ControlGym

ControlGym是一个包含36个工业控制环境和10个基于偏微分方程（PDE）控制问题的大型数据集，旨在为强化学习（RL）算法提供基准测试环境。该数据集集成在OpenAI Gym/Gymnasium框架内，支持连续、无界的动作和观察空间，适用于真实世界的控制应用。ControlGym特别设计了高维和潜在无限维系统，以评估RL算法的可扩展性。数据集适用于探索RL算法在控制策略学习中的收敛性、基于学习的控制器的稳定性和鲁棒性以及RL算法在高维系统中的可扩展性等关键问题。

ControlGym is a large-scale dataset comprising 36 industrial control environments and 10 partial differential equation (PDE)-based control tasks, developed as benchmark environments for reinforcement learning (RL) algorithms. Integrated into the OpenAI Gym/Gymnasium framework, it supports continuous and unbounded action and observation spaces, catering to real-world control application scenarios. ControlGym is specially designed with high-dimensional and potentially infinite-dimensional systems to assess the scalability of RL algorithms. This dataset enables the exploration of critical research topics in RL-based control, including the convergence of control policy learning, the stability and robustness of learning-driven controllers, and the scalability of RL algorithms when applied to high-dimensional systems.