The Open Data Network
for Physical AI.
物理世界智能的
数据基础设施。
A centralized + decentralized global collection network, with a compliant open marketplace for every dataset. UMI, ego-centric, real-robot, multimodal — licensed, structured, and training-ready. 中心化品控 × 去中心化全球采集网络,配套开放的数据集交易与合规授权市场。UMI、第一视角、真机遥操作、多模态数据——结构化、合规可交付、开箱即进训练。
Four things no one else puts together.四件别人凑不齐的事。
Hybrid Network Architecture混合网络架构
Not a single factory. Not fragmented crowdsourcing. We run a centralized quality backbone layered on top of a globally distributed collection network — enterprise-grade consistency at decentralized scale.不是单一工厂,也不是碎片化众包。我们在全球分布式采集网络之上,统一运行一套中心化品控主干——规模化产能,企业级质量,两者同时实现。
- Multi-node collection across the US, Mexico, China, and Pakistan美国、墨西哥、中国、巴基斯坦多节点并行采集
- 24/7 teleoperation pool across time zones24/7 跨时区遥操员池,无时差缺口
- Unified AI + human QA pipeline on every dataset每条数据统一经过 AI 预审 + 人工复审 + IoU 一致性校验
The First Open Data Marketplace for Embodied AI首个开放的具身智能数据交易市场
Every dataset on our network can be discovered, licensed, and purchased — directly. No custom procurement cycles. No opaque pricing.网络内所有数据集均可被检索、授权、直接采购。无需漫长的定制采购流程,无不透明报价。
- Browse by data type, task, embodiment, scene, or modality按数据类型、任务、机器人本体、场景或模态自由筛选
- One-time license, volume subscription, or exclusive rights一次性 License、按量订阅、独占授权——多种采购方式
- Clear IP chain on every record — compliant by design每条记录权属链清晰,合规从数据生产端内建
Compliance Built In, Not Bolted On合规是产品能力,不是附加条款
Enterprise buyers can't afford data liability. Every dataset ships with a full compliance package.企业买方承受不起数据权属风险。平台上的每个数据集,出厂即附完整合规包。
- US-domestic collection: ITAR-aware, CCPA-structured美国本土采集符合 ITAR / CCPA 要求
- Standardized licensing with tiered use-case coverage标准化 License 协议,按用途分层(训练 / 商业部署 / 衍生模型)
- Training, deployment, and derivative model rights — defined upfront合规尽调材料随数据集一起交付
Training-Ready Across All Modalities全模态,训练就绪
Raw footage is not training data. Ours is.原始录像不是训练数据,我们的是。
- Structured trajectories, not video files结构化轨迹数据,而非视频文件
- Multi-sensor alignment: RGB + depth + tactile + proprioception + languageRGB + 深度 + 触觉 + 本体感知 + 语言指令,多传感器同步对齐
- Standard formats: LeRobot · RLDS · Open X-Embodiment · HDF5标准交付格式:LeRobot · RLDS · Open X-Embodiment · HDF5
Five data types. One network.五类数据,一张网络。
| Type类型 | Description说明 | Scale规模 |
|---|---|---|
| UMIUniversal Manipulation Interface通用操作接口 | Human demonstrations via proprietary collection gloves, haptic + slow-motion capture.自研采集手套采集的人类操作示范,含触觉 + 慢动作轨迹。 | 200k+ trajectories轨迹 |
| Ego-centric第一视角First-Person | Human-perspective operation recordings with 6DoF trajectory alignment.人类视角操作录像 + 6DoF 轨迹对齐。 | 200k+ episodes |
| Real-Robot Teleop真机遥操作Cross-embodiment | Cross-embodiment real-machine trajectories across 100+ robot platforms.100+ 机器人本体跨本体轨迹数据。 | 10M+ episodes |
| Multimodal Fusion多模态融合 | RGB + depth + tactile + proprioception + language instructions, multi-sensor aligned.RGB + 深度 + 触觉 + 本体感知 + 语言指令,多传感器同步对齐。 | All types全类型覆盖 |
| Sim-to-Real Paired仿真配对 | Omniverse / Isaac Sim simulation paired with real-machine data for sim-to-real transfer.Omniverse / Isaac Sim 仿真与真机配对,支持 sim-to-real 训练。 | On demand按需定制 |
Tasks: 1,700+ atomic tasks · 5,000+ object types · 9 material classes · 12 core skill categories 场景覆盖:家庭 · 办公 · 文旅 · 商超零售 · 餐饮 · 工业
任务量:1700+ 原子任务 · 5000+ 物品类别 · 9 类材质 · 12 类核心技能
An order of magnitude larger than the biggest public benchmarks.比最大的公开数据集大一个数量级。
10× the size of the largest open corpus. Cross-embodiment, not single-fleet.比最大公开数据集大 10 倍。Cross-embodiment,而非单一本体堆量。
License it, commission it, or build it with us.采购、定制、或共建。
Ready to stop building your data pipeline from scratch?不需要再从零搭数据采集体系。
Talk to our team. We'll walk you through the marketplace, share a 100-episode sample pack, and map your roadmap to the right data type and engagement model.联系我们的团队。我们会带你浏览数据市场、提供 100 条样本数据供当晚测试,并根据你的训练 Roadmap 推荐最匹配的数据类型和合作方式。