Load Optimization Assistance Device

Inspiration

Not all hackers wear capes - but not all capes get washed correctly. Dorming on a college campus the summer before our senior year of high school, we realized how difficult it was to decipher laundry tags and determine the correct settings to use while juggling a busy schedule and challenging classes. We decided to try Google's up and coming AutoML Vision API Beta to detect and classify laundry tags, to save headaches, washing cycles, and the world.

What it does

L.O.A.D identifies the standardized care symbols on tags, considers the recommended washing settings for each item of clothing, clusters similar items into loads, and suggests care settings that optimize loading efficiency and prevent unnecessary wear and tear.

How we built it

We took reference photos of hundreds of laundry tags (from our fellow hackers!) to train a Google AutoML Vision model. After trial and error and many camera modules, we built an Android app that allows the user to scan tags and fetch results from the model via a call to the Google Cloud API.

Challenges we ran into

Acquiring a sufficiently sized training image dataset was especially challenging. While we had a sizable pool of laundry tags available here at PennApps, our reference images only represent a small portion of the vast variety of care symbols. As a proof of concept, we focused on identifying six of the most common care symbols we saw. We originally planned to utilize the Android Things platform, but issues with image quality and processing power limited our scanning accuracy. Fortunately, the similarities between Android Things and Android allowed us to shift gears quickly and remain on track.

Accomplishments that we're proud of

We knew that we would have to painstakingly acquire enough reference images to train a Google AutoML Vision model with crowd-sourced data, but we didn't anticipate just how awkward asking to take pictures of laundry tags could be. We can proudly say that this has been an uniquely interesting experience. We managed to build our demo platform entirely out of salvaged sponsor swag.

What we learned

As high school students with little experience in machine learning, Google AutoML Vision gave us a great first look into the world of AI. Working with Android and Google Cloud Platform gave us a lot of experience working in the Google ecosystem. Ironically, working to translate the care-symbols has made us fluent in laundry. Feel free to ask us any questions,

What's next for Load Optimization Assistance Device

We'd like to expand care symbol support and continue to train the machine-learned model with more data. We'd also like to move away from pure Android, and integrate the entire system into a streamlined hardware package.