Q-Learning is an off-policy reinforcement learning algorithm that learns an action-value function (Q-function). It updates Q-values by sampling transitions and using the Bellman optimality equation. Q-Learning can eventually learn the optimal policy, even if actions are selected randomly.