Matanya Horowitz founded Louisville, Colorado-based AMP Robotics in 2014 with a mission to change the way materials are recycled. Through AMP’s artificial intelligence (AI)-powered robotic sorting solutions, operators at material recovery facilities (MRFs) that process municipal solid waste and recycling facilities in construction and demolition (C&D), e-waste and metal scrap can cut down on contamination while reducing their reliance on manual sorters.
Horowitz talked with Waste Today about how he got started in the industry, the role robotics play in the industry and what advancements in AI learning mean for the future of recycling.
Waste Today (WT): How did you get involved in robotics, and why was recycling the industry you chose to target?
Matanya Horowitz (MH): Growing up, I was always fascinated by robotics, and that inspired my education. During college, there were many new developments occurring in robotics and machine learning. My Ph.D studies focused on robotic control theory, path planning and computer vision, as well as research with the U.S. Defense Advanced Research Projects Agency (DARPA) on a number of AI-related projects. When I graduated, I sought to apply these latest technologies and looked at a number of industrial applications. One of these industries was recycling. The convergence of machine learning and robotics unlocked major opportunities to automate what had historically been tasks that were labor intensive, high cost, inconsistent and limiting. Now cost-effective automation could be applied to extract more value from these very complex, heterogeneous material streams. The challenge was very appealing to me, especially as I felt that it could solve a very important and unmet need.
WT: The National Sword policy caused a real shift in industry dynamics and end markets. How did that affect your business?
MH: The National Sword policy increased the importance of our technological innovation to address the urgency to reduce costs, increase efficiency and improve purity. These market pressures raised the need to radically change the economics of recycling. This timed well for the commercialization of AI-powered robotics systems to help transform recycling economics, which drove demand for our application. As a pure-play AI and robotics company, we were able to completely focus on this need and organically develop the AMP Neuron AI platform for the recycling industry. We were also able to rapidly modify it to meet individual customer requirements and continuously improve our AMP Cortex robotic system. These market pressures also elevated the need for facility operators to make data-driven decisions to optimize their businesses. Our systems enable the identification of trends, like the daily characterization of material flows. Gaining this knowledge creates advantages for our customers to measure and improve their operations, thus allowing them to extract the most value from material recovery. In sum, the market demand for our technology has been tremendous. We are gaining customers at a rapid rate while getting repeat orders as our customers scale automation in their facilities.
WT: Can you talk about how AI learning might allow operators to be more nimble in targeting their materials?
MH: Our AI platform, AMP Neuron, uses advanced computer vision and machine learning to train itself by processing millions of material images. It teaches itself to look for different visual attributes on the line such as size, color and texture. Due to its learning-based approach, it learns from experience and is able to identify more specific categories of material and adapt to new material (and packaging) types. This helps operators monitor and adapt to changes in their material stream. For example, one of our customers sought to sort and process coffee cups. Historically, it wasn’t economically feasible to recycle coffee cups, but with our technology, they could. This broke a historical barrier, created a new commodity bale and met the demand for a material type that was increasing in volume.
WT: How does a MRF’s incoming material stream influence its sorting needs and the potential usefulness of robotics?
MH: Two major influences on the incoming material stream at a MRF are the complexity of products and packaging, as well as contamination rates. Household and consumer products are becoming more visually diverse and built from a variety of materials. This product diversity leads to consumer confusion and many non-recoverable objects being put into the single stream and increasing contamination rates. The contamination creates several challenges that affect MRF sorting needs. First, high rates of contamination increase the cost to sort single stream. Second, contamination finds its way into recyclables entering the commodity markets and decreases product quality.
Robots are very useful in addressing these challenges. First, the artificial intelligence that powers the robotics perceives and learns all of the objects in the single stream. The more objects the fleet of robots sees, the more they learn how to identify recoverable objects and contaminants. The robots can effectively behave like a sensor to detect and report contamination levels to MRF managers. Many robots are deployed at quality control stations where the robot can remove contamination and also quantify the quality of the material destined for the commodity market. Robots are also deployed in fiber and container line recycling, where they tackle the increasing costs to sort desired material. Robots are priced to be competitive with human sorters, and are picking and placing objects at up to twice the speed of a worker. This cost savings becomes more pronounced as MRFs operate two or three shifts per day.
WT: Robots in MRF applications are a new proposition for many. Where do you see the biggest need?
MH: Robots in MRF applications may seem like a new proposition for many, but one can quickly find robots that have been operating for several years and solving MRFs’ greatest sorting challenges. These include quality control, container line and residue line recovery. Each of these stations within a MRF has been traditionally staffed by people. Most MRFs are challenged to recruit and retain people in these roles and are chronically understaffed, resulting in decreased throughput and product quality. Every new sorter has to be trained on what commodities are in the complex single stream, and that knowledge is lost when they depart. Robots are rapidly deployed into the footprint of the sorting station, competitively priced and operate each shift of every day. Operating at up to twice the speed of a person creates value for the operator on day one. Robots have been operating in these positions for years now and have amassed knowledge of the material stream, which is no longer lost when an employee leaves.
WT: How do you see robotics/AI technology evolving over the next five to 10 years?
MH: The use of AI in recycling is only just beginning. In the near future, you’re going to see very precise categories for identification and purity rates that are difficult to achieve with alternative technologies. With these capabilities, what is exciting is that we’re starting to see robotics applications that we never thought of. You now have the ability to measure nearly any facet of the material stream that you care to know, and we can only imagine the possibilities that is going to open up.
We believe the reliability and performance of these robotics systems will continue to improve quickly, and they will become a staple of the industry. Speaking five to 10 years in the future, I think it will be unusual to visit a recycling facility that doesn’t have a robot, or perhaps one that is not almost fully staffed with robots. I think in that time horizon, you will also have several facilities that are fully autonomous.
An abridged version of this interview ran in the July-August issue of Waste Today.