lunar lander model #4, using PPO trained with learning rate 0.0005 for 500K timesteps

initial commit