Neural network controler.
Recently on my collage class we had a lesson on pid controllers. This subject would be nice and interesting, but unfortunately the hardware we were forced to use was very old PLC devices (BTW. which had been not working correctly back then). And our lecturer told us that it's impossible to make controller based on machine learning with behaviour better than PID. Since PID does not guarantee optimal control and it's pretty easy to reproduce discrete PID with simple neural network. After short analysis of the problem, I taught that it might be great experience to try that. Also, since the learning data can vary lot, the training data can be created using simple simulations.
This is pid-like regulator designed to work with embedded systems. This project uses c++14, and tinny-dnn, as machine learning library (library was chosen based on need to support embedded systems).