Driving a car in the most convoluted possible way (Q-learning with tensorflow in python)
The results of this weren't amazing. To be fair, it was my first attempt as Q-Learning, so I can't expect too much. The car successfully completed a lap a couple of times, but usually failed between the first and second turns, and was very slow. I expect a continuous reward function and a penalty for time would improve training, but I think I'll move on to some simpler Q-Learning examples before coming back to this one.