Saving the world by reinventing how we manage waste
One of the greatest threats facing the planet and mankind is the accumulation of man-made waste. Furthermore, the over-extraction of finite resources from the earth is vastly unsustainable. Recycling and composting help to minimize these negative effects. But over 90% of plastic and compostable items still end up in landfills rather than recycling centers and compost heaps. This can be because people don’t have access to recycling, don’t know how to correctly recycle, or think it is too much of a hassle to use different trash bins for different wastes. We wanted to find a solution that nullifies all the potential problems of recycling and composting, allowing for 100% efficiency regardless of user error.
What it does
EcoSort is a smart waste-disposal container. It has a single entrance point for waste, and scans the waste as it is tossed in. Based on the kind of waste, the EcoSort will drop it into one of three inner-containers, corresponding to trash, recycling, and composting. This saves the user the trouble of trying (or not trying) to figure out which bin to use for which waste item, and ensures that all recyclables end up in the recycle bin and all compost ends up in the compost bin. No item of waste will end up in an incorrect bin. EcoSort will also periodically transmit data about the waste it receives, (e.g. how many items are in each bin), to a central application, which will store and analyze the data for many potential uses (e.g.. when to empty each container).
How I built it
Building EcoSort involved various hacks, ranging from hardware to software to front end to prototyping.
Our software was primarily run off of Microsoft Azure. We used Azure’s cognitive services, including machine learning with data samples (images taken by internal camera of what was deposited in the ecoSort), custom vision, image processing, classification, and predictions based on different iterations and training images of the objects.
As for hardware, we use a powerful camera with image recognition. We implemented our motor by using a battery and a conveyor belt made out of paper. As for our sensors (ultrasonic), we used RaspberryPi and connected it to the motors. An ultrasonic sensor triggers the conveyor belt based on if there is an object present.
For the front end, we built a panel for our app using react and node js, implementing everything via Firebase. As for our prototype, we used whatever we could get our hands on around the building. A table, plastic bins, duct tape, plastic forks and knives, zip ties, paper, power strips, and more.
Challenges we ran into
One of our primary challenges was creating the hardware aspect of the project, given the limited materials. Building a conveyor belt was a particularly difficult challenge, as the belt needed to be strong enough to hold the trash but flexible enough to rotate without getting stuck. In response to this problem, we used paper reinforced with duct tape, and repurposed paint rollers to rotate the belt. Additionally, creating the platform’s frame was difficult, as it had to be raised above the ground to allow the items to fall into separate bins. We used a combination of broom handles, metal stands, and cardboard boxes tied together with duct tape and zip ties. We also had to tackle the challenge of removing the items from the conveyor. We decided to use servo motors with improvised arms to push the items off.
We also ran into several challenges in terms of software. The accuracy and speed of the image processing was one of our foremost concerns, as the objects must be identified in a limited amount of time and the choice made must be final. We had to make many API requests and quickly reached the limit, so we used the cloud credits to get more requests.
Accomplishments that We are proud of
We are proud of the hardware hacks that went into building the prototype. We had to be very creative with the available materials in order to make it work, scavenging everything from toothbrushes to broom handles. We’re also proud of the image recognition component of our project. The classifications were initially tentative and quite slow, but we were able to speed them up enough to be used with a moving conveyor belt in real time.
What We learned
Every hack we conducted to make this project come to life taught us new skills. This was our first time working with Microsoft Azure, so most of our software writing was taking what we know about machine learning theory and implementing it on Azure. This extends to all the functions we utilized from Azure. We did not have that much experience with RaspberryPi’s, so equipping this ultrasonic sensor was a new experience that required some research and thought. Many of us were also exposed to React for the first time, and that was a very useful skill to pick up.
What's next for EcoSort
We have many ideas we will be pursuing with EcoSort. First, we would like to finalize the product by incorporating additional sensors and compartments to sort out each individual type of recyclable (i.e. metal, plastic, paper). We will also need to implement a different power source. One option is to attach solar roof to our outdoor bins, allowing for the bins to operate independently of the power grid. We would also like to explore the use of biofuel, making use of the composting section to power the EcoSort itself. After we finalize the design, we would distribute EcoSort to a college campus that previously used 3-4 separate bins for each type of waste. We would collect data about trash volume in our centralized client, and distribute the information to the school, so they would know when they need to pick up each EcoSort x. If this trial run works, we would attempt to work with the city to expand into the surrounding region, as defined by the nearest waste sorting center. This would alleviate the need for the sorting center, as all trash will be directly sent to a landfill, all recyclables will already be sorted into their respective types, and all compost will be sent to farms. Without need for the sorting center, the pre-sorted recyclables will make their way directly to a repurposement facility, reducing work done by trucks and thus emitting less greenhouse gasses. Once we can prove to the local government that the EcoSort saved them money, helped the environment, and made people’s lives easier, we can expand nationwide by implementing a similar deployment strategy.