Skip to content
Backed byYZi Labs

From “Seeing” to “Doing”

The real-world embodied data layer for physical AI

Join Alpha
A human hand and robotic arm reach toward a three-dimensional embodied-data cube

4D Fusion Data Engine

One pipeline from raw capture to licensed models, scalable and verifiable at every stage.

  1. 01

    Capture

    A multi-device capture matrix, from phones to professional kits.

  2. 02

    Data

    On-chain rights registration and encrypted storage.

  3. 03

    Processing

    Cleaning, annotation and automated quality checks.

  4. 04

    Training

    AXON world-model and skill-model training.

  5. 05

    Distribution

    Model Store, API access and industry licensing.

Three core technology modules

Visual-tactile fusion · ViTam

Vision fused with touch, so demonstrations carry contact forces and material response, not just pixels.

Cross-embodiment mapping

Universal adaptation across robot arms, humanoids and quadrupeds, so one dataset serves many bodies.

AXON world-model engine

Physically consistent world models built through a calibrate–scale–simulate loop.

Data quality and verifiability

Automated QA and human spot checks close the quality loop. Rights are registered on-chain at capture, so every episode ships with clean provenance and a verifiable quality score.

  1. Capture
  2. QA
  3. Rights on-chain
  4. Vault

Built for

VLAWorld ModelsDexterous ManipulationGeneralist Embodied Agents
Data provenanceVerified
A robotic operator capturing a physical interaction
Capture
Robotic hands holding a motion-processing chip
Motion
A verified data network connecting physical devices
Chain
A secure verified-data vault
Vault

Product matrix

Hardware and apps feed one data engine. Every product is a capture surface for the same 4D pipeline.

In productionFlagship

Ego Suite

Native first-person 4D data combining RGB, IMU and tactile-glove streams in one wearable capture suite.

For scale contributors and field teams · Collects RGB, IMU and tactile-glove streams · Outputs first-person 4D episodes

In production

DexUMI

A standardized end-effector with force sensing establishes a reliable data-quality baseline for fundamental grasping tasks.

For labs and data teams · Collects force-labeled grasp trajectories · Outputs manipulation training sets

In production

EgoBio

Stretchable tactile arrays capture force interaction and deformation reconstruction for fine manipulation.

For fine-manipulation research · Collects tactile force and deformation fields · Outputs touch-annotated episodes

Zero barrierCompanion path

Phone 3D spatial scanning

Anyone with a phone can upload scans and earn token rewards. A 3D home model can fetch $10+ on the marketplace.

Derived apps · Optimal camera placement · Daylight simulation · Renovation preview · Accessible 3D data collection

APP STORECOMING SOONGOOGLE PLAYCOMING SOON

Model Store

Four model asset classes trained on real-world data, with checkpoints, benchmarks and licensing in one place.

Model subscriptionAPI callsPrivate deploymentIndustry licensingPay-per-taskRevenue share
A chrome geometric foundation-skill model

Foundation skill models

Core manipulation skills such as grasping, placing and inserting, trained on force-labeled real data.

A chrome scene-task model globe

Scene-task models

End-to-end task policies for kitchens, warehouses and other captured scenes.

A chrome industry-specific model cluster

Industry-specific models

Models tuned for manufacturing, power-grid inspection and medical assistance.

A chrome embodiment-adaptation model

Embodiment-adaptation models

Retarget any skill to an arm, humanoid or quadruped body.

B2B data marketplace and applications

License task-specific embodied datasets or commission custom capture campaigns across our contributor network.

World Model / VLA data licensingEmbodiment-adaptation dataIndustrial · power · medical capture3D scene simulation
A chrome data cube depicting industrial manufacturing

Industrial manufacturing

Assembly, kitting and tool-use data from tasks modeled on real production lines.

Modality
RGB-D + hand pose + contact
Scale
50K+ episodes (sample)
A chrome data cube depicting power-line inspection

Power-line inspection

Outdoor grid-inspection routines with spatial context, defect tags and route maps.

Modality
RGB-D + GPS-aligned scenes
Scale
Campaign-based
A chrome data cube depicting medical assistance

Medical assistance

Instrument handling and precise bimanual procedures from trained professionals.

Modality
DexUMI force trajectories
Scale
Expert-sourced
A chrome data cube depicting home services

Home services

Kitchen tasks, laundry and tidying in real homes, including the long tail of tasks robots struggle with.

Modality
Egocentric RGB + hand pose
Scale
180K+ episodes (sample)
A chrome data cube depicting logistics and warehousing

Logistics and warehousing

Pick-pack-sort episodes across varied bins, parcels and shelving layouts.

Modality
Egocentric video + depth + IMU
Scale
From 10K episodes
A chrome data cube depicting research training grounds

Research training grounds

Curated multi-embodiment sets for labs, benchmarks and training facilities.

Modality
Full-stack 4D episodes
Scale
Benchmark-sized cuts

Backed by

YZi Labs

Start contributingreal-world data to physical AI

Join Alpha