Control Tasks for Partially Observed Dynamical Systems

该数据集包含了一系列涉及部分可观测动态系统控制的任务，包括线性化倒立摆、其非线性版本、小车摆、车辆横向动力学、摆动机器人以及高维电力系统等多个控制任务。实验结果展示了在收敛状态下得到的奖励和样本轨迹，这证明了所提出方法相较于常规RNN控制器的性能优势。该数据集规模覆盖了多个复杂度不同的任务，主要任务是控制部分可观测的动态系统。

This dataset comprises a series of tasks focused on the control of partially observable dynamic systems, including multiple control tasks such as the linearized inverted pendulum, its nonlinear variant, the cart-pole system, vehicle lateral dynamics, pendulum robots, and high-dimensional power systems. Experimental results present the rewards and sample trajectories obtained under convergent states, which verifies the performance superiority of the proposed method over conventional RNN controllers. This dataset covers tasks with diverse complexity levels, and its primary task is to control partially observable dynamic systems.