For my first of two capstones, my team of 4 was tasked with creating a Stewart Platform for Unsupervised Machine Learning. To be put plainly, we were asked to create a machine that would solve a marble maze using Neural Networks and / or machine learning. The goal of this project was to explore the clear challenges behind training a neural network on physical stimuli (Position, Velocity, and Acceleration), while giving the machine a chance to reset itself whenever an error was encountered. This means that, in theory, the machine can run unsupervised for weeks, gradually improving its interactions with the physical world, in spite of the challenges listed above.
Physical Data in (photos) -> Mechanical Control of Physical Object (Marble Acceleration)
I was responsible for all mechanical systems, as well as some electrical and back-end programming.