9-DoF Real-World Dexterous Manipulation Task

该数据集涉及一个需要9自由度“DClaw”机械手完成旋转阀门180度的操控任务，其评估依赖于人类操作者的偏好查询。该任务采用原始图像观察，并依赖于基于偏好的学习方式，无需手动设计奖励函数。该规模为单一任务，具有特定的操控目标，属于基于视觉的机器人操控任务。

This dataset centers on a manipulation task where a 9-degree-of-freedom "DClaw" robotic hand is tasked with rotating a valve by 180 degrees. Its evaluation relies on preference queries from human operators. The task employs raw visual observations and adopts preference-based learning, eliminating the need for manually engineered reward functions. As a single-task dataset with a specific manipulation goal, it falls into the category of vision-based robotic manipulation tasks.