Restaurant prep videos for AI task and action recognition

A useful kitchen clip shows what the cook started with, what they changed, and the result. Figure’s Index announcement identifies cooking and restaurant tasks among its capture examples. That is evidence that one robotics company is gathering human task video. An individual restaurant can use a scoped pilot to see whether its own footage meets a specific project brief. Figure Index

Start with a cold prep station

Record one assembly task using reusable training materials or an approved cold station. A cook receives a sample ticket, selects the correct vessel and utensils, portions three items and verifies the result. Capture one alternative order and one labeled correction after a wrong item is selected. Record the order version, ingredient identifiers, timestamps and final match against the mock ticket. Use a stable camera with the hands and work surface in view.

A buyer building a video understanding model might need time ranges for each action. A buyer testing a planning assistant might want the order, the expected sequence and the decision to correct a mistake. Confirm the target first, then decide whether video, event labels, or paired text is needed.

Separate the shoot from food service

Get permission from the operator and everyone you will record. Keep customers, live tickets, payment screens, proprietary recipes and staff schedules out of frame. Stage examples away from active hazards. Your kitchen’s safety and food-handling requirements still apply to a shoot. FDA Food Code

Start by describing one repeatable prep task. Include available camera views, timing, permission scope and the type of variation you can demonstrate. Describe collection capabilities