Modernizing Waste Management through AI


One of the most challenging aspects of managing any municipality or community is managing the waste generated by consumers, businesses, and the public sector. This waste accounts for 230 million tons of trash each year, according to the Annenberg Foundation. While specific efforts to reduce the use of packaging material may ultimately reduce the amount of garbage produced over time, all communities have produced and will continue to produce waste. This is defined as a combination of materials that can be recycled, materials that are destined for landfills, and materials that must be disposed of carefully due to their contents.

Managing this waste traditionally was a largely manual process, but artificial intelligence (AI) is starting to be utilized in some communities to remove much of the labor, and therefore costs, involved with managing and processing waste. By incorporating a variety of technologies, including machine learning (ML), deep learning (DL), and computer vision, a number of solutions have emerged that are likely to improve the efficiency and productivity of waste management.

Waste Sorting

One way AI is being utilized is by training waste sorting robots that can be used at garbage dumps. Rather than needing to have workers sort through garbage, these autonomous robots are trained using ML algorithms to identify and process waste based on the type of garbage. The algorithms are trained on images of various types of waste. Using computer vision, the robots can sort through waste and match garbage based on specific characteristics, much in the same way humans might compare pieces of garbage. Most importantly, the machines will continue to learn over time and are more efficient than humans.

For example, SamurAI, a robotic innovation developed by Machinex showcased at the 2018 Waste Expo conference, can use AI to recognize recyclables such as cartons, plastic bottles, and containers. The robot then uses a suction cup to pick up the item and place it into the correct bin. After the robot identifies the object, it uses a suction cup to pick it up and place it into the correct bin. According to the company, SamurAI can perform up to 70 picks every minute, twice as productive as humans. Lakeshore Recycling Systems, a large recycling company in Illinois, installed SamurAI in its Forest View recycling plant. It does the work of two humans and saves the company up to about $130,000 a year.

Intelligent Trash Bins

Intelligent trash bins are fitted with computer vision sensors to identify the type of garbage being thrown inside them. For example, a system developed by Bin.e uses an ML algorithm to train the system to identify and categorize the type of trash being thrown away, and then the waste is sorted into bins by type. As such, all of the sorting is carried out as the waste is being disposed, eliminating the need to sort through large piles of waste at the waste processing center.

Further, the system can sense when the trash bin is full, allowing the collection schedule to be optimized. Instead of sending collection trucks out on a predefined schedule, when bins are partially filled, collection routes can be optimized to only visit the locations where bins are filled. Such optimization improves collection speed, lowers human labor costs, and reduces fuel costs.

Waste Stream Sorting

Ideally, waste would be sorted at the point of collection, as in the example above. But some waste cannot be separated by the consumer, and AI technology is being developed to handle this downstream waste sorting task.

In 2019, TOMRA unveiled its GAIN technology, which is a DL-based sorting technology designed for accurate, high throughput sorting tasks. The GAIN technology uses DL to remove PE silicone cartridges from a polyethylene (PE) stream via the use of computer vision information. Separation of silicon from the cartridges is required to ensure the purity of the waste stream.

The GAIN technology is fed thousands of images of waste types, and the system employs DL to learn how to connect the artificial neurons to classify objects. It can thus more quickly identify different types of silicon cartridges, double cartridges, and even deformed or partially destroyed cartridges without explicitly being trained on images depicting these objects. According to TOMRA, the GAIN system has achieved an overall ejection of 99% of the cartridges using two systems in a sequence.

Improving Efficiency and Reducing Human Labor Costs

The overall goal with this AI use case is to address waste management at all phases: at the initial point of waste disposal, during waste collection, and at waste processing plants. By using ML, DL, and computer vision technology, waste collection and management processes can be made more efficient and effective, and the human labor costs required to accomplish these tasks can be reduced.

These waste management use cases are just a part of the new AI use cases that are likely to power the development of smart cities. This is the subject of Tractica’s forthcoming report, Artificial Intelligence for Smart City Applications, which is projected to be released by 1Q 2020.

Comments are closed.