Hands-On Intelligent Agents with OpenAI Gym

Actor-critic algorithm

Let's start with a diagrammatic representation of the actor-critic architecture, as shown in the following diagram:

There are two components in the actor-critic algorithm, as evident from the name and the preceding diagram. The actor is responsible for acting in the environment, which involves taking actions, given observations about the environment and based on the agent's policy. The actor can be thought of as the policy holder/maker. The critic, on the other hand, takes care of estimating the state-value, or state-action-value, or advantage-value function (depending on the variant of the actor-critic algorithm ...