First-person human data for Physical AI.
AIMR POV — our first-person capture series. AIMR collects real-world egocentric data from skilled frontline workers doing their actual jobs — across floriculture, culinary, horticulture & facility farming, livestock & veterinary, cleaning & facility, office & document processing, and campus & classroom environments.
AIMR POV
Built to a written protocol.
Every AIMR POV dataset is built to a written protocol. Task taxonomy, device, capture specification, metadata schema and diversity quotas are agreed before capture begins.
Protocol-based collection
Device, scenario, capture specification and metadata schema defined by you. We collect to spec rather than offering whatever we happen to hold.
Skilled operators, real jobs
Our collectors are trained practitioners performing their own occupation. Not actors, not scripted demonstrations.
Cross-industry diversity
The same action performed across many environments teaches a model more than the same action repeated in one room.
Consent & traceability
Every deliverable ships with its chain of authorization and an auditable ledger.
Approximately 1,000 trained operators across partner schools, research institutes, enterprises and farms in China.
Seven environments, one protocol standard.
Floriculture
Fresh-cut flower harvesting, open-stage judgement, grading, bunching and packing, captured across multiple flower parks and greenhouse bases.
Culinary
Knife work, wok handling, heat and doneness judgement, plating and service, captured in real commercial and training kitchens.
Horticulture & Facility Farming
Orchard pruning, grafting, greenhouse vegetable work captured across growth stages and difficult lighting conditions.
Livestock & Veterinary
Animal handling, husbandry, milking, barn cleaning, and clinical procedures captured under veterinary supervision.
Cleaning & Facility
Commercial cleaning, janitorial work, waste handling and housekeeping in real office and industrial environments.
Office & Document Processing
Filing, sorting, scanning and document handling by administrative staff. Collected using de-identified sample materials on request.
Campus & Classroom
Lectures, note-taking, laboratory and library sessions, and campus daily-life routines, captured in real classroom environments.
Your scenario isn't listed?
AIMR collects to protocol, not from a fixed catalogue. Tell us the task, environment and device — we design the capture spec and run a pilot before scaling.
From protocol to delivery in four steps.
Define
We agree the protocol: task taxonomy, device, capture spec, metadata schema, diversity quotas.
Pilot
We deliver a 20 to 50 hour validation sample across multiple operators and environments.
Validate
Your team runs the sample through your own quality and training checks. Acceptance criteria are agreed in writing before scaling.
Scale
Volume ramps against the agreed specification, with batch-level quality reporting.