For robotics teams
The demonstrations your policy is missing.
Define the task, the viewpoint, and the environments. QwestLab recruits the workforce, runs the QA, and delivers structured real-world video with consent and provenance attached to every clip.
Example task protocol
- task
- grocery_putaway
- viewpoint
- egocentric
- duration
- 60–300s
- hands_visible
- ≥ 95%
- environments
- home_kitchen
- bystander_faces
- 0
Matching capture
EGOCENTRIC · 1080p30 · HANDS 97%
A workforce on day one
Embedded in Kumu — 100,000+ monthly actives — and open beyond it
First-person capture
Head and chest mounted, hands in frame
Real environments
Homes, kitchens, warehouses, retail floors
Enterprise-grade QA
Automated and human review on every clip
Physical AI
Help AI understand the physical world.
Robots learn from examples of how people move, handle objects, complete tasks, and navigate real environments. QwestLab coordinates contributors around the world to collect structured real-world demonstrations for Physical AI.
- Household tasks
- Kitchens
- Warehouses
- Retail
- Hospitality
- Food service
- Industrial environments
- Agriculture
Folding laundry
EGOCENTRIC · BOTH HANDS
Retail restock
EGOCENTRIC · OBJECT HANDLING
- Task
- Real-world capture
- Quality review
- Structured data
- Robot training
The deliverable
The camera sees what the worker sees.
Egocentric frames of real tasks in real workplaces — hands, objects and the state of the work, with faces and bystanders excluded by protocol rather than by luck. Commercial sites come through our Real-World Network: Environment Partner businesses that have signed location releases, so provenance is part of the deliverable.
Every accepted clip ships with consent lineage, device provenance and QA results attached. Illustrations of the viewpoint, not photographs of QwestLab sites or customers.
What we collect
One workforce. Every form of real-world video.
Projects specify the viewpoint, the task length, and the environments they need covered. Contributors record to that specification.
Egocentric
Head mounted, first person. Both hands in frame throughout the task.
Third person
Fixed or handheld camera capturing the whole body and the working area.
Multi-camera
Synchronised viewpoints of the same demonstration for depth and occlusion coverage.
Long-horizon tasks
Complete activities start to finish — cooking a meal, unpacking a delivery — not isolated motions.
Object manipulation
Grasping, moving, opening, and placing, with the objects and surfaces a robot has to generalise across.
Environment coverage
The same task across many homes and sites, so a model sees variation rather than one kitchen.
Quality
Raw human activity becomes AI-ready data.
QwestLab tracks quality at every step, including contributor reliability, device metadata, capture requirements, rejection reasons, and project-specific acceptance criteria.
- 01Contributor
- 02Task
- 03Capture
- 04Automated checks
- 05Human QA
- 06Structuring
- 07Accepted clip
- 08Client delivery
Workforce intelligence
The right people for the right data.
QwestLab builds a continuously improving workforce profile so projects can be matched to contributors based on real capability — not just location.
Example contributor profile
- Location
- Cebu, Philippines
- Occupation
- Restaurant worker
- Equipment
- Samsung S25, Chest mount
- Environments
- Home kitchen, Commercial kitchen
Capture rating
94%
Task acceptance
96%
Global network
A workforce that already exists.
QwestLab recruits into communities where the people already are, so a project does not begin with a recruitment campaign. The first of those communities is Kumu — over 100,000 monthly active users in the Philippines — which is why collection there can start immediately. Contributors anywhere else join QwestLab directly, with no Kumu account.
The Philippines is where the workforce runs deepest today, not where the platform ends. Other regions open country by country as communities and local operators come online.
- Southeast Asia
- South Asia
- North America
- Latin America
- Europe
- Australia
- Africa
Why QwestLab
Infrastructure, not a crowdsourcing app.
Global workforce
Recruit people by geography, occupation, equipment, and the environments they can actually reach and film in.
First-person capture
Head and chest mounted recording from the wearer's point of view — the perspective a robot has to learn from.
Quality first
Every clip goes through technical, task, privacy, and human QA before it counts toward a project.
Built for scale
Projects run from a handful of carefully designed demonstrations to thousands of accepted hours.
Rights and provenance
Consent, usage rights, contributor identity, project rules, and data lineage are tracked per clip.
Unstaged by design
Tasks are recorded in real homes and workplaces during real work, because staged footage teaches robots the wrong thing.
For AI companies
Need real-world data? Tell us what you need.
QwestLab turns your capture specification into a managed global collection program — from contributor recruitment and task execution to QA, rights, and delivery.
- 1Define the demonstration
- 2Match workforce
- 3Collect video
- 4Verify quality
- 5Structure and label
- 6Deliver dataset
Platform capabilities
- Configurable task design
- Global contributor recruitment
- Geographic quotas
- Device and mount requirements
- Specialised environments
- First-person video
- Multi-camera capture
- Hands-visible scoring
- Automated QA
- Human QA
- Consent and provenance
- Client dashboards
- Capacity forecasting
Also commissionable — delivered on a separate platform
Annotation and labelling
Action labels, segmentation, and metadata applied to delivered clips.
Speech and audio
Voice and conversational data collection in local languages and code-switched speech — Tagalog, Taglish, and Filipino English among them.
Human evaluation
Model output review, ranking, and task-completion judgement.
Define the requirement with us alongside a capture project. This module’s contributor workforce records video only; that work runs on a separate QwestLab platform.
What data does your AI system need next?
From 50 carefully designed demonstrations to 10,000+ hours of distributed real-world capture, QwestLab can build the workforce and workflow around your specification.