Original video
MP4 or WebM recordings of complete physical tasks. Source files stay private; licensed download access depends on the order’s confirmed payment and status.
Human activity.
Structured for
physical AI.
Find the physical tasks your research needs. Explore original footage, task context and timed actions by activity, then check the live catalogue for availability.
A package you
can inspect.
The recording, its labels and the evidence behind them remain distinct. Check each collection’s specifications before choosing it.
Structured context
JSONL metadata and a CSV index connect recording references, activity, viewpoint and technical fields. Explicit provenance distinguishes measured properties from contributor declarations.
Timed actions
Action intervals describe visible steps on the recording timeline. Automatically accepted footage also includes a full analysis JSON with the model, processing versions and limitations.
Inspect the manifest fieldsUnderstand the analysis JSONRead the data standardsFollow the collection pipeline
Accounts and private uploads are open. New automatic analysis is waiting for its production connection. Paid checkout and contributor payouts will open later.
Choose the
physical task.
Twelve activity areas share one catalogue taxonomy. These guides describe capture and annotation choices; the marketplace reports current supply.
Gardening
Repotting, watering, pruning and organizing garden tools.
Browse gardening Read the capture and annotation guideCooking
Preparing ingredients, pouring, mixing and handling utensils.
Browse cooking Read the capture and annotation guideHousekeeping
Folding textiles, cleaning surfaces and putting objects away.
Browse housekeeping Read the capture and annotation guideLogistics
Picking, packing, sorting and moving goods between locations.
Browse logistics Read the capture and annotation guideManufacturing
Positioning components, assembling parts and handling materials.
Browse manufacturing Read the capture and annotation guideCrafts
Working with textiles, materials and hand tools through a complete task.
Browse crafts Read the capture and annotation guideMaintenance
Inspecting, adjusting and repairing equipment in authorized settings.
Browse maintenance Read the capture and annotation guideHospitality
Setting a table, preparing a service station and arranging supplies.
Browse hospitality Read the capture and annotation guideRetail
Restocking shelves, arranging products and preparing customer orders.
Browse retail Read the capture and annotation guideAgriculture
Handling crops, sorting produce and performing authorized farm tasks.
Browse agriculture Read the capture and annotation guideConstruction
Preparing materials and performing safe, authorized installation tasks.
Browse construction Read the capture and annotation guideOther physical activities
A clearly defined physical activity with visible steps, objects and outcomes.
Browse other physical activities Read the capture and annotation guideA MORE SPECIFIC REQUIREMENTStart with
Start with
the task.
Make the brief concrete.
Describe what you want a model to observe, the variation you need and the evidence required to evaluate it. Availability and any additional capture or annotation work must be agreed for that request.
- Visible task steps, initial state and intended outcome
- Target hours, camera views, resolution and frame rate
- Objects, tools, environments and evaluation exclusions
- Action vocabulary, annotation layers and delivery format
Human video can support action understanding and learning from demonstrations. Ordinary footage does not supply measured robot commands, calibrated pose, depth, masks or force. Specialist labels need a separate specification and suitable capture.
Write a dataset brief