Foundations of Deep Reinforcement Learning: Theory and Practice in Python

14. States

Solving a new problem using deep RL involves creating an environment. We will therefore now shift our focus from algorithms to the components of environment design— which are the states, actions, rewards, and the transition function. When designing an environment, we first model the problem and then decide what information, and how, our environment should present to its users.

It is essential that an RL environment provides sufficient information to an algorithm so that it can solve a problem. This is one crucial role of states, which are the subject of this chapter.

First, we will give some examples of states both in the real world and in RL environments. In the sections that follow, we consider the following questions which are ...