DEXHAND-70K/dexhand-70k-review-sample

DexHand-70K是一个用于研究高自由度灵巧操作扩展规律的多模态数据集。它包含70,000个通过16自由度LEAP手进行的遥操作抓取演示，同步采集了RGB（4个摄像头）、触觉力（51维）和本体感觉（16维）数据，采样频率为30 Hz。数据集涵盖了101个对象×5个环境×10个光照条件的全因素设计。该数据集支持三个研究问题：高自由度动作空间中的数据扩展规律、视觉与触觉互补性、以及视觉域扩展。数据集还提供了详细的元数据、分割方案和验证方法，适用于机器人学、强化学习和模仿学习等领域。

DexHand-70K is a multi-modal dataset for studying scaling laws in high-DOF dexterous manipulation. It contains 70,000 teleoperated grasping demonstrations with a 16-DOF LEAP Hand, synchronized RGB (4 cameras), tactile forces (51-dim), and proprioception (16-dim) at 30 Hz, collected across 101 objects × 5 environments × 10 lighting conditions under a controlled factorial domain design. The dataset enables three research questions: data scaling laws in high-DOF action spaces, vision-tactile complementarity, and visual domain scaling. It includes detailed metadata, splits, and verification methods, making it suitable for robotics, reinforcement learning, and imitation learning.