RouteAble

RouteAble Logo

Inspiration

We were inspired by fellow UMass Amherst students who identified the lack of accessibility here on campus including dangerous ramps, hidden entrances, and unsafe roads. Students have the option to wait for "Classroom Access Assistants" and Accessible Van Services provided by the university, which are not the most consistent and are a common source of frustration. In our research, we found there were no commercially available apps which offer the convenience and range that our app does. Current solutions suffer from lack of upkeep, difficult to navigate UI, and overall software pitfalls that do not best help disabled users.

Our application ensures that disabled people know ahead of time which buildings have accessible entrances and where they are -- all using crowd-sourced user images. Using machine learning, pictures of the same location but from different angles are filtered out using our object similarity model. We also use machine learning to generate tags of certain characteristics (stairs, ramps, etc). For quality control, we additionally have an edit tag function to manually edit generated tags. We also identified the challenge of incentivizing users. Hence, we developed a cryptocurrency wallet that allows users to collect tokens if their image is accepted by the threshold model.

We are excited about being part of a more inclusive future and hope that our app can help bridge structural barriers for disabled people.

What it does

RouteAble allows users to search for a specified location. These locations are stored in a database. If other users have added tags or images already, then these tags and images show up on the user's screen. Additionally, tags are auto-generated by our machine learning model in the back-end. Similar images uploaded by users are filtered out using an image similarity model. Pictures taken from different angles of the same area are not added to the database.

Our application also allows users to filter all tagged locations with specific requirements like ramps. These locations have pictures to accompany the tag categorization.

Users have the option to add tags and images for their current location only for quality control reasons. Other functionality includes filtering all tagged areas for characteristics (e.g. ramps) and searching up another address to see if any crowd-sourced data is available. This search yields tags and pictures associated with the location.

Lastly, we used Ergo Blockchain to incentivize and reward users who successfully upload a unique image.

RouteAble is accessible via Android and iOS devices.

How we built it

Machine learning: We used TensorFlow for object detection (tag generation) and PyTorch for image similarity.

App development: React Native is used for Android and iOS configurations. Map is generated using Google Maps (Android) or Apple Maps (iOS) API. In-house machine learning models hosted on Node.js.

Backend: Supabase Relational Cloud Database for image database and retrieval. NestJS Typscript framework is hosted on Linode Ubuntu VPS using Docker.

Version control: We split our project into three directories all under the RouteAble organization on GitHub: routeable-mobile, routeable-ml-backend, and routeable-backend. This ensures that our app can be scaled up and there are less conflicts in code between members. Pull requests can be more easily addressed.

Challenges we ran into

For all of us, this was our first time working with JavaScript to develop an app. We also struggled with machine learning accuracy given our data limitations, the database set-up, and crypto wallet set-up.

Accomplishments that we're proud of

We made the app! It works on Android and iOS! Yay! We also made the front-end look fairly nice considering our time constraints. We exceeded our goals including being able to make an authorization page, in-app support for several images of one location, and convenience for use on multiple platforms.

What we learned

We have now gained a newfound sense of respect for front-end developers. Making a tiny button is a lot harder than it looks.

What's next for RouteAble

We are excited about using this app to help alleviate the frustration of heading to a space and finding the entrances aren't easily accessible. We want to add image support of different types of roads (like gravel, dirt, rough roads). We also want to train the machine learning model with more balanced, versatile data.