Bin There, Sorted That! A Recycle Waste Sorter using AI

CWSF · 2026 Digital Technology

Thumbnail supplied by the source for Bin There, Sorted That! A Recycle Waste Sorter using AI

Overview

Improper waste sorting is a common but critical issue, often leading to recycling contamination and increased landfill use. With the introduction of Ontario’s producer-led recycling system in 2026, individuals and public spaces now play a greater role in correctly separating waste at the source. To address this challenge, the Recycle Waste Sorter (RWS) was developed as an autonomous system capable of identifying and sorting waste using artificial intelligence, machine learning, and mechatronics. The RWS uses a camera and a trained image-recognition model to classify items, while a microcontroller coordinates motors and sensors to physically sort them into the correct bins. By automating this process, the system reduces human error, improves sorting accuracy, and decreases the burden on recycling facilities. This project demonstrates how accessible, low-cost technology can support communities in reducing contamination, improving recycling efficiency, and advancing more sustainable waste management practices.

Video

Video

Hi, I am Ani,

and I am Tristan, and we have built this Recycle-Waste-Sorter.

Human error in sorting waste is a significant factor in recycle-contamination and waste going to landfill.

With Ontario’s blue-box program transitioning to a producer-led recycling system from Jan-2026, the residents have the crucial responsibility to sort their waste into acceptable categories before the collection day.

The Recycle-Waste-Sorter identifies teracycle and recycle waste and sorts it using AI and machine-learning.

The IR-sensor triggers opening of the main sorter door, allowing user to put the object into a trolley, powered by stepper motor and belt drive.

The onboard camera images the object, and the microcontroller compares it with the YOLOv8 image-library, generated using AI-based machine-learning.

The controller determines the right bin and actuates the trolley and door motors through driver HATs to drop the item in the bin.

We performed numerous tests to determine the accuracy of sorting. We also investigated methods to improve the prototype and expedite machine-learning.

It was a fun project giving us valuable insights into recycling challenges and using mechatronics to address those.

We hope our Recycle-Waste-Sorter will be used in our community someday soon.

Thank you and see you at the fair.

Why?

In January 2026, the Municipality of Kincardine transitioned to Ontario's new producer-led recycling system, introducing new rules on how residents, schools, and public spaces must separate their waste prior to disposal. Human errors in sorting waste is a significant factor in recycling contamination, resulting in excessive waste being directed to landfills. Connections to these issues are also drawn through Model UN, where global challenges such as pollution, sustainability, and climate action are regularly examined.

Unsorted waste is a key issue that needs to be addressed at its source before it becomes uncontrollable. Conventional and emerging technologies can help build a solution to sort waste while complying with the rules and regulations. This inspired the creation of the Recycle Waste Sorter (RWS) for use in the local community.

The objectives of the project were to:

Accurately and reliably categorize waste through the use of Machine Learning and Artificial Intelligence (AI)

Build an autonomous machine to sort the waste items into designated bins

While large waste management firms in Canada continue to deploy commercial-scale automatic sorters, the residences, schools, cafeterias, and public spaces where the waste originates, can certainly use small scale sorters. Such systems will help people to follow the new standards, keep recycling streams clean, reduce waste going to landfills, and make sorting easier for everyone, even those unfamiliar with the rules.

How?

The Recycle Waste Sorter (RWS) was developed through a series of phases including conceptualization, prototype testing, coding, fabrication, and assembly. Know-how about the fundamentals of building and working of various components was developed through consultation with experts, video resources and online articles. A conceptual layout of the RWS was first created. The individual components were then designed and selected based on the following criteria.

1. Microcontroller:

Raspberry Pi 4 was the chosen microprocessor due to cost, additional GPIO pins, Python compatibility, and AI potential.

2. Motors and Actuation:

MG995 servo to open the waste intake door.

NEMA-17 stepper motor to positionally move the sorting trolley.

DS3218 servo to open the trolley-to-bin drop-door.

Adafruit and servo motor control HATs.

Geared GT2 belt drive for the trolley due to non-slip characteristics.

The motor torque and belt tension carrying capacity were checked against the trolley’s intended maximum load.

3. Sensors and Camera:

IR sensor to detect items.

Raspberry Pi camera V2 to recognize waste items, selected for its capabilities and low cost.

All custom parts, including the trolley, door mechanisms, motor housings, camera holder, and IR sensor holder were drawn in Autodesk Fusion 360 and 3D-printed using biodegradable polylactic acid (PLA) material. The RWS frame was built using lightweight pine wood for durability and ease of transportation. Sub-assemblies were tested with sample codes to refine and build the final assembly.

