Custom implimentation of Q-Learning to try and balance a custom-built rotary inverted pendulum. Algorithm successfully tested on OpenAI's cartpole enviroment. On the real system we were unable to converge to the optimal solution.
The report is available in French.