Gymnasium is an open-source library providing an API for reinforcement learning environments. Its main contribution is a central abstraction for wide interoperability between benchmark environments and training algorithms. Gymnasium comes with various built-in environments and utilities to simplify researchers' work along with being supported by most training libraries. This paper outlines the main design decisions for Gymnasium, its key features, and the differences to alternative APIs.

翻译：Gymnasium 是一个为强化学习环境提供 API 的开源库。其主要贡献在于提供了一个核心抽象，以实现基准测试环境与训练算法之间的广泛互操作性。Gymnasium 内置了多种环境和实用工具，以简化研究人员的工作，并且得到了大多数训练库的支持。本文概述了 Gymnasium 的主要设计决策、关键特性以及与其他替代 API 的差异。