Advanced Robotics Sorting of Consumer Recyclables

How We Are Solving It

Led by Metaspectral, in partnership with Merlin Plastics and the University of British Columbia, the Advanced Robotic Sorting of Consumer Recyclables (ARSCR) is a complete automated system to sort consumer recyclables, consisting of an automated vision system powered by machine learning to detect objects on a conveyor belt and identify object material, along with a robotic system to pick selected items off the conveyor belt. Once completed, it will be the first able to detect, identify and efficiently sort materials—such as different types of plastic—that are identical to the naked eye, generating an extremely high-quality output.

The proprietary operating software will send the data back to a central computing system, allowing robotic systems in a specific sorting facility to use a single computing system to analyze the captured spectral data, lowering the unit cost of each robotic device. The vision and robotic system will be configurable through a user interface available on a display mounted next to the robotic system and require minimal operator training.

The industry will reap huge benefits from this project, as it will increase the production capacity of recycling plants by automating and accelerating repetitive tasks, grow the profitability and competitiveness of recycling plants by making them less vulnerable to fluctuations of market prices. It will result in a higher-quality finished product, while reducing the labour intensity of waste sorters and recyclers, increasing the profitability, resiliency, and growth potential of the recycling sector as a whole.

The project will include a comparative study, with the goal of benchmarking the developed technology. This research will be done in partnership with UBC, and co-led by Professors Maria Holuska and Ed Grant, and will evaluate different technologies for plastics characterization and identification in service of sorting/separation, integrating all technologies to supplement one another for quality control, and the resulting publications will be shared with consortium partners, and eventually to target customers of the finished product. The findings from the study will allow for standardization of the techniques used for precise and accurate characterization of plastics and to validate and certify the quality of the final recycled product and benchmarking of the robotic sortation process.

This joint venture, upon successful implementation of the technology put together by consortium members to market the developed technology to the recycling industry, anticipates to gradually increase its revenue to $50 million in Year 5, and $100 million within 10 years of commercialization, capturing approximately 1% of the total addressable market of waste recycling plants globally by that time. It will reduce Canada’s carbon emissions and position B.C. as a global leader in industrial-scale recycling robotics and artificial intelligence.

The Result

This project developed consumer waste sorting technology to increase the sorting efficiency of recycling plants, thereby increasing their overall capacity. By leveraging ultra-high-resolution imaging and analyzing the data in real-time, previously indistinguishable materials can be differentiated, increasing the quality of the final recycled material.

The project’s original scope was adjusted to reduce the focus of the robotic gripper production and increase the scope of Metaspectral’s vision system development beyond its initially planned capabilities to be able to detect more materials. This expanded scope additionally envisioned the ability for the vision system to be installed in more locations along the plastic recycling production line, thus increasing its ability for intervention in removing or diverting non-recyclable or improperly recycled materials on moving conveyor belts. Metaspectral was able to test their technology throughout the project in an active Merlin Plastics facility.

The project successfully automated the auditing process of incoming bales of material, which was previously a manual and sporadic process. The machine learning powered vision system is able to sort between a variety of black or non-black materials and identify materials such as glass, organics, metals, cardboard, tetra packs and more in real-time with an average 95% accuracy rate. It is also notably capable of detecting challenging PVC packaging, which even at 1 part-per-million releases toxic, chlorine-based chemicals that were otherwise passing through other sorting technologies undetected.

An independent study by Ostrom Climate confirmed that, based on the expected impact of widespread deployment of Metaspectral’s developed technology on the climate, the total environmental benefits would be 10,779 kilotonnes of CO2e avoided by 2040. The benefits are estimated at 1.44 tonne of CO2e per tonne of material recycled.