Farm and landscaping work is some of the hardest for a robot to learn from a lab. Every plant is a different shape, the light changes through the day, and the ground is never flat. A picker in a citrus grove makes dozens of small decisions a minute about which fruit is ready, how to twist it off, and where to put it. None of that shows up in staged footage.
HiOps films this work where it happens. Our collectors wear head-mounted cameras while they feed animals, weed beds, pick and sort crops, tend greenhouse benches, and repair irrigation lines. Fixed cameras around the work area record the same tasks from the outside, synchronized with the first-person view.
We run collection on partner farms, nurseries, and grounds crews in five countries, so you get variation in crops, climates, tools, and techniques rather than one site filmed many times. Participants are consented and paid, and the footage is cleared for commercial model training.
Each episode is cut into tasks and sub-tasks, labeled in plain language, and tagged for success or failure. Hand and object tracks are available on request. You receive the data in the schema your pipeline expects, delivered into your own cloud bucket with a manifest and acceptance rates.
If you are training a manipulation or mobile model for agriculture, start with a sample. Tell us which of the tasks below matter and we will send annotated footage within a business day.