We turn hard-to-source human expertise into training data for Physical AI.
Tell us what your robot needs to learn. We find the people who already know how to do it.
A completed task is not always a good demonstration. Robots need examples of how the job should actually be done.
The difference isn't what gets done. It's how.
Start with the skill. We source people who already know how to do it and design the capture around your training needs.
Capture the small decisions behind safe human assistance.
Experienced caregivers can demonstrate mobility, transfer, positioning and everyday assistance tasks for robot learning.
Capture skills learned through years of work at sea.
Source experienced maritime professionals for maintenance, inspection, equipment handling and operational demonstrations.
Turn hands-on craft into robot training data.
Capture tool use, manipulation and work sequences from people who perform these tasks professionally.
Capture the dexterity behind everyday food preparation.
Cooks and kitchen professionals demonstrate the hand skills and sequences that make food preparation fast, safe and consistent.
Tell us what your robot needs to learn. We'll find the people who already know how to do it.
No fixed catalog. No one-size-fits-all capture. Every project starts with what your robot needs to learn.
One running example throughout: Inspect a diesel engine.
Buyer defines robot, environment, task, success criteria and desired output.
Start with the behavior, not the camera setup.
Sourcing is filtered on the experience the task actually demands.
We source for the skill your task actually requires.
Capture layers are configured per project, not fixed to one rig.
The capture protocol follows the learning objective.
Sample first. Scale only after the data fits your pipeline.
Start with a small evaluation set. Inspect the episodes, metadata and provenance before discussing volume.
Professional background, task, consent and QA can travel with the episode so your team knows where the data came from.
// illustrative schema · values are placeholders until a real episode is published { "episode_id": "EED-0001", "task": "engine_room_inspection", "professional": { "role": "marine_engineer", "years_experience": "10+", "countries_of_practice": ["…"], "credential_status": "verified_when_required" }, "capture": ["first_person_video", "hands", "trajectory"], "qa": { "sensor": "…", "skill": { "reviewer_role": "…", "verdict": "…" } }, "consent": { "data_use": "…", "likeness": "…" } }
First-person video · hands · motion · trajectories · additional sensors on request
Episodes · task steps · metadata · professional context · QA
Standard robotics formats where supported by the project. Custom schema discussion available.
Consent · licensing options · transfer/exclusivity discussion · private delivery options
InnomaLabs is building its professional sourcing network from the Philippines, a market with deep talent across maritime, care, hospitality, construction and skilled trades.
Instead of recruiting generic operators and teaching them a task from scratch, we can source for prior professional experience when the project requires it.
We're working with a small group of robotics teams to shape the first Expert Embodied Data projects. You bring the task. We design the capture, source the right professionals and build an evaluation set around your requirements.
A rough description is enough. We'll work out the capture details with you.
That is the model. Start with the task and we scope the professional profile, environment, capture and output around it.
No. The professions are examples. We source around the skill required by the project.
Some tasks contain technique, sequencing and judgment developed through practice. We are testing where that prior experience creates useful training data. The evaluation set lets your team judge it directly.
We scope output around the buyer's pipeline. Standard robotics formats can be supported where appropriate. Confirm the exact schema during scoping.
Verification can include employment history, credentials and partner records depending on the profession and project requirements.
Licensing, transfer and exclusivity are defined per project.
Yes. The preferred path is a small evaluation set before a larger collection.
InnomaLabs is building its initial sourcing and capture operations from the Philippines, with project scope defined around each task.
Give us the task. We'll find the people who already know how to do it.
