Real-world training data for physical AI

Train robots on the environments they will actually face.

DXTRAL captures consented, first-person video and sensor data from factories, warehouses, workshops, and other operational environments across Latin America and Asia. We deliver rights-cleared, task-level datasets structured around your model, schema, and deployment requirements.

Consent-cleared and privacy-first

Commercial-use rights included

Task-level annotations

Custom protocols and schemas

toolbox
hand · grasp
pliers
FIRST-PERSON VIEW
RGB + IMU
Task
Tool retrieval
Modality
RGB + IMU
Region
Monterrey, Mexico
Labels
Hand pose, grasp, object state
Rights
Commercial training use
00:12 / 00:45
2,800+
Hours collected
1,200+
Workers onboarded
180+
Facilities covered
15
Countries operating
Operating in LATAM, North America and Asia.
Trusted by AI labs and robotics companies
What we capture

The workplaces you can't stage.

Industrial

Industrial floors

Assembly, machining, packing, food and textile production. Hands and tools in frame.

Commercial

Commercial & retail

Warehouses, kitchens, markets, and service counters. Everyday work at real speed.

Logistics

Logistics & ports

Loading, sorting, and last-mile handling across major trade corridors.

Process

From contract to labeled data.

01 / CONTRACT

Contract

We already hold the agreements. Capture happens inside working businesses.

02 / CAPTURE

Capture

Operators wear head- and chest-mounted rigs. RGB, depth, and hand pose.

03 / ANNOTATE

Annotate

Consent verified. Activities, task boundaries, and scene context labeled.

04 / DELIVER

Deliver

Formatted to your schema. No cleanup on your end.

Coverage

Two continents. Around the clock.

Capture runs on local working hours in both hemispheres, so volume scales without adding to your team.

Latin America

  • Monterrey MX
  • Saltillo MX
  • Guadalajara MX
  • Mexico City MX
  • Durango MX
  • Bogotá CO

Asia

  • Jakarta ID
  • Ho Chi Minh City VN
  • Chennai IN
  • Bangkok TH
  • Manila PH
  • Bengaluru IN
How we compare

Pick the right source for the job.

SourceWorks whenWatch out for
Academic datasetYou need a baseline benchmarkWrong license or task fit
Internal labYou own the rigs and operatorsSlow to scale, narrow scenes
Synthetic / simYou want cheap volumeBreaks on contact-rich tasks
DXTRALYou need real industrial and commercial work, at scaleBest with a clear task spec
Built for
Humanoid & robotics labs Foundation model teams World-model research Manipulation & policy learning
Get started

Tell us the task. We'll get the data.

Send the environments and tasks you need. We scope a collection run and get you a sample before you commit.

Request a sample dataset →
Sample
A labeled clip set from your target scene
Discuss a collection project →
Direct