Markov Decision Process

Discrete-time stochastic control process
Agent traverses system by estimating state-action value function, Q(s,a)

Model-free and model-based systems both converge to optimal solution in a stationary environment (i.e. reward contingencies and state-transition probabilities do not change with time)