DEXHAND-70K/DEXHAND-70K

DexHand-70K是一个用于研究高自由度灵巧操作中缩放定律的多模态数据集。它包含了70,000个16-DOF LEAP手的遥控抓取演示，同步记录了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个物体×5个环境×10个光照条件的组合，设计了一个控制因子域的实验。数据集支持三个研究问题：高自由度动作空间中的数据缩放定律、视觉-触觉互补性以及视觉域缩放。数据集的总大小为2.09 TB，存储为gzip压缩的HDF5格式，并提供了详细的元数据和索引文件以便于数据加载和过滤。

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 with RGB (4 cameras), tactile forces (51-dim), and proprioception (16-dim) at 30 Hz. The dataset covers 101 objects × 5 environments × 10 lighting conditions under a controlled factorial domain design. It enables three research questions: data scaling laws in high-DOF action spaces, vision-tactile complementarity, and visual domain scaling. The total size of the dataset is 2.09 TB, stored in gzip-compressed HDF5 format, and includes detailed metadata and index files for easy data loading and filtering.