In reinforcement learning, an agent changes its states to maximize its goals. There are four distinct concepts here: agent, state, action, and reward. Let's take a look at these in more detail:

• Agent: This is the program we train. It chooses actions over time from its action space within the environment for a specified task.
• State: This is the observation that's received by the agent from its environment and represents the agent's current situation.
• Action: This is a choice that's made by an agent from its action space. The action changes the state of the agent.
• Reward: This is the resultant feedback regarding the agent's action and describes how the agent ought to behave.

Each of these concepts has been illustrated in the following diagram:

As shown in the preceding diagram, reinforcement learning involves an agent, an environment, a set of actions, a set of states, and a reward system. The agent interacts with the environment and modifies its state. Based on this modification, it gets rewards or penalties for its input. The goal of the agent is to maximize the reward over time.