Can artificial intelligence solve a key problem in the high-quality recycling of household waste? With WERTIS-KI, an interdisciplinary project team is working on an AI-supported solution to an everyday challenge of the circular economy: the correct disposal of household items and packaging. “The disposal and recovery options available to me as a citizen depend on the respective municipality or city, its contracts and the existing facilities, and are often not intuitive for citizens to understand,” explains Lisa Klatt from Ostfalia University of Applied Sciences, who is responsible for coordinating the project.
According to Klatt, there are also numerous region-specific options when it comes to higher-value R-strategies beyond recycling, such as repair or reuse. What is lacking, she says, is low-threshold real-time feedback that can provide a local, correct and fast solution at the moment of disposal and that also works independently of language. Taken together, all these factors create a highly complex and inconsistent logic of disposal recommendations and options across regions. The WERTIS-KI project therefore aims to create an easy-to-use solution that enables citizens to dispose of their recyclables correctly.
The idea of developing a simple app for a complex circular economy problem came from Lisa Klatt’s research colleagues, who were working on closing new material loops. “Imagine a company specializing in a certain type of plastic promoting an additional collection scheme for that plastic, for example from Bobby Cars,” Klatt explains. “People who want to dispose of exactly that object or material type could then be informed and reached in a targeted way.” In the best-case scenario, locally specific recovery routes could even be communicated directly when certain objects are identified.
To make this possible, a solution is needed that delivers up-to-date disposal information directly and individually to citizens. That is exactly what the new app is intended to do.
To implement the project, the researchers at Ostfalia University of Applied Sciences brought experienced partners on board: dida Datenschmiede, ge-T GmbH and the waste management company A+B Peine.
As a technology company, ge-T GmbH is responsible for developing the app. Dida Datenschmiede, in turn, handles all matters related to AI, including the selection and development of AI models responsible for object recognition and data processing. “Throughout the entire project, the waste management company A+B Peine has been our partner for all questions relating to local waste recovery and the applicable regulations. Together with A+B Peine, we at Ostfalia University compiled the data set used to train the AI models,” says Klatt. In addition to data collection, the university has taken on project management and coordination and has worked intensively on developing a classification scheme for household items.
The technical and practical expertise of the project partners created a solid basis for developing the app. However, as the project progressed, it became clear that unexpected hurdles also had to be overcome on the way to implementation. One of these lay in the app’s original concept, which was initially based solely on object recognition. “As the project progressed, it became clear that much of the information influencing the disposal route cannot be captured in an image. One obvious example is the fill level of an item,” the project coordinator recalls. A full paint bucket has to be disposed of differently from an empty one.
To “feed” the artificial intelligence, the team developed a special app that can be used to systematically collect data on household items. With the help of this data collection app, it was possible to record photos of the items, barcodes, symbols indicating disposal routes, product names, manufacturers and additional information. “This allowed us to assign the items directly to the project’s internal classification, which creates an adaptable link to local recovery routes,” Klatt explains.
At first glance, that sounds simple enough. But the large number of different individual regulations governing waste disposal in Germany did not make life easy for the developers. These rules had to be translated into an adaptive classification while also reflecting the full spectrum between what is “sensible” from a recycling perspective and strictly rule-based decisions. “What would make technical or ecological sense does not always correspond to what contracts, manufacturer agreements or legal requirements stipulate,” Klatt emphasizes, adding: “These regulatory frameworks are often formulated for a very broad field of application and fall short when it comes to detail – and those are exactly the details we wanted to make visible and manageable with our app-based solution.”
As an illustrative example, she cites plastic nets such as those used for citrus fruits at Christmas time or for Christmas trees. From the perspective of manufacturers and Germany’s dual systems, they are classified as packaging and should therefore be disposed of in the yellow bag or the recyclable waste bin. “In practice, however, they create major problems for sorting facilities: on conveyor belts that can stretch for several kilometres, the nets often get caught in the machinery and can cause malfunctions or even damage,” reports the project coordinator. “With the app, we can now provide guidance indicating that such nets should ideally be disposed of differently.”
This places the project in an area of tension: legally, the classification may be clear, but technically and operationally it is problematic. “For our classification scheme, this means we have to map such special cases while at the same time taking formal requirements into account – a challenge that is both legal and technical,” says Klatt.
With its AI-based app, the project addresses key challenges in waste recycling: high complexity, great diversity and constant change, driven in part by product design and marketing. “A system capable of processing enormous volumes of data for sorting and recognition and of learning continuously would certainly be very helpful,” Klatt says. However, it is not yet clear when the WERTIS app will be able to make a contribution, because the project did not receive an extension and the app could therefore not yet be completed to market maturity. “However, we are currently working on follow-up projects that can build on and continue the work of this project,” says Klatt.
Author: Alexander Stark, Editor FACHPACK360°
