Q-Learning and SARSA
Monte Carlo and TD Learning only evaluated a fixed policy. This page turns TD prediction into control — algorithms that learn to act well, with no model of the environment — and the two most fundamental such algorithms differ by exactly one term in their update rule.