ChingMu 1000-Hour Embodied Motion Dataset
青瞳1000小时具身智能动作数据集
High-precision optical motion capture data for humanoid robots, dexterous hands, embodied AI, and virtual production.
面向人形机器人、灵巧手、具身 AI 和虚拟制作的高精度多模态动作数据。
| Duration / 时长 | 1000+ hours @ 120 Hz |
| Scenarios / 场景 | 15+ real-world scenes |
| Tasks / 任务 | 500+ standardized tasks |
| Objects / 物体 | 1000+ tracked props (6D pose) |
| Modalities / 模态 | Skeleton · Finger · Object 6D · Video · Labels |
| Formats / 格式 | BVH · Retargeted CSV · NPZ |
🔒 Access note: Metadata and samples are public. Full data requires Request access.
🔒 访问说明： 元数据和样本公开，完整数据需申请访问权限。
Key Features / 核心特点
- Optical ground truth – sub-mm accuracy, 120 fps, no estimation errors.
光学真值 – 亚毫米精度，120 fps，无估计误差。 - Dexterous hands – 20+ DoF per hand, synchronized with object 6D pose.
灵巧手数据 – 每只手 20+ 自由度，与物体 6D 位姿同步。 - Robot-ready – pre-retargeted to Unitree G1; custom retargeting available.
机器人就绪 – 预重定向至 Unitree G1，可定制其他机器人。 - Real-world diversity – 15+ scenarios, 500+ tasks, 1000+ objects.
真实场景多样性 – 15+ 场景，500+ 任务，1000+ 物体。 - Multi-modal – full-body skeleton, finger motion, object pose, multi-view video, semantic labels.
多模态 – 全身骨骼、手指运动、物体位姿、多视角视频、语义标签。 - Quality assured – every take passes automated cleaning + manual inspection; quality flags provided.
质量保障 – 每条数据经自动化清洗和人工审核，提供质量标记。
Dataset Summary / 数据集概述
ChingMu 1000H is an optical motion capture dataset designed for training and validating embodied AI and humanoid robot controllers. It covers full-body skeleton, finger articulation, object 6D pose, multi-view video, and semantic labels across 15+ real-world scenarios (industrial, household, retail, healthcare, logistics, agriculture, performance). All data is cleaned, quality-assessed, and robot-retargeted.
青瞳 1000 小时数据集是为训练和验证具身 AI 与人形机器人控制器而构建的光学动捕数据集。它覆盖全身骨骼、手指关节、物体 6D 位姿、多视角视频和语义标签，跨越 15+ 真实场景（工业、家庭、零售、医疗、物流、农业、表演）。所有数据均经过清洗、质量评估和机器人重定向。
Data Format Specifications / 数据格式规范
| Component / 组件 | Format / 格式 | Details / 详情 |
|---|---|---|
| Raw motion / 原始动作 | .bvh |
Y-up, 120 fps, ZYX rotation, cm, 47–67 joints |
| Retargeted trajectories / 重定向轨迹 | .csv |
Root position (m), quaternion, joint angles (rad) |
| Object 6D pose / 物体 6D 位姿 | .csv |
Position (m) + quaternion, 120 Hz |
| Multi-view video / 多视角视频 | .mp4 |
4–8 cameras, co-registered |
| Semantic labels / 语义标签 | .jsonl |
Task, scenario, action, object |
📸 Sample Visualization / 样本可视化
Display is a snapshot from our motion capture studio showing a subject performing a box-moving task, with real-time skeleton overlay and object tracking:
以下是我们动捕棚的截图，展示了一名受试者执行搬箱子任务的过程，包含实时骨骼叠加和物体追踪：
Figure: Optical mocap data visualized with skeleton and tracked object.
图：光学动捕数据可视化，显示骨骼和追踪物体。
🎥 Preview Video / 预览视频
Watch a short demonstration of the motion capture data in action:
观看动捕数据的简短演示：
Demonstration of full-body motion capture with real-time skeleton overlay and object tracking.
全身动作捕捉演示，包含实时骨骼叠加和物体追踪。
Intended Uses / 预期用途
- Imitation learning / motion policy training for humanoids
- 人形机器人的模仿学习 / 运动策略训练
- Dexterous manipulation datasets (hand-object interaction)
- 灵巧操作数据集（手-物交互）
- Motion generation & retrieval (text/motion cross-modal)
- 运动生成与检索（文本/运动跨模态）
- Sim-to-real validation (MuJoCo via retargeted trajectories)
- 仿真到现实的验证（通过重定向轨迹在 MuJoCo 中验证）
- Virtual production & animation reference
- 虚拟制作与动画参考
Task & Scenario Taxonomy / 任务与场景分类
All takes are indexed in metadata/index.csv. Key filter columns:
所有数据片段均在 metadata/index.csv 中索引。关键筛选列：
| Column / 列名 | Values / 取值 | Use / 用途 |
|---|---|---|
scenario |
industrial, household, retail, healthcare, logistics, agri, performance |
Filter by scene / 按场景筛选 |
task_category |
locomotion, manipulation, dexterous_hand, tool_use, interaction |
Broad category / 大类 |
task_label |
walk_carrying_box, screw_with_driver, pinch_grasp_bottle … |
Specific task / 具体任务 |
has_finger_data |
true / false |
Needs hand DoF? / 是否需要手指数据 |
has_object_6d |
true / false |
Needs object tracking? / 是否需要物体追踪 |
quality_flag |
pass / warning / fail |
Skip bad takes / 跳过低质量数据 |
retarget_available |
g1 / none |
Robot format / 机器人格式 |
Full taxonomy includes locomotion, manipulation, dexterous hand, tool use, object interaction, social contact, and performance.
