Have you ever second-guessed yourself when tossing a plastic takeout container into a recycling bin, wondering, “Was that supposed to go in there?”
Because of incidents like this, more recycling ends up in local landfills because it’s expensive to sort recyclable materials by conventional methods, Joe Castagneri, a technical lead at AMP, said.
In recent years, the rate of recycling in Colorado has continued to decline as low as 15.9 percent in 2019, according to CPR News. The national average of successful recycling sits at a higher 35 percent, and this is where companies like AMP see room for positive change, thanks to innovative machine learning capabilities.
According to Castagneri, machine learning is the modern equivalent of the “steam engine in the 19th century or computers in the 20th” – game-changing and available across many industries with an abundance of opportunities. We caught up with Castagneri about using high-speed robot systems with human-like eyes and what it’s like to see your everyday work effect real change in the world.
Louisville-based AMP understands that in order to reimagine the world’s recycling infrastructure, they must utilize AI and robotics to improve the failing material sorting process. AMP’s AI platform processes millions of images to map out complex material patterns and to ensure that the greatest amount of recyclable product is reused.
What's a surprising or interesting way your company is using machine learning?
Machine learning represents an exciting paradigm shift — one that could rival the effect of computers for modern business or the steam engine for the 19th century economy. Modern machine learning techniques allow us to extract incredibly complex patterns from a broad range of sources in the blink of an eye. This type of targeted intelligence allows us to begin building systems that efficiently solve problems that were previously thought intractable.
At AMP, we’ve zeroed in on the problem of recycling, where huge volumes of recyclable material end up in landfill. This is because it’s expensive to sort recyclable materials by conventional methods — sometimes more so than these materials are worth when resold into recycling markets. AMP is driving down these costs by using ML and robotics to automate sortation.
We’ve built high-speed robotics systems that use ML-powered eyes, similar to a human’s, to process image data live, and recognize and remove recyclables from the belt faster and more accurately than ever before. AMP’s mission is to enable a world without waste, and ML technology is the enabler.
What impact has machine learning had on your business, product or the customer experience?
AMP is a machine learning company, so the impact of ML can be felt in everything we do. Our core technology uses deep learned neural networks to drive our robotics systems and data insights.
By using advanced machine learning techniques, our customers benefit from higher quantities of high-purity recyclables they can sell for better prices back into the circular economy. They observe far lower costs from carting what they don’t sort off to landfill. They also see the systems improve over time as we continue to develop our AI.
We regularly leverage the data we generate with our systems to build new algorithms that tell us which systems are underperforming, when systems should be upgraded, and more. Last year, we were able to automate much of our software rollout process by building an intelligent, multistage system that can monitor fleet health, stage software on targets, and even roll back units in bulk if we observe problems. This has accelerated our ability to field test new features, which brings more value to our customers more quickly.
Our products continue to benefit from ML because we are always seeking new ways to apply it to the toughest problems in our technology stack.
“It’s rare to work on the bleeding edge of technology while also having an immediate impact on the world, but that’s what we get to do at AMP.”
What excites you most about the work you're doing?
I’m so excited to work on technology that’s in production! It’s rare to work on the bleeding edge of technology while also having an immediate impact on the world, but that’s what we get to do at AMP. We read recent publications from the literature, implement them, extend them, and experiment with the cutting-edge of computer vision. Whatever we find that works can be in production within weeks. It really drives home the work that we do.
A dry, scientific result might note, “We pushed our precision numbers by X% with this change,” which can feel arbitrary or too calculated to drive passion. Here, if we improve the performance of our systems by some measure, we divert material from landfill as soon as a week later. This is uniquely motivating in a modern world where our impact is often far removed from our day-to-day work.
