Markov decision process
Also: MDP, MDPs
A tuple (S, A, P, R, γ) in which P(s′ | s, a) gives the probability of the next state and R the reward. The Markov property says the next state depends only on the current state and action, not on the earlier history.
MDP = (S, A, P(s′ | s, a), R(s, a), γ)
Example
In a slippery gridworld, choosing "up" moves up with probability 0.8 and sideways with 0.1 each; with γ = 0.9 a +1 reward three steps away is worth 0.9³ ≈ 0.73 now.