Ecobins

tl;dr we are enabling global sustainability straight from home through a smart trash-can that allows for accurate sorting of waste, empowers civilian education on what materials are recyclable, and helps break us out of the feedback loop fueling climate change.

Inspiration

The Earth is under an extreme crisis. Only one-third of those in the US recycle and out of everything that is recycled, only 5% of those materials actually make it through the recycling process. While we don’t pay much attention to where our waste goes once we toss it, this issue has profound consequences. Lack of recycling depletes key limited resources that we depend on, expands landfills that pollute nearby environments, and makes the Earth increasingly uninhabitable. Inspired by this resounding issue, we sought to develop a low-cost, practical solution that enables the average household to mitigate their waste management.

What it does

Ecobins is a smart trash-can with an array of features that enables users to efficiently sort their waste and play an individual part in mitigating the global climate change crisis. Most central is Ecobin’s image recognition algorithm that differentiates between recyclables and non-recyclables and informs users in real time so that they can correspondingly dispose of their waste. Ecobins uses a motion sensing lid to react to the user’s presence and allow for practical disposal of waste, making the user experience smoother. Depending on the output of the AI algorithm that classifies any given waste as recyclable or as trash, the user presses buttons that control the flow of waste so that waste is effectively sorted. This interactive process not only informs users of how to dispose of any given waste, but it also equips them with an expanded knowledge of how items should be disposed of.

How we built it

The Ecobins deep learning algorithm that classifices waste as trash or recyclable was built using TensorFlow Lite, a low computational power tool that allows rapid determination of the best way to dispose of waste. The hardware was built using Arduino Uno circuitry and the corresponding software was coded in C++.

Challenges we ran into

Technological limitations: The first challenge we faced was the limited variety of IoT hardware that could be used to embed the machine learning model onto the trash-can system. Without a functioning Raspberry Pi or an Arduino Portenta + Vision Shield, we integrated our system with a web based-app as an innovative alternative. Practical limitations: In order to keep our solution low-cost, low-resource, and hence widely accessible, we placed additional constraints on ourselves during the design process to ensure the final product would meet these visions. For example, in order to ensure that our system could accommodate low product costs that would allow this trash-can to become an everyday household item, we focused on designing innovative software that wouldn’t rely on an expansion of hardware in order to further revise the product.

Accomplishments that we're proud of

Approaching a multi-faceted issue through a multi-faceted solution: We’re proud of how we combined various components central to resolving the waste crisis to tackle this extreme issue. Not only does our platform allow for efficient waste sorting, which serves as an important smart home function, but it also allows for civilian education on proper waste disposal, changing the public view on recycling, and reduces instances of improper recycling, mitigating anthropogenic impacts on the climate crisis. An interdisciplinary approach to designing a technological solution: In order to develop our final product, we drew from software coding skills in C++, hardware construction skills, societal knowledge of the state of the recycling crisis, and more. We are happy to see how we were able to draw from different skill sets to develop a cohesive solution that is a product of interdisciplinary collaboration. Our multidisciplinary approach allowed us to come up with a more comprehensive solution that embeds essential knowledge from different fields, and we’re happy to see how these all came together in the end to support a more robust final product. Integrating interests and strengths: We are proud that our final product is a mosaic of everyone’s interests and strengths. While we integrated Sam’s strengths Arduino circuitry design, we coupled these with Selin’s interest in mitigating non-recycled waste production, and Ola’s strengths in hardware construction. When we look at our final design, we see a reflection of our own ideas and visions as well as those of our teammates’, each time being able to pinpoint how ideas were proposed and how they developed through collaboration to become a part of this final mosaic.

What we learned

Throughout this weekend, we learned how to overcome challenges by optimizing our product design path. As we faced technological limitations, we creatively brainstormed suitable alternatives that would allow us to preserve the initial project vision but reach that vision through an alternate path, such as by using a web-based app for integrating the image classification model. Additionally, even when we achieved our general vision, we still performed iterations of testing to find alternate approaches that potentially worked even better. For example, once we implemented a preliminary machine learning model, we iterated through multiple optimization rounds to revise the model and afford a more accurate prediction. Additionally, given that we were under an extreme time constraint this weekend, we learned the importance of fully thinking through ideas early on before diving headfirst into the build phase. We learned that 5 minutes of early brainstorming can save 5 hours down the road and that fully fleshing out ideas gives a stronger team vision and paves a clearer development path. We particularly experienced this when deciding on how to develop our trash-can mechanism; we realized there were many variations in how we could construct the box and assemble the various electronic parts to it. Uncertainty on how to approach this decision as we were initially beginning created some hesitation and we realized that the best course of action at that time was to thoroughly address the decision by drawing thorough product sketches before moving forward with a half-clear idea in mind. Once we discussed and came to a conclusion, we felt much more confident in development and were able to resume at a quicker pace than before, achieving a more cohesive vision at the end.

What's next for Ecobins

We plan to further revise the Ecobins model to develop a solution with real-world impact. This weekend has planted many ideas in our minds about how we can build upon this idea and further advance our solution, so we hope to pursue those visions to realize the potential for impact that this product holds. In particular, we hope to embed our machine learning algorithm onto a Raspberry Pi device equipped with a camera for real-time and portable waste sorting, without the need for a computer interface. We also hope to explore other materials that may be an appropriate fit for a trash-can, especially those that are durable, low-cost, mitigate bacterial contamination, and visually pleasing, all of which are considerations we have deemed important for such a solution. Finally, we hope to expand our machine learning algorithm to find a more robust solution that preserves the low computational power of our current model. By expanding the dataset that our machine learning algorithm is trained on, we hope to further improve the accuracy of our model’s predictions and enable a solution that provides consistent reliability for an optimal user experience. Additionally, this would allow us to make further iterations of our product without having to alter product costs, as software innovation does not incur further construction costs as compared to hardware innovation. Ultimately, we hope to turn our solution to a low-cost, practical solution that can be implemented in average households across the globe.

Built With

• arduino
• c++
• image-classification
• machine-learning