Hastings (Belleville) and Frontenac County are among the only Ontario counties that require non-urban individuals to sort their recycling waste and garbage. As one of our team members recognized, one of the greatest challenges upon moving to the region was sorting through their recycling (and understanding why their garbage wasn't being collected for weeks!). We all agreed that sorting recycling is one of the most complex subjects in terms of waste sorting and management, as it is a subcategory that is worth millions but is less talked about. We embarked on the journey of creating Roll-E to help solve this issue!

What it does

Roll-E is extremely easy to use. A user would simply deposit their recycling waste into the deposit (as with any recycling bin). Roll-E, with the help of neural networks and AI, will then assess what category that object falls into and use robotics (Arduino robots, motors and 3D-printed gears) to position it to sort into that waste bin, ready for easy collection and disposal.

How I built it

There are many individual parts and engineering that went into developing Roll-E, however after a successful build the tools themselves are inexpensive and easy to work with: Roll-E is physically built using MD4 Plywood and 3D printed flipper parts and bins. It also features a webcam. To manage and recognize each object, Roll-E uses a camera and passes images into a custom TensorFlow neural network to assess object properties and classifications to break down what pile it should go into. It then uses a Python script to execute this recognition and call the necessary Arduino sketches to toss the object into that bin and reset to prior bin structure. Arduino sketches and stepper motors allow Roll-E to have custom command sets that create a pathway to each container and disposal bin with accurate angling and reset, while also adding an animate feel to Roll-E.

Challenges I ran into

There were many major challenges that arose from creating Roll-E since our team was beginners in the field of hardware as well as neural networking. To list a few snags however:

  • Coming up with an efficient and compact design that would house 3 different component waste types.
  • Setting up our neural network to distinguish between three different categories of recycle waste.
  • Setting up our Arduino and hardware base, as well as configuring and setting up stepper motors to work/function correctly with the right motor shield.
  • Training our model to pick up and identify recycle waste attributes to categorize.
  • Getting the wiring to work correctly for our motors and rigs, including figuring out how to reverse the motor to rotate backwards.

Accomplishments that I'm proud of

In short, our greatest achievement would be to have created our desired product and end result considering our aforementioned challenges. We were able to build the product and tool as we had envisioned it and that alone is a feat that makes us proud as beginners with neural networks, hardware, and robotics. Furthermore, we are very proud of our design because of its use of cheap components, lesser pieces and its uniqueness in that it is capable of handling and sorting many different forms of trash and recycling (not binary).

What I learned

As a team, we learnt many things working on Roll-E. Roll-E was a passion-based project that we all wanted to commit to, despite the fact that NONE of us had any experience with hardware or a hardware hack and had to learn Arduino, TensorFlow and almost all tools and technologies involved from scratch. Some of us came from a web-development or geomatics background and learnt to use hardware, Arduino boards, sensors, and motors as well as 3D architecture and building to create the physical version of Roll-E. Others had programming experience but learned neural networks from scratch and terminal commands.

What's next for Roll-E

Roll-E's flipper based "tossing" waste management system is especially unique since it is very flexible and easily extendable. We specifically see ourselves adding more waste recognition (using neural networks), adding new bins (food, electronics, etc.) and using Arduino to have more flippers and command sets so that Roll-E can toss trash back and forth to new bins for more collection items. Furthermore, while Roll-E is a sustainable and affordable AI waste collector, we would love to see how to scale it to large scale while cutting costs down for parts.

Built With

Share this project: