Markov decision process

An MDP is the mathematical setup behind reinforcement learning. It defines states, actions, the probability of moving between states after each action, and rewards. The 'Markov' property says the future depends only on the current state. The goal is a policy that maximizes long-run reward.