MotionDecode

ChingMu 1000-Hour Embodied Motion Dataset 是一个面向具身智能、 humanoid机器人和虚拟生产的高精度光学动作捕捉数据集，包含超过1000小时、120Hz采集频率的真实人体运动数据，覆盖15+真实场景、500+标准化任务以及200+带六自由度姿态跟踪的物体。数据集提供全身骨骼、手指关节、物体6D位姿、多视角视频和语义标签等多模态数据，经过自动清洗与人工质检，并完成机器人动作重定向，可直接用于人形机器人模仿学习、灵巧操作训练、运动生成、仿真验证和机器人控制策略研究。ChingMu 1000-Hour Embodied Motion Dataset is a high-precision optical motion capture dataset designed for embodied AI, humanoid robots, and virtual production. It contains over 1,000 hours of real-world human motion data captured at 120Hz, covering 15+ real-world scenarios, 500+ standardized tasks, and 200+ tracked objects with 6D pose annotations. The dataset provides multi-modal data including full-body skeletons, finger articulation, object 6D poses, multi-view videos, and semantic labels. All data is cleaned, quality-assessed, and robot-retargeted, making it suitable for imitation learning, dexterous manipulation, motion generation, simulation validation, and humanoid robot control research.