Machine Learning For Game Testing

Machine Learning should be used to test the game "Dirt 5." There should be 10 - 15 mins of continuous input where the car will try to stay on the tracks and go around the tracks. The goal is not to win the game but to simulate someone playing the game. If the car hits a barrier or another car or gets off track of the course, it correct itself and get back on track. OpenCV, TensorFlow and Keras should be used. No other deep learning languages.