Can robots help automate the recycling sorting process?

Reduce. Reuse. Recycle. And, if AMP Robotics has its way, robotics.

The Colorado-based robotics startup specializes in building waste-managing robots, that operate on a conveyor-belt system and organize piles of recycling and garbage based on their characteristics. The company has raised $74.5 million since its founding in 2015, from investors like Sidewalk Infrastructure Partners, an Alphabet-backed offshoot of Sidewalk Labs, and Sequoia Capital.

These recycling robots are trained to identify and then sort a variety of plastics, cardboard, paper, and aluminum metal, using computer vision. Matanya Horowitz, founder and CEO of AMP Robotics, told Emerging Tech Brew that for over two years, the robots have been able to correctly identify and characterize recyclable material with 95%–98% accuracy.

“There really is a lot of value in all of this material that we put in the recycling bin. Its value [is] in the plastics, the metals, and the paper. The problem is the cost of separating it is high because it requires a lot of work and requires a lot of heavy equipment that doesn’t work too well. And robots help reduce that cost of separation,” Horowitz told us. AMP is looking to grab a slice of the $7.6 billion recycling-facilities industry, but declined to share its revenue.

Separation anxiety

Part of the problem surrounding the recycling industry is the rates at which some countries recycle their waste. A 2019 study found that Canadians only recycle 9% of their waste, while in the US, the EPA has set a goal of recycling 50% of its waste by 2030. The UK had a similar goal to recycle half its local-authority-collected waste by 2020, but fell short of its goal at 44.0%, but in Japan, up to 86% of plastic waste was recycled.

The recycling industry might be well-intentioned, but it is not a perfect industry, and the world is taking a harder look at the efficacy of recycling. Recycling itself costs energy and resources, while, according to the EPA, only 32.1% of municipal waste in the US is actually recycled. AMP claims it can help solve these problems by moving faster as well as reducing costs and human labor.

The amped-up recycling robots are capable of identifying and sorting through municipal waste, e-waste, and construction and demolition debris, and use machine learning and AI to spot out and differentiate between types of waste. Horowitz said that the robots are capable of identifying over 100 types of waste in minute detail, categorizing them by color, size, shape, opacity, form factor, brand, and more.

AMP operates globally, deploying over 200 robots in places like Japan, Canada, and the Western Europe, but also has domestic partners in companies like Waste Connections and Waste Management and municipalities like Emmet County, Michigan and Boulder County, Colorado.

For companies looking for a helping hand, Horowitz said the price of the robots varies depending on the buildout AMP’s customers want, but could cost “hundreds of thousands of dollars.” Horowitz claims most customers see an ROI in two years, regardless.

“Recycling facilities really aren’t able to extract value from material because they can’t quantify the performance of their systems. When they separate out plastics, they kind of eyeball it, and they’re like, ‘Okay, it looks like it’s 90% pure or 95% pure,’ but they don’t really know,” Horowitz said. “What it means is all the buyers of plastics have to assume the worst, so they don’t pay for the material’s actual worth. And then…the material doesn’t end up going to the highest and best use because they have to assume that the chemical properties have been degraded, even if they haven’t.”

But, but, but: On top of all this, the algorithms that the machines are trained on are in constant need of fine tuning and training to identify types of trash. Horowitz said that the algorithm needs to be refined to identify things like the plastic composition of clear Solo cups or the difference between torn and intact paper, for example.—JM