# rlenvlab — The RL Environments Lab

> A curated, link-checked directory of reinforcement learning environments.
> Find the right env for your project, compare landmark benchmarks, and learn RL.

## What this site covers

- **Environment directory** (https://rlenvlab.com/#directory): 51 RL environments across
  8 categories — classic control, Atari & retro, robotics, multi-agent, board & card games,
  navigation, procedural, and real-world simulators. Every repo link verified live (Oct 2026);
  maintenance status checked against real commit history. Filterable by category, difficulty,
  framework, and maintenance status.
- **Env picker** (https://rlenvlab.com/#picker): answer 4 questions (goal, experience, compute,
  preference) and get a matched shortlist of environments with reasons.
- **Benchmarks board** (https://rlenvlab.com/#benchmarks): 10 landmark results that defined
  the field — Agent57's 57/57 Atari, Rainbow, AlphaGo, AlphaZero, AlphaStar, OpenAI Five,
  plus classic solve thresholds (CartPole, LunarLander).
- **Learn track** (https://rlenvlab.com/#learn): 12 canonical RL resources — courses
  (David Silver's UCL lectures, Berkeley CS285, HuggingFace Deep RL Course), books
  (Sutton & Barto), tutorials (Karpathy's Pong, CleanRL), libraries (SB3, RLlib), key papers (DQN).
- **FAQ** (https://rlenvlab.com/#faq): best env for beginners (CartPole-v1), standard
  benchmarks (ALE, MuJoCo, Procgen), what is maintained in 2026 (Farama Foundation suites,
  Isaac Lab, MuJoCo Playground), Gymnasium vs PettingZoo.
- **Changelog** (https://rlenvlab.com/#changelog): monthly ecosystem notes.
- **Blog** (https://blog.rlenvlab.com/): long-form posts on the RL environments ecosystem.

## Key facts (verified Oct 2026)

- Best RL environment for beginners: CartPole-v1 (Gymnasium).
- Standard deep RL benchmarks: Arcade Learning Environment (57 Atari games), MuJoCo
  locomotion suite, DeepMind Control Suite, Procgen (generalization).
- Actively maintained suites: Gymnasium, PettingZoo, MiniGrid, ALE (Farama Foundation);
  Isaac Lab (NVIDIA); MuJoCo Playground (Google DeepMind).
- Archived/unmaintained classics: OpenAI Retro, DeepMind Lab, NetHack Learning Environment,
  SMAC, RLCard, PySC2, BabyAI (envs moved into MiniGrid), Crafter, MiniHack, Neural MMO.
- Microsoft AirSim is still actively maintained (commits as of Sept 2026).
- Gymnasium = standard API for single-agent RL. PettingZoo = multi-agent counterpart.

## Contact

rlenvlab.com is for sale — inquiries via the footer link on the site.