Environment

Topics: Reinforcement Learning

Definition

The environment is the world in which the agent live and interacts. At every step, the agent can see the full or partial state of the environment, take an action and receive a reward, which is a number that tells the agent how good or bad the action was given the current state.

A fully observable environment is one where all the states are observable.
A partially observable environment is one where some states, not all, are observable.

States & Observations

• State: full representation of the environment
• Observation: partial representation of the environment
  They are represented by real-valued vectors, matrices, or high-order tensors.

References

1. https://spinningup.openai.com/en/latest/spinningup/rl_intro.html#