Reinforcement Learning (RL) methods are well suited for control problems as they learn to determine the best action to take in a given environment to maximise a given reward. They have advantages over classical optimisation methods, such as efficiency. This talk will introduce the fundamental concepts in model-free RL, such as Q-learning and various types of agents, and review a number of successful applications in the control of particle accelerators, such as the CERN injector chain.