Using Python code and the camera, 630 images of fourteen types of terracycle and recycling items were captured with different orientations inside the trolley. The images were then annotated using Roboflow and a custom dataset was built in Google Collab for use in the main code. Each component was tested independently before full integration and repeated trials were conducted to evaluate detection accuracy and sorting reliability.

What?

The RWS was successfully developed to autonomously identify and sort waste into the appropriate disposal bins. Operation begins when a user waves an item in front of an IR sensor, which signals the Raspberry Pi microprocessor to open the intake door. The user places the item into the trolley at the home position, after which the intake door closes. A Pi Camera then captures images of the item, which are then analyzed using a custom trained YOLOv8 object detection model and compared against the project dataset. Based on the classification result, the Raspberry Pi commands a stepper motor to move the trolley to the correct bin position. A servo-actuated drop door releases the item into the selected bin, and the trolley returns to the home for the next sorting cycle. The sequential coding logic provides safety interlocks, ensuring both intake and drop doors remain closed while the trolley is in motion.

Iterative testing identified mechanical limitations that were systematically addressed throughout the development. Initial trials revealed excessive trolley mass and friction, reducing traverse speed and increasing motor load. These issues were mitigated using lightweight 3D-printed PLA components, improving rail alignment, and utilizing low friction ball rollers. Belt instability during motion was corrected through an adjustable tensioning system to eliminate abrupt trolley motion and inconsistent positioning. Vibration and noise generated by motor operation and frame resonance were reduced through reinforcement of mounting points and improved alignment of rotating shafts. The intake and drop door mechanisms were refined through testing and redesign of the motor housing. The camera’s field-of-view limitations at close range were resolved by redesigning the camera mounting bracket, improving image framing and object centering during classification.

Electrical testing revealed stepper motor grinding, skipped steps, and inconsistent motion, which were primarily caused by insufficient current delivery and incorrect driver tuning. Applying rated current values and optimizing voltage settings significantly improved torque consistency, positioning accuracy, and overall system reliability. Voltage limitations of the motor control HAT restricted the use of higher speed motors, establishing a practical upper limit for trolley speed in the current prototype. Wiring layout and connector stability were improved to enhance durability during repeated cycles.

Machine learning validation required extensive dataset refinement to achieve reliable classification performance. Over 630 training images were captured and annotated across three waste bin categories to create the dataset. Two validation trials were conducted. In the known item validation trial, 280 tests were performed using items included in creating the dataset. The model achieved an overall classification accuracy of 83.6% with a Macro F1 score of 82.4% indicating balanced performance across all bin categories. A second generalization trial was conducted, consisting of 435 tests using new waste items with similar characteristics. This test evaluated classification of unfamiliar objects, with performance influenced by item shape, color, orientation, lighting conditions, shadows, and visual similarity. The results of this test yielded 44% overall accuracy.

Integrated system testing confirmed reliable coordination between item sensing, image classification, and actuation subsystems across repeated autonomous sorting trials.

So What?

The results of this project show both the potential, and current limitations of AI-based waste sorting. In Trial-1, the system achieved 83.6% accuracy and a comparable Macro F1 score, demonstrating that the RWS can reliably classify and sort items it has been trained on. This confirms that a low-cost system using machine learning, sensors, and mechanical actuation can successfully perform autonomous waste sorting in controlled conditions.

However, Trial-2 revealed a significant drop in performance, with accuracy decreasing to 44% when new, unfamiliar items were introduced. This gap highlights that the model is currently overfitted to the training dataset and struggles to generalize to real-world waste, where item appearance, shape, and packaging vary widely. It also showed that high confidence predictions do not always mean correct classification.

These findings are important because they reflect a real challenge in deploying AI systems outside controlled environments. For the RWS to be effective in homes, schools, or public spaces, it must handle a wide variety of waste items, not just those it has seen before. This makes improving dataset diversity, image quality, and training methods essential.

The project also reinforced that performance depends on the integration of software, electronics and mechanical components. Improvements to trolley friction, belt alignment, motor tuning, and camera positioning, all contributed to more reliable operation.

With the planned improvements, including an expanded dataset, the RWS has strong potential to evolve into a practical system for small-scale use that can reduce recycling contamination and support real-life waste management goal.

What's Next?

The RWS developed in this project is a flexible machine for sorting various waste types at their source of generation.

The following are ongoing improvements to ensure the RWS performs effectively outside of a controlled environment:

Expanding the image dataset and conducting additional trials

Using sturdier screw drive to reduce noise

Introducing an iris door mechanism

Optimizing overall size

Increasing trolley speed

Future iterations would include:

Investigating spectroscopy/ capacitive/ inductive sensors for material identification

Implementing weight sensors to differentiate terracycle waste

Sort multiple items

Upcoming trials will deploy the RWS at the local municipality office and community center as well.

