Agent-Environment Interface

Agent: The learner and decision-maker Environment: The thing the agent interacts with

In a finite MDP, the sets of states, actions, and rewards are finite. So we have well defined discrete probability distributions.

$p$ specifies a probability distribution for each choice of $s$ and $a$, tells us how likely we are to end up in a new state $s^{\prime }$.

$p$ is decided by the environment, agent cannot modify this. ${\sum }_{s^{\prime },r}p(s^{\prime },r|s,a)=1$

Instead, the agent modifies $\pi$, the policy.