Reinforcement learning is a branch of machine learning in which an agent learns to make sequences of decisions by interacting with an environment and receiving rewards or penalties. The core idea is trial and error guided by feedback. Unlike supervised learning, where correct answers are given, the agent must discover good actions by exploring and exploiting what it has already learned.

Formally, the problem is often described as a Markov decision process with states, actions, transition probabilities, and a reward function. The objective is to learn a policy that maximizes cumulative reward over time. This is typically done through value functions that estimate how good a state or state action pair is, or by directly optimizing the policy.

Classical methods include dynamic programming, which assumes a known model of the environment, and model free approaches such as Monte Carlo methods and temporal difference learning. Temporal difference methods like Q learning and SARSA update value estimates from experience without needing a model of the dynamics.

More recently, deep reinforcement learning combines these ideas with deep neural networks to handle high dimensional inputs such as images. Techniques like deep Q networks approximate value functions, while policy gradient methods and actor critic architectures directly learn parameterized policies, often with improved stability and performance.

Research also addresses exploration strategies, credit assignment over long time horizons, sample efficiency, and safety. Applications range from game playing and robotics to resource management and autonomous systems, where the ability to learn from interaction makes reinforcement learning a powerful framework for sequential decision making under uncertainty.