Housekeeping video data for physical AI
Scope housekeeping recordings for organizing, folding and routine cleaning while protecting personal spaces and distinguishing visible outcomes.
Housekeeping tasks reveal how people organize objects, handle flexible materials and work through a changing space. The research question might concern a sequence of placements, a folded item or the relationship between a cleaning tool and a surface. A focused brief makes these ordinary activities easier to capture and evaluate.
Homes and occupied rooms also contain unusually personal context. Choose an approved area and remove sensitive material before recording begins. A useful dataset does not need family photographs, correspondence or unrelated residents to explain how a towel is folded or a shelf is organized.
Tasks to include in the brief
Folding textiles
Record a complete fold with the item initially visible and its final arrangement shown. Vary garment or towel dimensions deliberately, and retain an ordinary repositioning when the material slips or bunches.
Organizing a shelf
Show selecting an object, finding a position and placing it among other approved items. Specify whether maintaining object identity or understanding the arrangement is the main research goal.
Clearing a work surface
Capture moving a defined group of objects into agreed destinations. Keep the before and after states observable; a final empty surface alone does not explain where the items went.
Routine surface care
Use an already approved cleaning task with the tool and target area in view. Label the visible motion and coverage, rather than assuming that a surface has been sanitized.
Capture requirements
- Frame the object manipulation at useful scale. A fixed view can preserve the room layout, while a declared first-person view may show the working hands more clearly.
- For textiles, avoid losing the item at the bottom of the frame. Include its initial shape and the completed or incomplete fold so the outcome is interpretable.
- Vary object size, material, shelf height and surface clutter within the agreed scope. Keep room and session identifiers so related clips can stay together in evaluation splits.
- Record natural interruptions or retries only when appropriate. Mark an obscured destination as uncertain instead of treating a smooth-looking movement as proof of correct placement.
Useful annotation layers
- Temporal actions may include unfold, align, fold, lift, place, wipe and return. Define whether repeated wiping motions form one interval or several events.
- Object and surface descriptions can support retrieval, but visible appearance is not a reliable measurement of cleanliness, contamination or material composition.
- Ordinary video does not provide fabric tension, contact pressure, exact surface coverage or a robot’s manipulation trajectory. Specify any such measurements separately before collection.
Permissions, boundaries and exclusions
- Exclude personal documents, medication, photographs, mirrors revealing other people and screens with accounts or messages. Check the full frame, not only the working surface.
- Do not record occupied bedrooms, bathrooms or guest spaces without the necessary explicit permissions. Approval from one participant does not cover everyone associated with the room.
- Avoid chemical handling, ladders, heavy furniture movement and any task made unsafe by the recording setup. Never create a hazardous mess to demonstrate recovery.