Reinforcement learning is a type of learning in which an agent often a computer learns by receiving feedback in the form of rewards or punishments. It is based on the idea that rewards encourage behaviour that lead to positive outcomes, while punishments discourage behaviour that lead to negative outcomes.