Finger-Gaiting Manipulation Dataset

该数据集旨在学习使用次优控制器进行手内操作技能的指关节控制。它包含了多种方法的结果，包括标准的非策略强化学习以及行为克隆损失版本。数据集涵盖了多种物体，如立方体、球体、圆柱体、十二面体和二十面体，针对的任务是灵巧的手内操作。This dataset is designed to learn joint control for in-hand manipulation skills using suboptimal controllers. It contains results from multiple methods, including standard off-policy reinforcement learning and versions with behavior cloning loss. The dataset covers a diverse set of objects, such as cubes, spheres, cylinders, dodecahedrons and icosahedrons, targeting the task of dexterous in-hand manipulation.