Industries / Food, farming and forestry / Agriculture and farming
Produce packing
Produce packing takes fresh fruit and vegetables in from the field and washes, sorts, grades, packs and cools them, so that they reach buyers clean, graded and fresh.
1893
Southern California citrus growers set up the Southern California Fruit Exchange, a cooperative whose local associations each packed their members' fruit in their own packing houses under their own brands; until then much citrus had been packed by hand in the groves.
38
Milestones achieved across 6 domains.
9
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
In the early California citrus trade, fruit was packed by hand in the groves and sold in local markets. As railways and refrigerated rail cars opened markets in the east, packing moved into packing houses beside the tracks, and from 1893 growers' cooperative associations ran them under their own brands. The first houses did no washing, and packers graded the fruit by eye as they packed; washers, sizers, cooling rooms and legal rules on containers and grades followed.
Where it stands
The world grew about 2.1 billion tonnes of fruit and vegetables in 2023, by the count of the UN Food and Agriculture Organization (FAO). Large packhouses now grade each fruit by camera and near-infrared light, fill some trays by robot, cool produce by forced air or vacuum within hours of picking, and work under food-safety rules and audits written after deadly outbreaks traced to spinach in 2006 and cantaloupes in 2011. Elsewhere much of the work is still done by hand with little cooling: FAO estimates that 13.3% of all food was lost between harvest and retail in 2023, and 23.0% in sub-Saharan Africa.
What is open
Many checks are still made by eye or on paper: how long a bin waited before it was cooled, how big a bruise is, which crops draw in cold wash water, which records lead back to a contaminated lot. Open jobs include a public clock on harvested bins, a dye test of wash-water uptake across crops, measured cooling times for the packages on sale, an open robot arm that learns to pack fruit into trays from a few demonstrations, and a trace-back drill on which people and AI agents can be scored.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Receiving from the field
Bringing the harvest in: moving produce from the field in bins, deciding whether a load is ripe and sound enough to accept, and emptying it onto the packing line without bruising it.
5 · 1 · 2
Last achievedLight reads mango maturity2017
Watch forAI berry checks at intake
2
openWashing and food safety
Cleaning produce and stopping germs from spreading from one fruit to the next, through the wash water, the equipment, the building, and the standards, rules and audits that check them.
7 · 1 · 2
Last achievedEurope caps chlorate residues2020
Watch forGerm counts from the wash
2
openSorting and grading
Sorting and grading divide each harvest by size, colour, defects and, more recently, hidden inside quality, into the classes that buyers pay for.
5 · 2 · 2
Last achievedDeep learning grades apples2022
Watch forMRI scans inside the fruit
2
openPacking
Packing puts graded produce into the baskets, cartons, trays, bags and punnets it is sold in, in set counts or weights.
7 · 2 · 2
Last achievedFrance's plastic ban annulled2024
Watch forA packer for round fruit
2
openCooling and storage
Taking the field heat out of produce soon after harvest, then holding, colouring or ripening it at the temperature and air that keep it best.
7 · 2 · 2
Last achievedWindow units chill farm rooms2006
Watch forRipening rooms run by prediction
2
openTracing produce
Marking produce and keeping the records that show where each item came from and where it went, so that a problem can be traced to its source.
7 · 1 · 2
Last achievedStickers that rot at home2023
Watch forLot data on fruit stickers
2
open
Advances to build with
2 research notes

NVIDIA GR00T 2
Five demonstrations per task
By NVIDIA's own test, its robot model GR00T 2 averages 95% success after five demonstrations of a task, on 50 tasks picked from those it learns that way.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.
Open in produce packing
3 of 12 open deliverables
Open · Receiving from the field
Bin photos against the grader
A public test of whether photos of the top of a bin, taken at receiving, predict what the packing line's optical grader later finds in the whole bin: the spread of sizes, the colour and the defects. Photograph each bin, run it through the grader on its own, and release the images, the grader's results for each bin and an open sizing model.
Built with what exists
Phones or a fixed camera over the receiving line; the packhouse's existing optical grader; an open object-detection or segmentation model such as Ultralytics YOLO or Meta's Segment Anything 2 for finding and measuring each visible fruit; Python for the comparison; any commercial bin-sizing app the packhouse already uses, tested the same way.
Open · Washing and food safety
Which crops drink their wash
A public table, with its method and photos, of how much water each common fresh crop takes in when dipped in water colder than itself, measured with a food dye instead of germs: for each crop, how far the dye spreads inside the fruit at several temperature gaps, depths and dipping times.
Built with what exists
FD&C Blue No. 1 (Brilliant Blue FCF), a permitted food colour; temperature-controlled water baths; a phone camera on a copy stand or a flatbed scanner for cut sections; Fiji (ImageJ) to measure the dyed area; Python for the statistics.
Open · Sorting and grading
Defect areas, measured not guessed
An open test set of about 2,000 apples of one or two varieties in which every defect has a measured size. Photograph each apple from six sides in a light box beside a printed scale, and have two people trace each bruise, scab spot and skin fault on the photos so its area is known in square centimetres, which fixes the class the fruit should get under the EU standard. Then have three graders trained on the standard grade the same fruit by eye as Extra, Class I, Class II or out of grade, and score open vision models on the photos.
Built with what exists
A turntable and light box; a phone or USB camera; a colour and scale card; OpenCV to measure defect areas; Label Studio for the graders' entries; open vision models such as DINOv2 or OpenCLIP, fine-tuned in PyTorch.
Where it sits
Broader