Waste management companies in the USA have implemented robot in recycling plants designed to increase the quantities of waste sorted and the quality of the selection. They recognize the different categories of household waste using artificial intelligence.
And they work 1.5 times more efficiently than a human operator. However, they must be skillfully adjusted in advance.
Such robot works hard on the assembly line. This new employee at sorting centers receiving tons of waste per year (all household waste streams combined, except glass) has a mission: to increase the volumes of waste sorted and, above all, the quality of sorting. This robot, included in the research budget of the junk disposal companies, is driven by artificial intelligence. A first in Florida, in the field of household waste.
They have to be irreproachable. China has finally closed its borders to secondary raw materials that are of insufficient quality in its eyes, recalls the General Manager of such a Recycling and Waste Treatment activity. Since its launch, the robot’s mission has been to sort the bulk of the store, a flow of small papers and small cardboard boxes, and to eliminate cans, bottles, plastics, and even shoes that may have escaped previous sorting, using a trommel and an overband. For the time being, it has an error rate of 10%.
A recycling operator remains present for quality control
At the end of the chain, an operator remains present to help the recycling robot at a sustained rate: he performs quality control, like most of his colleagues who work in a semi-automated factory (including sorting remotely using a tablet). They take on additional tonnages with the same number of operators, so they secure employment. The missions are evolving: they have to improve working conditions, as the sorter’s job is not the easiest.
Visual learning of waste elements
To graft a robot using artificial intelligence onto an existing sorting chain, the recycling robot first had to adapt to the specific sorting instructions and packaging in force. Hundreds of thousands of color images were added to its database. The use of tele-operated sorting, already underway in some factories, made the operation easier.
The robot can be controlled from a console if necessary. The images are pre-recorded.
Then, the algorithm had to be perfected: they work with deep neural networks (deep learning) and supervised learning. They have to differentiate PET bottles from non-PET bottles, explains the director of recycling and waste recovery research programs at the Panama City sorting center. Finally, the teams added a dashboard to monitor the activity of the chain, and integrated a system (a suction spider). The waste research center also has a copy of the robot for different batteries of tests in real conditions.
Other recycling flows in preparation
Ultimately, the robot will process different flows, including plastics brought by dumpster rentals– the bulk of the work is only the first step in its implementation. It is already proving to be quite efficient, carrying out 3,600 waste picks per hour, compared to 2,200 for a human sorting operator. In 2024, a new unit will accommodate two robots for the over-sorting of clear PET (bottles, bottles) and the sorting of a flow mainly composed of polyethylene. Other robots, without artificial intelligence, are already deployed in the group.
To reduce the volume of waste stored in Florida by 50% in 2025, the recycling sector estimates the necessary investments at 4 billion dollars and, beyond robotization, at 25,000 additional jobs.