DeepTrash @ MakeHarvard 2019

5 min readFeb 6, 2019

Last weekend I was lucky enough to attend MakeHarvard 2019, a 24-hour makeathon hosted in Cambridge, MA. My team built DeepTrash, a smart trashcan that sorts your waste for you. We had a ton of fun and ended up winning best overall. Here’s a little bit about the project and my experience at the event.

Thanks

First, I’d like to give huge thanks to the MakeHarvard team for a really awesome event! The good food, cool hardware, and helpful mentors made it a really special place to work. Thanks a ton for the great experience. It was well worth the flight from Florida!

The problem

My team wanted to take on a very challenging problem and see what we could come up with. We were confident that we had a good mix of talent and wanted something to really push ourselves. After mulling over a few ideas, we settled on trying to tackle the world’s excessive waste problem.

Here’s more on the issue:

Failing to handle waste properly is loss to our health, our economy, and our environment. These are just a few of the ways it harms us and our planet:* Unrecycled products release harmful chemicals and greenhouse gasses polluting our soil and air* Habitat destruction is furthered instead of reusing the materials we already have* Huge amounts of energy are wasted by mining raw materials instead of using recycled onesIn spite of these harmful effects and inefficiencies, Americans only recycle about 30% of their waste, leaving 180 million pounds to decompose in landfills and pollute our oceans. [1] At current rates, we are completely unsustainable.When polled about their low recycling rate, 9 out of 10 Americans said that they would recycle if it were 'easier'. [2] Clearly, there needs to be a better way for Americans to dispose of their waste.

The Idea

To solve this need for easier recycling, our team proposed DeepTrash, a robotic trashcan that sorts your trash for you! DeepTrash revolutionizes the centuries-old trashcan with advanced new technologies.

The product is enabled by deep learning models and robotic components that quickly sort compost, recyclables, and trash into separate bins within one main trashcan. The user simply tosses an item into the can and it’s automatically sorted!

Here’s a more specific breakdown of how DeepTrash works:

(1) The user tosses the waste item into the top of the can and the item is quickly funneled down to a sorting platform via gravity.(2) Here, the detection occurs. The camera feeds the video image of the item to a laptop. The laptop then predicts the type of waste using a convolutional neural network.(3) Based on this prediction, the laptop signals the Arduino into action. The Arduino first rotates the bin and then turns the sorting platform dumping the item into its compartment.

The Team

I was super lucky to work with 3 very talented individuals on this project. Here’s a little bit more about them and the great contributions they made.

Bill Chen is a sophomore at Columbia University studying computer science. Bill brought the idea for DeepTrash to the table and was the overall architect for the project. During the hackathon, he helped keep everything on track while making key design decisions for the product. Although he’s traditionally a coder, he did an amazing job creating the beautiful aluminum chassis of the can. Some might compare his design to an Apple product, but I think that would be selling him short.
Grace Hu is a sophomore at Stanford University studying materials science. Grace was a big help in so many different areas. In addition to making key design decisions, she helped precisely sculpt the product’s walls (she’s better than a laser cutter), put together and design the trashcan’s main bin, and labeled over 1,500 training examples for the convolutional neural network. In addition to all that, she also led the team to a great victory with her world-class presentational abilities on the final day.
Raziq Mohideen is a senior computer engineering student at Cargenie Mellon University. Raziq’s expertise with circuitry, hardware design, and low-level programming was invaluable to the team. He was wholly responsible for the precise wiring and tuning of the Arduino and its components. Because of his work, the physical and programmatic pieces of the project were able to seamlessly work together.

What I learned

On top of having a ton of fun at the event, I learned a lot in both technical and non-technical domains.

On the technical side, I got to delve into transfer learning, a deep learning technique that allows a model to get better results with fewer data, and apply it to a real-world problem. In addition to that, I gained some hands-on engineering experience when I helped build pieces of the can. It was really cool to work with all the different materials and combine them to accomplish a structural goal.

On the non-technical side, I sharpened my ability to work under pressure with a team (albeit a very good one) and got more valuable experience working with people face-to-face, rather than through my computer screens.

Conclusion

All in all, MakeHarvard 2019 was a really awesome event. I learned a ton, made some really cool friends, and got to work on a high-impact project. Thanks again to all the organizers and sponsors that made it possible. Hope to see you again in 2020!