All Tools
E
AgentsFreeOpen Source
ENVPOOL
High-performance C++ vectorized environments for reinforcement learning
Apache-2.0
ABOUT
Python subprocess vector envs waste CPU on serialization and process overhead, so large-scale RL training is bottlenecked by environment steps. EnvPool runs a C++ thread-pool of batched envs with Gymnasium-compatible APIs so PPO and SAC trainers collect millions of frames per second without rewriting the algorithm.
INSTALL
pip install envpoolINTEGRATION GUIDE
1. Speed up Atari PPO training by replacing gym.vector with EnvPool batched envs
2. Run MuJoCo or DeepMind Control rollouts at millions of steps per second on CPU
3. Plug EnvPool into Stable-Baselines3, CleanRL, Tianshou, or ACME trainers
4. Add a custom C++ env to the pool when Python wrappers are too slow
TAGS
pythoncppreinforcement-learningvectorizedgymnasiumatarimujoco