Skip to content
Reinforcement Learning

The RL Loop

Markov Decision Processes

9 min+85 XPStep 1/10

Formalizing the loop

An MDP packages the RL loop into five pieces:

  • S — the set of states the environment can be in
  • A — the actions available to the agent
  • P — transition probabilities: taking action a in state s lands you in state s′ with what chance?
  • R — the reward function
  • γ — the discount factor: how much future rewards count

Nearly all of RL is an attempt to solve an MDP — usually without ever being told P or R.