A basic Deep Reinforcement Learning agent and environment for driving in a 4 lane highway made with keras and Pygame.
You can run the spardriver_env.py and play the game by arrow keys without the agent to get a sense of what's going on.
It is possible to train the agent based on pre existing network so the previous data will not be thrown away.
The dqn_model.h5 file has a network trained for 10 hours and its performance in the test area is relatively well. Also there is a backup of that file.