# sail-sg/envpool

C++-based high-performance parallel environment execution engine (vectorized env) for general RL environments.

Repository: https://github.com/sail-sg/envpool
Canonical: https://ross.abutalabs.com/products/envpool
Homepage: https://envpool.readthedocs.io
Language: C++
License: Apache-2.0
License Family: permissive
Topics: reinforcement-learning, parallel-processing, cpp17, pybind11, reinforcement-learning-environments, threadpool, atari-games, vizdoom, gym, high-performance-computing, lock-free-queue, mujoco, box2d, dm-control, robotics, dm-env
Last push: 2026-08-26T19:20:36+00:00

## Health v2 (maintenance only)
Score: 94/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 84, longevity 100
- inputs: {"age_days": 1778, "days_push": 7, "days_rel": 105, "gap_med": 3.5, "n_releases_24m": 11}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1506, forks 141 (observed 2026-08-28T04:04:55.003325+00:00)

## What it is
EnvPool is a C++-based batched environment pool with pybind11 bindings and a thread pool for high-performance parallel RL environment execution. It provides Gymnasium and dm_env compatible APIs and ships prebuilt implementations of many popular environments like Atari, MuJoCo, and DeepMind Control Suite.

## Use cases
- run reinforcement learning environments at millions of frames per second
- vectorize atari game environments for RL training
- parallelize mujoco physics simulations for RL
- batch dm_control suite environments with async stepping
- speed up RL training data collection pipelines
- replace gym vector envs with a faster C++ backend

## When to choose
- you need maximum environment throughput for RL training
- you want prebuilt, optimized implementations of common RL benchmarks like Atari or MuJoCo
- you need Gymnasium or dm_env compatible APIs with sync and async stepping

## When to avoid
- you need a custom environment not among the supported list and don't want to write C++
- you only run a handful of environments and don't need vectorization
- you need a pure-Python solution for easy modification of environment logic

## Facets
- artifact type: library
- maturity: active
- function: simulation, machine-learning, concurrency, sdk
- domain: reinforcement-learning, machine-learning, simulation, robotics
- platform: python, cpp, cross-platform
- tags: reinforcement-learning-environments, vectorized-environments, gymnasium, dm-env, pybind11, thread-pool, atari, mujoco, high-performance, game-development, linux, macos

## Member repositories
- sail-sg/envpool (main) score 94

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:55.003325+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:32:40.821378+00:00, confidence not recorded.
  - readme: https://github.com/sail-sg/envpool (fetched 2026-08-28T04:04:55.003325+00:00, sha 28f5eab7ecb5)
- Data as of 2026-08-30T08:39:29.467469+00:00.