TOMRA and NEXTLOOPP are currently working on a project that uses cutting-edge artificial intelligence technology to make critical recycling processes more efficient. In this article, Professor Edward Kosior, founder of Nextek and NEXTLOOPP, tells more about this groundbreaking collaboration.
It’s been almost four years since we launched our groundbreaking multi-participant project, NEXTLOOPP, to close the loop on post-consumer food-grade polypropylene (PP) packaging.
The world was still reeling from the emergence of COVID and travel, let alone human interaction, was limited. But Nextek had already spent more than a decade researching and testing science-based solutions to accelerate the efficient sorting and vigorous sanitization of what we had identified as a key missing link in the recycling stream, food-grade PP.
We were ready to launch our cutting-edge technology into the commercial world. The timing was perfect: in 2022 the global PP market was estimated at £60 billion, expected to rise to £70 billion by 2028. Yet only 1% of this is recycled.
In Europe alone, 44 percent of PP is used in consumer packaging, divided between rigid (22 percent) and flexible packaging (16 percent). Brand owners were looking for low-carbon solutions for the circular economy.
The sorting dilemma
NEXTLOOPP’s progress since 2020 has been well documented. The project’s innovative fluorescent marker technology was one of the leading candidates to address the problem of distinguishing food-grade plastic packaging from the waste stream.
However, we were not the only ones. There were a number of advanced solutions that aimed to do the same thing, resulting in a lack of consensus on which technology should be adopted.
The overall disadvantage was that marker technology by its nature requires changes to labels or packaging, and the industry as a whole disagreed on which technology should become the standard.
NEXTLOOPP’s 50 global participants trialled and deployed Nextek’s technologies for both sorting and decontamination, keeping an open mind as to which technology would best serve the industry.
The clear belief that authentic collaboration can accelerate results recently paid off when the project achieved a major sorting breakthrough that has solved the dilemma of which sorting technology to adopt universally.
Beyond expectations
Since the start of NEXTLOOPP, one of the main focuses has been the efficient separation of food and non-food packaging, for which the team has very successfully trialled UV markers in collaboration with NEXTLOOPP participant TOMRA.
At the time, this was the most effective spectroscopic sorting technology to separate the same polymer into food and non-food fractions by adding an additional, fully integrated, coded sorting dimension to the standard NIR/VIS sorting systems.
Even before NEXTLOOPP launched the leader in sensor-based sorting technology, TOMRA had introduced the industry’s first AI-based deep-learning sorting solution to separate silicone cartridges from PE streams and later for wood sorting (in 2019).
In early 2024, TOMRA introduced the food-safe application to address the industry-wide challenge of food-safe separation.
NEXTLOOPP supported the PP field validations conducted by TOMRA to test GAINnext’s capabilities in industrial conditions, but given that the AI had to be learned and NEXTLOOPP already had our own highly efficient marker technology ready plug-and-play, we thought we would start with markers and then gradually introduce AI.
We did not anticipate the increasing speed at which TOMRA expanded the capabilities of GAINnext.
By early 2024, TOMRA had accelerated its GAINnext deep learning technology to separate food and non-food plastics and used it, among other things, to identify PP packaging. It didn’t take long for me to realize that this was a real game changer.
The system correctly identified more than 95% of previous food packaging contents, an excellent result that will enable brands to meet the sorting standards required to meet the strict requirements of food safety authorities.
AI-assisted sorting
Since these initial tests, TOMRA and NEXTLOOPP have conducted a series of groundbreaking tests using TOMRA’s near-infrared, visual spectrometry system AUTOSORT in combination with their latest deep-learning technology GAINnext to demonstrate how deep learning, a subset of AI, could use markers replacing it with sturdy packaging and completely solving the food grade PP sorting problem.
In the latest large-scale tests, AUTOSORT with GAINnext sorted five tonnes of mixed PP plastic packaging per hour and exceeded 97 percent food-safe contents in the sorted output.
This development is an invaluable boost for NEXTLOOPP, whose participants confirm that TOMRA’s new sorting system has the potential to be rolled out to all PP packaging sorting facilities, as it focuses on package design features rather than any form of additional markings.
Accelerating the next step of recycling: decontamination
By offering a sorted food-grade PP-PCR stream, AUTOSORT with GAINnext can now accelerate the delivery of food-grade rPP via the NEXTLOOPP decontamination process in many more recycling operations worldwide, without further delays associated with new label or marker requirements.
As a result, this breakthrough will positively impact the production of valuable food-grade PP-PCR streams.
Transition from markings to AI
Less than twelve months ago, NEXTLOOPP’s focus was on package design guidelines to facilitate sorting packages into fractions of a single polymer using markers.
Now TOMRA’s latest innovation has turned this element of the design guidelines on its head. Rather than a system that relies on labels with specific markings, the AI system’s neural network is trained to identify a wide range of shapes and packaging attributes. Through structured training, the company learns to distinguish food contact from non-food contact packaging.
Design for deep learning
The next step is to revise the current design guidelines to take into account how the AI ’thinks’ to continuously improve the capabilities of both GAINnext and other existing sorting solutions. The proposed changes to packaging will certainly be simpler and more cost-effective than relying on labels and markings. The more stereotypical the packaging, the better.
The principles with which GAINnext recognizes a packet are based on object recognition. By segmenting a range of different package design factors, the AI collects the different triggers to build a contextual memory of each package shown.
Using the analogy of road signs, with the iconic stop sign being internationally recognized, the AI is trained on the shapes, sizes, dimensions and other criteria of food packaging that are commonly encountered. Transparency, opacity, printing, shapes and colors alert the system designed to strive for accurate recognition of the sorted PP packaging.
Next generation packaging
The more stereotypical the pack shape, including easily identifiable features, the higher the degree of identification. Since PP food trays are predominantly unpigmented or white and rectangular, they are easy to remove.
However, ice cream cups, which are often plain white, are likely to be rejected from the food packaging stream because they can be identified as non-food dishwasher capsule packaging. This is where design features and deeper learning can promote proper recovery.
This brings us back to NEXTLOOPP’s original suggestion to use color or design features to indicate whether a package belongs in the food or non-food category. Using color or design features to identify a package’s past use would increase the AI’s ability to define the package’s destination.
By making packaging as easy to identify as possible, brand owners now have the opportunity to communicate the recycling identity of their packaging without having to rely on a label. This has huge implications for data collection during sorting and opens up the opportunity for comprehensive producer responsibility reporting and recovery of individual packaging flows for brand owners through new PRF configurations.
GAINnext’s training is such that even in a scenario where the pack is crushed, torn or otherwise damaged, it can pick up enough points of differentiation to make an effective sorting decision in a split second.
This AI system quickly builds a wide range of signals to identify and differentiate packaging, providing an ideal opportunity for brand owners to adapt their packaging to the way AI ‘thinks’.
This breakthrough in fast and accurate AI sorting, exemplified by TOMRA’s AUTOSORT with GAINnext Deep Learning technology, is already having a significant impact on the recycling industry. As traditional markers become redundant for rigid packaging, sustainability and circularity will soon be effortlessly integrated into packaging design, delivering both environmental and economic benefits.
If you liked this story, you might also like:
The ultimate guide to the Packaging and Packaging Waste Regulation in 2024
How are the top brands making progress in packaging sustainability?
Sustainable Innovation Report 2024: current trends and future priorities
Everything you need to know about global plastic sustainability regulations
