cleanrl

Why we included this project

CleanRL is the rare RL library that prioritizes readability over abstraction. Every algorithm variant, including PPO, DQN, SAC, TD3, C51, and PPG, lives in a single standalone Python file, so you can trace the whole implementation without digging through subclass hierarchies. That makes it a strong reference for anyone learning deep reinforcement learning and a convenient base for prototyping custom research ideas. It also includes the practical scaffolding researchers expect, such as seeding for reproducible runs, TensorBoard and Weights & Biases logging, gameplay video capture, and published benchmark results across dozens of environments. Just know it is not built to be imported as a dependency; treat it as a study guide and experimentation kit.