Gymnasium
An open-source reinforcement learning environment interface library that provides standardized APIs for game and simulation environments. It enables researchers worldwide to compare algorithm performance on unified benchmarks and serves as a foundational infrastructure for reinforcement learning research.
Core Facts
Timeline (last 90 days)
The developer built a Snake engine in C++17, provided Python bindings via pybind11, and wrapped it as a Gymnasium environment.
The developer has turned the commander seat into an open bot kit that includes a local practice server (running real match rules at about 100 times real-time speed without requiring a game client), a Python library using only the Python standard library, a Gymnasium-style environment with action masking, a MaskablePPO example from Stable-Baselines3, and rule-based bots and learned-policy opponents.
Most classic-control competitions on Kaggle remain compatible with Gym or Gymnasium, and beginners reproducing Notebooks need to watch for version differences, otherwise errors in unpacking env.step() return values easily occur.
After 2022, maintenance of Gym was transferred to the Farama Foundation and renamed Gymnasium, with minor incompatible API changes (such as step() returning a five-tuple instead of a four-tuple).
OpenAI Gym has been renamed to Gymnasium.
Gymnasium (formerly OpenAI Gym) provides an environment interface convention that allows simulation environments and real robots to share the same definition of observation and action spaces.
OpenAI Gym (now Gymnasium) and PettingZoo can be used to hands-on implement classic reinforcement learning environments such as mazes and CartPole.
The researchers conducted computational experiments on two types of benchmarks: the Feynman dataset (physical formula regression tasks) and Gymnasium reinforcement learning benchmarks.
MuJoCo has become one of the core simulation backends in the OpenAI Gym / Gymnasium ecosystem.
The environment follows the Gymnasium API standard; the observation space is a 13,977-dimensional float vector and incorporates the historical record of cards played by the opponent.
18 more timeline events
All Facts (20)
Gymnasium, maintained by the Farama Foundation, succeeded OpenAI Gym and defines the environment interface standard
70%UnverifiedOpenAI Gym provides a standardized environment interface, but its officially built-in environments are almost exclusively Atari games and MuJoCo physics simulations.
60%UnverifiedThe developer built a Snake engine in C++17, provided Python bindings via pybind11, and wrapped it as a Gymnasium environment.
50%UnverifiedMost classic-control competitions on Kaggle remain compatible with Gym or Gymnasium, and beginners reproducing Notebooks need to watch for version differences, otherwise errors in unpacking env.step() return values easily occur.
50%UnverifiedAfter 2022, maintenance of Gym was transferred to the Farama Foundation and renamed Gymnasium, with minor incompatible API changes (such as step() returning a five-tuple instead of a four-tuple).
50%UnverifiedOpenAI Gym has been renamed to Gymnasium.
50%UnverifiedGymnasium (formerly OpenAI Gym) provides an environment interface convention that allows simulation environments and real robots to share the same definition of observation and action spaces.
50%UnverifiedOpenAI Gym (now Gymnasium) and PettingZoo can be used to hands-on implement classic reinforcement learning environments such as mazes and CartPole.
50%UnverifiedThe researchers conducted computational experiments on two types of benchmarks: the Feynman dataset (physical formula regression tasks) and Gymnasium reinforcement learning benchmarks.
50%UnverifiedMuJoCo has become one of the core simulation backends in the OpenAI Gym / Gymnasium ecosystem.
50%UnverifiedThe environment follows the Gymnasium API standard; the observation space is a 13,977-dimensional float vector and incorporates the historical record of cards played by the opponent.
50%Unverifiedgym-pybullet-drones provides a native Gymnasium interface that is directly compatible with mainstream RL libraries such as Stable-Baselines3 and RLlib
50%UnverifiedBot-friendly reinforcement learning environments typically follow standard interface specifications similar to OpenAI Gym/Gymnasium, providing standard methods such as observation, action, reward, and step/reset
50%UnverifiedMuJoCo is the default physics backend for classic control tasks in OpenAI Gym/Gymnasium.
50%UnverifiedOpenAI launched Gym in 2016, which has since evolved into Gymnasium, now maintained by the Farama Foundation.
50%UnverifiedGym/Gymnasium defines a unified interaction interface reset(), step(), render() for reinforcement learning environments, but standardization stops at the environment interaction API and does not cover higher-level engineering processes such as training loop management and experiment configuration.
50%UnverifiedGymnasium将原版Gym中的done信号明确拆分为terminated（任务自然结束）和truncated（因时间限制截断）
50%Unverified2022年OpenAI将Gym的维护工作移交给Farama基金会，并更名为Gymnasium
50%UnverifiedGymnasium提供统一的环境接口（reset、step、render），覆盖CartPole、MountainCar、Atari游戏、MuJoCo物理仿真等环境
50%Unverified作者为《雨世界》环境添加了Gymnasium封装，使其可直接使用Stable-Baselines3中的算法
50%