Egocentric video
First-person recordings of real tasks in residential, commercial and operational environments.
For AI labs
Human Data Infrastructure for Embodied AI and World Models
Real-world human experience, operational environments and managed workforce across Brazil — built for physical intelligence at scale.
What we provide
An operating layer between AI teams and the physical world
You define the data. We build the workforce, activate the environments and deliver reviewed, consented human demonstrations.
Egocentric human demonstrations
First-person capture of people performing real tasks in the environments where those tasks actually happen.
Real-world task execution
Full task sequences with natural variation, interruptions and recovery behaviour — not staged re-enactments.
Operational environments
Warehouses, fulfilment centres, production floors, service areas and other authorised working sites.
Skilled workforce
Contributors recruited by profession, trained on the capture protocol and qualified before production.
Multimodal collection readiness
Video-first today, with the operational structure to extend to audio, pose, depth and sensor streams.
Structured quality assurance
Submissions can be reviewed against explicit client criteria, with issue classification and approved-duration accounting.
Consent and compliance
Informed contributor consent, company authorisation, privacy controls and a full audit trail.
Regional scalability
Brazil-first operations designed to extend across Latin America without rebuilding the model.
Built for embodied intelligence
Human experience for world-model pretraining
MotionGrid opens a governed path to the physical world: diverse people, places, objects and long-tail tasks captured with provenance and client-defined structure.
Human and environmental diversity
Natural variation across people, homes, workplaces, tools, layouts and regional contexts.
Long-tail physical tasks
Everyday and skilled activities that are difficult to reproduce reliably in a laboratory.
Structured demonstrations
Task episodes designed around intent, sequence, interaction, recovery and observable outcomes.
Consent and provenance
Traceable contributor consent, site authorisation, protocol versioning and delivery records.
Collection modalities
Start with proven capture. Extend with the programme.
Availability is stated explicitly so research teams can distinguish deployable capabilities from client-led extensions.
Narrated demonstrations
Human demonstrations paired with spoken task context when the protocol requires it.
Structured metadata
Participant, environment, task, device, duration and quality fields mapped to client requirements.
Wearables and sensors
Capture using client-approved devices and sensor protocols, subject to equipment validation.
Depth, pose and motion
Additional streams can be introduced when hardware, synchronization and acceptance criteria are defined.
Robot and teleoperation data
A future extension for programmes in which the client supplies the robotic stack and operating protocol.
Activation roadmap
A measurable path to 10,000 qualified contributors
MotionGrid can mobilise at national scale while protecting quality through staged activation, qualification and production gates.
By day 60
1,000
Qualified contributors activated across initial regions and task profiles.
By day 120
4,000
Expansion by geography, environment and professional segment after pilot validation.
By day 180
10,000
National mobilisation target supported by regional coordination and measured quality.
Activation capacity is scoped against the approved protocol, geographic coverage, equipment availability, task supply and quality criteria. Production volume is confirmed after pilot performance is measured.
How it works
From specification to approved delivery
Every project runs through the same operational sequence, with measurable checkpoints at each stage.
- Step 01
Client specification
- Step 02
Task design
- Step 03
Workforce matching
- Step 04
Environment activation
- Step 05
Capture
- Step 06
Quality assurance
- Step 07
Compliance review
- Step 08
Approved delivery
Use cases
Training data for embodied and multimodal systems
Vision-language-action training
Paired visual context and task intent for action-grounded models.
Human-object interaction
Grasping, handling, tool use and object manipulation in cluttered real settings.
Task planning
Long-horizon sequences with sub-steps, ordering and real-world constraints.
Manipulation learning
Repeated fine-grained manipulation across operators, objects and sites.
Imitation learning
Consistent demonstrations of the same task by many qualified contributors.
Robotics evaluation
Reference human execution for benchmarking robot performance on the same task.
World-model data
Environment dynamics, layouts and human movement through operational spaces.
Multimodal AI
Structured episodes ready for enrichment with additional modalities.
Contributors
50–100 qualified
Initial scope
250–500 hours
Launch market
Brazil
Scale model
Stage-gated
Launch a Pilot in Brazil
Bounded scope, defined acceptance criteria and transparent reporting before any volume commitment.