate into fine powder and are lost to the ferrous fraction from which they cannot be recovered using today’s technology,” it says. Liam and Daisy remove the acoustic modules so Apple can send them to companies that specialize in converting rare earth magnets into the raw material needed to make new magnets. Visit www.apple.com.