Thanks

We would like to sincerely thank everyone who supported the development of this project.

We are especially grateful to Devkinandan Tokekar for providing materials, assisting with design and construction of the Recycle Waste Sorter, and offering technical guidance throughout the build process.

Thank you to Anji Cottrill and Maisie Cottrill for helping revise our written content, improve formality, and strengthen the presentation of our project board.

We also thank Lorenzo Demarni for reviewing our code, improving our imaging process, and providing valuable technical feedback.

We appreciate the support of our teachers, Mr. Morris, Ms. Shipp, and Ms. Chalmers, for reviewing our work and offering helpful guidance.

Finally, we thank the Bluewater Regional Science and Technology Committee, especially Diane Wall and Heather Christie for organizing the trip to Edmonton and supporting our preparation for CWSF.

We are deeply grateful to our families for their encouragement, patience, and support throughout the project.

References

Journal articles:

[1]     Habib, H., Siddiqui, S., & Xu, C. (2023). A first comprehensive estimate of electronic waste in Canada. Journal of Hazardous Materials, 443, Article 130189. https://www.sciencedirect.com/science/article/abs/pii/S0304389423001474

[2]     Roy, J. W., & Bickerton, G. (2021). Organic contaminants of emerging concern in leachate of historic municipal landfills. Science of the Total Environment, 765, Article 142785. https://www.sciencedirect.com/science/article/pii/S026974912100052X

[3]     Masoumi, H., Safavi, S. M., & Khani, Z. (2012). Identification and classification of plastic resins using near infrared reflectance spectroscopy. World Academy of Science, Engineering and Technology International Journal of Mechanical and Mechatronics Engineering, 6(5), 3.

[4]     Core Electronics. (2023, February 16). Setting up and using the Adafruit 16-channel servo HAT for Raspberry Pi. https://core-electronics.com.au/guides/servo-hat-raspberry-pi/

[5]     Core Electronics. (2025, February 2). How to use stepper motors and DC motors with a Raspberry Pi: Adafruit DC & stepper motor HAT. https://core-electronics.com.au/guides/raspberry-pi/raspberry-pi-dc-stepper-motor-guide/

Images:

[6]     Environment and Climate Change Canada. (2024). Solid waste diversion and disposal [Figure]. Government of Canada. https://www.canada.ca/en/environment-climate-change/services/environmental-indicators/solid-waste-diversion-disposal.html

[7]     Flaticon. (2026, April 21). Landfill [Icon]. https://www.flaticon.com

Webpages:

[8]     CBC News. (2018, April 9). Many Canadians are recycling wrong, and it's costing us millions. https://www.cbc.ca/news/science/recycling-contamination-1.4606893

[9]     CBC Radio. (2019, May 2). Why your recycling may not actually get recycled. https://cbc.ca/amp/1.5099103

[10]   Cleantech Group. (2025, September 23). AI in waste sortation: Robotics, algorithms, and more. https://cleantech.com/ai-in-waste-sortation-robotics-algorithms-more/

[11]   Environment and Climate Change Canada. (2024). Solid waste diversion and disposal. Government of Canada. https://www.canada.ca/en/environment-climate-change/services/environmental-indicators/solid-waste-diversion-disposal.html

[12]   Global News. (2019, March 28). Toronto recycling: Contamination in blue bins on the rise. https://globalnews.ca/news/5104582/toronto-recycling-blue-bin-contamination/

[13]   Government of Ontario. (2024). Landfill sites map. Ontario.ca. https://www.ontario.ca/page/landfill-sites-map

[14]   Library of Parliament. (2019). Global marine plastic pollution: Sources, solutions and Canada's role (Publication No. 2019-37E). Parliament of Canada. https://publications.gc.ca/collections/collection_2020/bdp-lop/bp/YM32-2-2019-37-eng.pdf

[15]   Machinex. (n.d.). Waste and recycling equipment manufacturer. https://www.machinex.com

[16]   Made in CA. (2026, January 22). Waste management statistics in Canada. https://madeinca.ca/waste-management-statistics-canada/

[17]   Made in CA. (2026, January 26). Recycling statistics in Canada. https://madeinca.ca/recycling-canada-statistics/

[18]   Mongabay. (2023, April 20). Mining may contribute to deforestation more than previously thought, report says. https://news.mongabay.com/2023/04/mining-may-contribute-to-deforestation-more-than-previously-thought-report-says/

[19]   Oceana Canada. (2022, July 6). Canada's plastic problem: Sorting fact from fiction. https://oceana.ca/en/blog/canadas-plastic-problem-sorting-fact-fiction/

[20]   Recycle BC. (2020). What is contamination? https://recyclebc.ca/what-is-contamination/

