October 13, 2021
7 Minute Read
PolyPerception provides an AI-powered waste management platform to plastics and material recovery facilities. Their technology tracks each individual piece of waste at an object level. This gives the facilities visibility into their waste streams so they can operate more efficiently and responsibly. The team has partnered with Sama to help fuel this technology, to further their mission of empowering stakeholders across waste management—from recyclers to municipalities to legislators—to make more sustainable decisions about waste and its impact on the environment.
As the volume of discarded waste continues to grow globally each year, so has the pressure on our landfills, resources, and environment. With waste disposal technology, infrastructure, and regulators struggling to catch up, one significant barrier blocks the way: visibility.
On average, 8 tonnes of waste passes through a waste sorting facility every hour. If this sounds like an almost inconceivable amount of waste, that’s because it is. For the operators of sorting centers, understanding and documenting the waste passing through each day remains a major challenge, forcing the waste management industry to operate with a lot of blind spots.
To circumvent this and better understand the makeup of the waste passing through, many facilities implement a sampling process, but sample sizes are typically small. Though 8 tonnes of waste passes through a facility every hour, sampling is often only done once a week. Even representative sample sets can be skewed by the subjectivity of human bias, making consistency a challenge.
These factors lead to datasets that don’t often show the full picture, forcing waste management facilities to largely operate on human intuition; a less-than-ideal approach to the decision-making required to meet federal regulations and to cut operational costs.
The sheer volume of waste passing through a facility is not the only obstacle for sorting center operators. In the world of waste management, tons of other parameters are constantly changing, transforming the efficiency and economics of the process.
For starters, brands often change their packaging in ways that impact how it’s sorted and where that packaging finally ends up. For example, Heinz is currently using a multi-layer PET bottle with a non-recyclable barrier, but this is soon slated to change in Europe. These changes often happen suddenly and without warning, giving facilities little time to react to make sure these bottles now end up in the appropriate output stream — and therefore at the right recycling facility.
Packaging also varies from region to region. If a facility onboards a new source of waste, they may not understand regional differences, and their machines may not be optimized to sort materials accordingly. Slight margins of error are amplified when you consider the number of sorting steps and the sheer volumes processed.
Keeping up with shifting packaging trends is only compounded by regulations that are also constantly in flux, and differ from one country to the next. All these variables from the outside require sorting facilities to constantly optimize their processes and plan strategically, but this is hard without quantifiable data.
This is where PolyPerception comes in: to give operators in facilities continuous visibility into their waste flow systems by adding cameras to monitor the end-to-end process. This means tracking each piece of waste at an object level. Traceability and transparency allow these facilities to operate more efficiently, bringing them better commercial terms and preparing them to tackle new legislation and packaging trends.
In order to deliver actionable insights and quantitative data to their clients, PolyPerception set out to build a robust multi-object tracking model, but quickly found that the accurate labeling of data would play a key role.
Crucially, they needed a labeling solution that could accurately label waste objects that travel quickly on conveyor belts in facilities with less-than-ideal lighting conditions. But they also needed a partner who could effectively annotate millions of waste objects while accounting for a wide range of packaging types, materials, processing speeds, and processing conditions.
Rafael Hautekiet, CEO of PolyPerception, was delighted by how feedback loops with Sama’s managed workforce of annotators resulted in better quality data for their model – annotations with an average Quality Score of 99%:
“The team quickly learned to distinguish between waste objects, which differ greatly from region to region. Communication channels remained open for feedback, and we had a continuous open discussion about how the efforts were progressing.”
This line of open communication was important to PolyPerception when they were evaluating different data labeling partners. They wanted to feel like annotators were an extension of their own team, both to maintain visibility into the labeling process and to ensure that the many nuances of their labeling needs were being met.
For PolyPerception, accurate data in waste management has cascading effects for society – a rising tide to lift all boats. With more reliable data:
The economic, social, and environmental benefits of the above are too numerous to list, but they are at the core of PolyPerception’s longer-term vision: of empowering stakeholders across waste management to make more impactful decisions, to move the world toward a circular future.
To PolyPerception COO Parshva Mehta, accurately labeled data plays a small but important part in this ambitious undertaking. This is why it was important for PolyPerception to do data labeling a different way: to work with a provider whose values aligned with theirs, and with their mission to make the world a better place. In Parshva’s words:
“There’s a possibility to make an impact on legislation and on the environment, but not without accurately labeled data. We appreciate that Sama started out by disrupting the status quo and by having a strong social mission. We really resonate with this.”
Amanda is the Content Marketing Manager at Sama, where each day she gets to learn about AI and how it’s transforming industry. She’s passionate about technology but periodically tries to get away from her computer to explore more analog things like collage and quilting.