Crop Picking, Planting or Sorting

Pick a strawberry too hard and it bruises. Pick it too early and it is sour. A person learns the difference in a week by touch and color, and then does it a few thousand times a day, bent at the waist, while shifting a tray with the other hand. That loop, perceive, decide, grasp, place, is where most robots still fail in the field.

Planting work means setting seedlings at the right depth and spacing from a tray, firming soil without crushing stems, and keeping pace with a transplanter. Sorting happens at the packhouse table, where workers grade by size, color, and blemish, discard culls, and fill trays to a count while a belt keeps moving.

The capture rig is a head-mounted camera on the worker and fixed cameras over the row or the sorting line, time-synchronized. We record full shifts in working farms and packhouses in five countries. The workers are paid participants who have given consent, and the data is cleared for commercial model training.

Episode segmentation diagramONE EPISODE01Assess02Pick03Place04Move05SortEACH STEP: LANGUAGE LABEL · OUTCOME TAG · OPTIONAL HAND / OBJECT TRACKS
A typical crop picking, planting or sorting episode, cut into the steps your labels attach to.

Deliveries are segmented into episodes, with steps labeled in plain language: locate ripe fruit, grasp, twist, place in punnet, move tray. Each step is tagged with a success or failure outcome, so a dropped or damaged item is marked. We can add hand, object, and tool tracking. The files arrive in your own schema, in your own storage.

A sample is one message away and ships within a business day.