Modified a TENS/EMS unit with an Arduino so that through a serial port I could control when the shock is released and the power of that shock, I then interfaced this device with different things so it would shock the user under different circumstances. Some examples implementations are a video game when the user took damage, a piano when the wrong note was played in a song, and haptic feedback in VR.
Built a Go-Kart that you could drive by pretending to drive. To explain this ridiculous idea a little bit further, the car would accelerate when it detected a specific sound which was the user mimicking an engine, and the steering was controlled by a gyroscope on the driver's glove so the user would pretend to hold a steering wheel. An ackerman steering system was designed and implemented from scratch.
The software here would watch a video game currently being played and detect when certain events happen based on visual input. Examples include player health loss, when certain abilities are used, etc
I'm currently in the process of exploring deep q-learning. The approach here is to 'record' various boxing moves such as punches and dodges, then play back these recordings randomly to train a network; rewarding it on a hit or a dodge and punishing it when it gets hit.