OpenDriveLab/FreeTacMan

FreeTacman是一个支持机器人视觉触觉数据收集系统研究的大规模、高精度数据集。它包含超过3000k的视觉触觉图像对和超过10k的轨迹，涵盖50个任务。数据集结构包括视频文件和轨迹文件：视频文件为MP4格式，包含手腕安装摄像头和视觉触觉摄像头的同步视频记录；轨迹文件包含工具中心点姿态和夹持器距离的详细跟踪数据。该数据集可用于系统复制、多模态模仿学习、触觉感知抓取和模拟到现实的转移等多种研究方向。

FreeTacman is a large-scale, high-precision dataset supporting research on robot visuo-tactile data collection systems. It contains over 3000k visuo-tactile image pairs and more than 10k trajectories across 50 tasks. The dataset structure includes video files and trajectory files: video files are in MP4 format, containing synchronized video recordings from wrist-mounted and visuo-tactile cameras; trajectory files contain detailed tracking data for tool center point pose and gripper distance. This dataset enables diverse research directions such as system reproduction, multimodal imitation learning, tactile-aware grasping, and simulation-to-real transfer.