safe-control-gym

safe-control-gym是一个用于安全学习和基于学习的控制以及强化学习在机器人领域的统一基准套件。该数据集由多伦多大学航空航天研究所动力系统实验室创建，旨在支持模型基于和数据基于的控制技术。数据集包含三个动态系统：cart-pole、1D quadrotor和2D quadrotor，以及两个控制任务：稳定和轨迹跟踪。通过扩展OpenAI的Gym API，safe-control-gym允许指定和查询符号动力学、约束，并可重复注入模拟干扰，从而支持安全控制算法的发展。该数据集适用于量化比较传统控制、基于学习的控制和强化学习领域的控制性能、数据效率和安全性，旨在解决机器人控制中的安全性和效率问题。

safe-control-gym is a unified benchmark suite for safe learning, learning-based control, and reinforcement learning in the field of robotics. Developed by the Dynamic Systems Lab at the Institute of Aerospace Studies, University of Toronto, this suite is designed to support model-based and data-based control techniques. It includes three dynamic systems: cart-pole, 1D quadrotor, and 2D quadrotor, alongside two control tasks: stabilization and trajectory tracking. By extending the OpenAI Gym API, safe-control-gym allows for the specification and querying of symbolic dynamics, constraints, and repeatable injection of simulation disturbances, thus facilitating the development of safe control algorithms. This benchmark suite can be used to quantitatively compare the control performance, data efficiency, and safety of traditional control, learning-based control, and reinforcement learning methods, with the goal of addressing the safety and efficiency challenges in robotic control.