EcoRush

Logo
App demo
Object detection model training
Predicted map of trash 1 month later given the initial trash, wind, currents, etc.
Object detection results
App demo of Object Detection model

Inspiration

All of our parents like to recycle plastic bottles and cans to make some extra money, but we always thought it was a hassle. After joining this competition and seeing sustainability as one of the prize tracks, we realized it would be interesting to create something that makes the recycling process more engaging and incentivized on a larger scale.

What it does

We gamify recycling. People can either compete against friends to see who recycles the most, or compete against others for a prize pool given by sponsors (similar to how Kaggle competitions work). To verify if a person recycles, there's a camera section where it uses an object detection model to check if a valid bottle and recycling bin are in sight.

How we built it

We split the project into 3 major parts. The app itself, the object detection model, and another ML model that predicted how trash in a city would move so users can move with it to pick up the most amount of trash. We implemented an object detection model, where we created our own dataset of cans and bottles at PennApps with pictures around the building, and used Roboflow to create the dataset. Our app was created using Swift, and it was inspired by a previous GitHub that deployed a model of the same type as ours onto IOS. The UI was designed using Figma. The ML model that predicted the movement of trash concentration was a CNN that had a differential equation as a loss function which had better results than just the vanilla loss functions.

Challenges we ran into

None of us had coded an app before, so it was difficult doing anything with Swift. It actually took us 2 hours just to get things set up and get the build running, so this was for sure the hardest part of the project. We also ran into problems finding good datasets for both of the models, as they were either poor quality or didn't have the aspects that we wanted.

Accomplishments that we're proud of

Everyone on our team specializes in backend, so with limited initial experience in frontend, we're especially proud of the app we’ve created—it's our first time working on such a project. Integrating all the components posed significant challenges too. Getting everything to work seamlessly, including the CNN model and object detection camera within the same app, required countless attempts. Despite the challenges, we've learned a great amount throughout the process and are incredibly proud of what we've achieved so far.

What we learned

How to create an IOS app, finding datasets, integrating models into apps.

What's next for EcoRush

A possible quality change to the app would be to find a way to differentiate bottles from each other so people can't "hack" the system. We are also looking for more ways to incentivize people to recycle litter they see everyday other than with money. After all, our planet would be a whole lot greener if every citizen of Earth does just a small part!