Toy Environment for Planning Problem

该数据集是为了分析提出的Grad+CEM方法在高维动作空间中与传统CEM方法的性能而创建的玩具环境。该环境包含了具有真实梯度的动力学模型，并允许通过弹簧常数调整接触硬度。其规模涉及高达20维的高维动作空间。任务则是利用强化学习技术在一个高维动作空间中进行规划。

This dataset is a toy environment developed to analyze the performance of the proposed Grad+CEM method against the conventional Cross-Entropy Method (CEM) in high-dimensional action spaces. This environment features a dynamics model with ground-truth gradients, and allows tuning contact stiffness via spring constants. It supports high-dimensional action spaces with dimensionality up to 20. The core task of this environment is to perform planning in high-dimensional action spaces using reinforcement learning techniques.