We built it using a roomba, raspberrypi with a picamera, python, and Microsoft's Custom Vision
Challenges we ran into
Attaining wireless communications between the pi and all of our PCs was tricky. Implementing Microsoft's API was troublesome as well. A few hours before judging we exceeded the call volume for the API and were forced to create a new account to use the service. This meant re-training the AI and altering code, and loosing access to our higher accuracy recognition iterations during final tests.
Accomplishments that we're proud of
Actually getting the wireless networking to consistently. Surpassing challenges like framework indecision between IBM and Microsoft. We are proud of making things that lacked proper documentation, like the pi camera, work.
What we learned
How to use Github, train AI object recognition system using Microsoft APIs, write clean drivers