Distributional Reinforcement Learning (Distributional RL) is an advanced area of reinforcement learning that models not just the expected return of actions but the entire distribution of possible returns, enabling agents to better capture risk, uncertainty, and variability in sequential decision-making. Foundational works introduced the C51 algorithm, which discretizes the value distribution, followed by quantile-based methods such as QR-DQN and implicit quantile networks (IQN) that improve representation and learning of return distributions. Recent research explores extensions to continuous action spaces, multi-agent settings, risk-sensitive policies, and combinations with deep reinforcement learning architectures for tasks in robotics, games, and autonomous systems. Distributional RL has been shown to enhance stability, convergence speed, and policy robustness compared to traditional expectation-based methods, and it also facilitates more informative exploration strategies. Current studies further integrate distributional approaches with hierarchical RL, meta-RL, and model-based RL, highlighting its growing significance in developing reliable and risk-aware decision-making systems.