Recycling Gets a Robotic Boost

The job of a recycler isn’t easy. According to Cottingham & Butler, the turnover rate is 22 percent and the injury rate ran around 4.7 per 100 workers in 2023, which was among the highest in the country per profession, per U.S. government data.  

Companies must often hire multiple times to fill critical recycling positions and this practice is unsustainable in the long run, said Areeb Malik, founder and chief technology officer at Glacier, a San Francisco startup that’s combining robotics and AI to revolutionize the recycling process. 
 

Eyes on the prize 

As a possible solution to the steadily decreasing number of human recyclers, Glacier’s robots specialize in collecting recyclable materials from conveyor belts and aim to replace the antiquated manual process of identifying, sorting, and picking up recyclable materials. 

“We can identify anything that needs to be recycled,” Malik said. 

These robots use AI to identify materials such as plastic, cardboard, paper, and cans. They can also identify non-recyclable items such as diapers and jeans. 

“Each AI model lives inside of a camera that’s looking at some conveyor belt in a facility,” Malik explained. 

More for You: Engineers Integral to a Successful Circular Economy

The camera systems are like a set of eyes that help the robot identify and sort material moving across conveyor belts. That information is fed to the robots in real-time, telling them exactly where and when to pick up specific items. 

“We spin some motors, send the arm over to where it needs to be, and then do a pick motion, and we hit the object exactly where we expect to because computers are very precise,” he continued. 

The robots aren’t humanoids by design, but instead look like static boxes that straddle the conveyor belt. As materials flow under the conveyor belt, the robots use suction to pick items up. 

Glacier’s technology enables the robots to pick up to 45 items per minute, far outpacing even the fastest of human sorters, Malik said. 

“We have about 20 robots installed across the country and an additional 20 camera systems as well,” he added. 
 

Plug and play 

Waste management facilities are struggling to hire personnel. A single recycling plant sorter job in California costs upward of $80,000, when accounting for salary and benefits. So, Glacier’s goal is producing a generalized robotic and camera system that can be installed in any recycling plant. 

“When we plug in, our machine can cost you as little as $30,000 to $50,000 a year, so we basically can beat that in theory,” Malik said. 

The Glacier system has two functions: it can either pick up or leave in place specific items on conveyer belts.  

The “positive” picks a specific class of recyclable items, such as plastic bottles or cardboard, while leaving others. The “negative” picks leave a specific class of recyclables such as plastic bottles or carboards on the conveyer belt, while picking out others. 

But there are challenges when it comes to robots identifying items. For example, the system can identify a plastic bottle, but may have a challenge identifying a severely twisted water bottle with a napkin in it. 

Discover the Benefits of ASME Membership

“The number of ways a bottle shows up under our camera is next to infinite. Imagine how many ways you can crush, twist, tear, cover, dirty that bottle that is going into your bin,” Malik said. 

In those situations, Glacier uses specialized tooling that identifies things that the AI may have missed. The items are also routed to human labelers—the human-in-the-feedback-loop approach helps improve the system as well. 

“We route that information to human labelers to do validation and correction, and subsequently reintegrate that into our training pipeline to improve the performance of our model,” he explained. 

Malik, whose background is in AI software, started the company in 2019 after seeing an opportunity in the climate sector. Over the years, he has hired a gamut of skilled mechanical and electrical engineers to build these robots.  

Structural engineers are figuring out the robotic frame, while motion engineers build the moving pieces, and pneumatic engineers deal with airflow. 

Glacier’s specialized recycling AI model is flexible and can be plugged in, even when new models come along. 

“There are new models every week, every month that are outperforming one another,” Malik said. 

Agam Shah is a business and technology writer in Phoenix.