Nobody wants to spend hours picking through a steady stream of garbage; this is why many recycling operations rely on automated systems to separate plastic containers, glass bottles, aluminum cans and mixed paper. But these systems have an imperfect track record, so human workers must stand by to nab what the machines fail to catch. Waste Management, a trash-hauling corporate titan with 100 recycling facilities in North America, employs about 3,000 human sorters—but it has difficulty finding workers willing to show up every day, and many quit within hours. This is one reason the company has begun testing new types of robots that could eventually join humans in the sorting lines.

“It’s a very hard role to staff, which is why robotics makes sense for that position,” says Brent Bell, vice president of finance for WM Recycle America, a Waste Management subsidiary.

The United States generates massive volumes of waste. Each American on average tosses about 2,555 pounds of trash per year, an estimated 75 percent of it recyclable. Machines can process the mess much faster than humans; an optical sorter developed by Quebec-based manufacturer Machinex separates recyclables at up to 3,000 objects per minute. This type of technology relies on magnets to pull out some metals, eddy currents to catch others that are nonmagnetic, and near-infrared light to help optical sorters detect different grades of plastic. But their accuracy is compromised by the high rate of contamination inherent to U.S. recycling systems.

Unlike most industrialized countries, the U.S. has overwhelmingly embraced a convenient single-stream recycling approach that lets people toss plastics, glass, metals and paper into one collection bin. This convenience leads to high recycling rates and relatively efficient collection. But it also worsens contamination: U.S. industry estimates suggest 20 to 25 percent of submitted items are unrecyclable trash. The problem stems partly from consumer confusion about what items qualify, along with “wish recycling”: taking a chance that chucking something in a designated bin will give it a new life. Many items also become unusable during collection, transit, or in-truck compacting that smashes everything together, staining paper with fluids and grinding broken glass into other items.

Thus, human workers have to step in. Otherwise, misidentified items can damage expensive equipment or temporarily shut down recycling operations. For example, light plastic and paper pieces are processed in giant spinning machines. Overlooked metal or glass containers that sneak in can turn into high-speed projectiles, says Susan Collins, executive director for the nonprofit Container Recycling Institute based in Culver City, Calif.

Instead of struggling to find humans to constantly oversee the process, some companies are testing AI-driven robots equipped with grippers or suction cups to pick out recyclable objects. They use cameras and other sensors, coupled with machine-learning software, to recognize visual patterns associated with specific items. In order to replace people, however, such bots will eventually have to outperform us—human quality-control workers pick out about 30 to 40 items per minute. “If this [robot] can pick two times or three times as many objects as human workers, then we could start looking at the economics and seeing if we can justify a purchase,” Bell says.

Last year Finnish company ZenRobotics debuted its Fast Picker robot, which can grab approximately 66 objects per minute. Now some recycling companies are talking about AI-driven “dark factories” or “lights-off sites” without human workers, says Janica Johansson, marketing director for ZenRobotics. Other companies still envision robots working alongside humans. “They’re not going to replace people—we will absolutely keep quality control people in our facilities—but they will allow us to process more tons per hour and provide cleaner recyclable products,” says Steve Sargent, director of recycling for Rumpke Waste & Recycling.

Like Waste Management, Rumpke hopes to integrate recycling robots into its existing operations, which include regional services in Ohio, Indiana, Kentucky and West Virginia. The company is especially keen on trying out a Machinex-developed robot called SamurAI at a recycling facility in Cincinnati. SamurAI can pick out about 70 objects per minute, so it works more slowly than Machinex’s standard sorting equipment. But its still-superhuman speed means it could really help in the quality control department.

Newer technologies could ratchet up robotic efficiency even more. The Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory (CSAIL) recently unveiled an experimental RoCycle system that uses soft Teflon “fingers,” which have fingertip sensors to detect object size and stiffness. This robot is much slower than humans; when it tried identifying objects on a simulated conveyor belt it only achieved 63 percent accuracy by touch alone (paper-covered tins posed a special challenge). Still, that accuracy could improve by combining tactile data with visual data from cameras, says Daniela Rus, a professor of electrical engineering and computer science and director of CSAIL at MIT.

“AI solutions allow us to preserve this convenience and widespread adoption [of single-stream recycling] while ensuring that we aren't just shifting the labor of sorting to more vulnerable populations,” says Lillian Chin, a PhD candidate in electrical engineering and computer science at MIT and lead author on the RoCycle paper. However, skeptics point out that robots in recycling facilities cannot fix some of the fundamental flaws of the U.S. single-stream system, including the contamination issue that begins at curbside collection. “The robots will not be able to unbreak glass if the glass is arriving at sorting facilities already broken,” Collins says. MIT researchers suggest robots could someday presort recyclables before curbside pickup—but it is unclear who would pay to install such robots, even if the technology matures.

Despite these limitations, the turn toward robots has gained new momentum after China rocked the global recycling industry by halting its imports of the world’s contaminated mixed paper and plastic waste in early 2018. No longer able to outsource the dull, dirty and dangerous task of trash-sorting, many American cities and towns have resorted to dumping recyclables straight into landfills or incinerators. Both the U.S. and overall worldwide recycling average for plastic is a measly 9 percent; another 12 percent of global plastic waste ends up incinerated, and 79 percent either goes into landfills or clutters up the natural environment.

Robots that can demonstrate superhuman sorting speeds without too many mistakes may prove good enough to join the recycling lines. But beyond new technologies, companies such as Waste Management and Rumpke still emphasize the need to educate customers about what they can recycle and how they should do it. That means focusing on the basic recyclable categories and teaching people to avoid adding items like garden hoses, Christmas lights and plastic bags into the mix. It is a long-term struggle to change humanity’s wasteful lifestyles—and everyone acknowledges that AI and robots cannot solve the recycling crisis without humans doing their part.