Hands-On Reinforcement Learning for Games

SARSA lambda and the Lunar Lander

As the algorithms we develop get more complicated, their capabilities also get more powerful. However, there are limits and it is important to understand the limits of any technology. To test those limits, we want to look at an example that pushes them. For this particular case, we will look at the Lunar Lander environment from Gym. This environment is modeled after the old classic arcade game of the same name, where the object is to land a lunar module on the surface of the moon. In this environment, the observation space is described in eight dimensions and the action space in four. As we will see, this can quickly go beyond our current computational limits.

The LunarLander environment requires the installation ...