Stable-Baselines3

A PyTorch-based library of classic reinforcement learning algorithms, with a simple API and reliable implementations.

Stable-Baselines3 is an open-source, PyTorch-based reinforcement learning library maintained by Antonin Raffin and colleagues at the German Aerospace Center's Robotics Institute (DLR-RM). It's the successor to Stable Baselines, itself derived from OpenAI Baselines, with the associated paper published in JMLR in 2021. It provides reliable implementations of algorithms including PPO, SAC, TD3, DQN, and A2C — training a policy on a Gymnasium environment can take just a few lines of code — with thorough documentation and tests, and it's often used as a starting point or a reference baseline. It mainly targets a single environment or a small number of parallel ones; for large-scale GPU-parallel simulation training, people more often reach for rsl_rl or rl_games instead.

Examplemodel = PPO(“MlpPolicy”, env); model.learn(100000) — a few lines of code are enough to train a policy on Gymnasium's cart-pole task.

Also called

SB3

Related

Reinforcement Learning · Proximal Policy Optimization · Soft Actor-Critic · Gymnasium · CleanRL · skrl

Sources

Stable-Baselines3 Docs
DLR-RM/stable-baselines3 (GitHub)