完整分类包括：移动、操作、灵巧手、工具使用、物体交互、社交接触和表演。
👀 Try the interactive Dataset Preview at the top of this page (select
metadataconfig) or downloadmetadata/index.csvfor offline filtering.
👀 尝试页面顶部的交互式数据集预览（选择metadata配置），或下载metadata/index.csv进行离线筛选。
Full Taxonomy (abridged) / 完整分类（简版）
- Locomotion → walk, jog, crouch-walk...
- 移动 → 行走、慢跑、蹲走……
- Manipulation (whole-body) → shelf-pick-place...
- 移动 → 行走、慢跑、蹲走……
- Dexterous Hand → pinch, precision-grasp...
- 灵巧手 → 捏取、精密抓握、旋盖、插拔、按键、工具握持……
- Tool Use → screwdriver, wrench...
- 工具使用 → 螺丝刀、扳手、剪刀、镊子、喷瓶……
- Object Interaction → door-open/close...
- 物体交互 → 开门/关门、抽屉、冰箱、橱柜、盖子开合……
- Social / Contact → handoff-object...、
- 社交/接触 → 传递物体、引导运动、双人交接……
- Performance → dance, martial-arts...
- 表演 → 舞蹈、武术风格动作、风格化手势……
👀 Try it live: Use the Dataset Preview panel at the top of this page to filter and explore the actual index table. Select the metadata config to browse available takes.
👀 现场体验： 使用页面顶部的数据集预览面板筛选和浏览实际索引表。选择 metadata 配置查看可用的数据片段。
ℹ️ The full index with all rows is best viewed locally. Download
metadata/index.csvto open in Excel or pandas for complete filtering.
ℹ️ 完整索引（所有行）建议在本地查看。下载metadata/index.csv后用 Excel 或 pandas 打开进行完整筛选。
Dataset Structure / 数据集结构
chingmu-1000h/
├── metadata/ ← Index files (CSV, taxonomy)
├── retargeted/ ← Robot-ready trajectories (gated)
│ └── g1_joint_trajectory/
├── raw_bvh/ ← Original BVH (gated)
├── annotations/ ← Semantic labels
├── samples/ ← Public preview (no gate)
└── LICENSE.md
🖥️ Interactive Showcase / 交互式展示网站
Visit our dedicated showcase website for interactive demos, comparison videos, and detailed visualizations:
访问我们的专属展示网站，查看交互式演示、对比视频和详细的可视化内容：
Includes: trailer video, modality breakdowns, robot retargeting comparisons, and more.
包含：预告片、模态分解、机器人重定向对比等。
Quick Start / 快速开始
pip install huggingface_hub
from huggingface_hub import hf_hub_download
repo_id = "ZIHLING/Chingmu-RobotData"
file_path = hf_hub_download(
repo_id=repo_id,
filename="samples/Move_the_box_001.bvh",
repo_type="dataset",
local_dir="./robot_samples"
)
print(f"Downloaded: {file_path}")
For full access, request permission via the Request access button, then use snapshot_download.
如需完整访问，请通过 Request access 按钮申请权限，然后使用 snapshot_download。
Quality & Limitations / 质量与局限
Quality controls: marker swap correction, gap-filling (≤6 frames), foot skating detection, manual review. Flags: pass, warning, fail.
质量控制：标记交换修正、缺帧填充（≤6帧）、足部滑动检测、人工审核。标记：pass、warning、fail。
Accuracy: joint error <1mm, object pose ±2mm / ±0.5°, temporal sync <1 frame.
精度：关节误差 <1mm，物体位姿 ±2mm / ±0.5°，时间同步 <1 帧。
Limitations: performer age skew (20–35), finger precision depends on calibration, object accuracy varies with marker cluster size.
局限：表演者年龄偏向（20-35岁），手指精度取决于校准，物体精度随标记簇大小变化。
Ethics: all performers consented; faces excluded from skeleton data; no biometric identifiers retained.
伦理：所有表演者已签署同意书；面部未包含在骨骼数据中；未保留生物特征标识。
Get Full Dataset / 获取完整数据集
Only a subset of the dataset is publicly available here. If you need full access to the entire 1000-hour dataset, please scan the QR code below to contact us via WeChat:
此处仅公开了数据集的子集。如果您需要完整访问整个 1000 小时数据集，请扫描下方二维码通过微信联系我们：
Alternatively, you can click the "Request access" button on the right side of this page to automatically gain download permissions for the complete dataset.
或者，您可以点击页面右侧的 “Request access” 按钮自动获得完整数据集的下载权限。
Note: Approval is automatic, but we kindly ask you to provide your contact information for licensing purposes.
注意：审批是自动的，但我们恳请您提供联系信息以便授权。
Or email us at: dataset@chingmu.ai
或发送邮件至：dataset@chingmu.ai
We look forward to collaborating with researchers and industry partners!
我们期待与研究人员和行业伙伴合作！
Contact / 联系方式
- Issues / Requests: Discussions
- 问题/请求：讨论区
- Email: dataset@chingmu.ai
- 邮箱：dataset@chingmu.ai
- Downloads last month
- 80