[21]   Recycleye. (2024, June 21). How AI robots help reduce the cost of waste sorting in MRFs. https://recycleye.com/how-ai-robots-reduce-cost-waste-sorting/

[22]   Statistics Canada. (2024, April 8). Biennial waste management survey: Waste diversion, 2022 [Daily release]. https://www150.statcan.gc.ca/n1/daily-quotidien/240408/dq240408b-eng.htm

[23]   Statistics Canada. (2025, June 5). More plastic diverted from landfills in Canada, but waste and pollution remain high. https://www.statcan.gc.ca/o1/en/plus/8174-more-plastic-diverted-landfills-canada-waste-and-pollution-remain-high

[24]   TOMRA. (n.d.). GAINnext™: Waste sorting machine. https://www.tomra.com/gainnext

[25]   Mattley, Y., & Guenther, D. (n.d.). Spectroscopy for plastics recycling. Ocean Optics. https://www.oceanoptics.com/blog/spectroscopy-for-plastics-recycling/

[26]   Visit Kincardine. (n.d.). Davidson Centre. https://visitkincardine.ca/profile/davidson-centre/2112/

[27]   Municipality of Kincardine. (n.d.). Contact us. https://www.kincardine.ca/our-services/contactus/

[28]   Municipality of Kincardine. (2026). Recycling. https://www.kincardine.ca/our-services/garbage-and-recycling/recycling/

[29]   Circular Materials. (2026). An enhanced recycling system is here! https://www.circularmaterials.ca/recycleontario/

[30]   Circular Materials. (2026). Kincardine recycling. https://www.circularmaterials.ca/resident-communities/kincardine/

[31]   Mitra Electronics. (2026, February 7). IR sensor tutorial: How to use an IR sensor with a Raspberry Pi 4 [Video]. YouTube. https://youtu.be/g-K3pkV_dys

[32]   Core Electronics. (2026, February 7). Controlling DC and Stepper Motors With A Raspberry Pi - How to use Adafruit DC & Stepper Motor HAT [Video]. YouTube. https://youtu.be/ea6tSppgZlY

[33]   FREEDOM TECH. (2026, February 13). custom Instance segmentation | raspberry pi camera | yolov8 Object Detection Instance segmentation [Video]. YouTube. https://youtu.be/fcObSMuCsaw

[34]   Engineer3D. (2026, March 14). Designing Ball and Socket Joints | Print in Place | Fusion 360 | 3D Printing [Video]. YouTube. https://youtu.be/9vAkUrdSFvo

[35]   Core Electronics. (2026, February 7). How to use an Adafruit 16-Channel PWM HAT with a Raspberry Pi to Control Sixteen Servos [Video]. YouTube. https://youtu.be/bB-xymRI8BY

[36]   TechWithDavid. (n.d.). Raspberry Pi 4 Setup - Software, Hardware, Remote Access [Video]. YouTube. https://youtu.be/E9JQdX9GFwE

[37]   Google. (2026, February 15). Google Colaboratory [Cloud computing environment]. https://colab.research.google.com

[38]   Roboflow. (2026, February 14). RPI cam sort main [Dataset]. Roboflow Universe. https://universe.roboflow.com/project-9todr/rpi-cam-sort-main-fr

[39]   Reynoso, M. (2025, March 20). Inductive vs. capacitive proximity sensors. MISUMI Mech Lab. https://us.misumi-ec.com/blog/inductive-capacitive-proximity-sensors/

[40]   GFL Environmental. (n.d.). Learn more about our company. GFL Environmental

[41]   Government of British Columbia. (n.d.). Regional districts in B.C. British Columbia Regional Districts

[42]   Metro Vancouver. (n.d.). Solid waste management plan. Metro Vancouver Solid Waste

[43]   Waste Connections of Canada. (n.d.). Company overview. Waste Connections Canada

[44]   Wikipedia. (2024, March 15). GFL Environmental. GFL Wikipedia

[45]   WM (Waste Management). (n.d.). WM Canada: Waste management & recycling services. WM Canada

[46]   WM Northwest. (n.d.). British Columbia services. WM British Columbia

[47]   Mindee. (2024, April 8). How to use confidence scores with machine learning models. Retrieved April 29, 2026, from https://www.mindee.com/blog/how-use-confidence-scores-ml-models

Images (25)

Awards (2)

  • Special Award
  • Selected for CWSF 2026

Competition history

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

Source: ProjectBoard / Youth Science Canada

Save projects to your library

Sign in with Google to keep track of projects you find interesting, organized into folders. An account also raises your daily allowance for “Has this been done?”, and lets you create a key for the MCP server with a much higher limit than anonymous use. Browsing stays public.

Continue with Google