BRENNEN

Industries / Food, farming and forestry / Agriculture and farming

Dairy

Dairy farming is keeping cows, buffalo, goats, sheep, camels and other animals for their milk, from breeding and feeding them to milking, cooling the milk and handing it over to the processor.

About 6500 BCE

Milk fat soaked into cooking pots in northwestern Anatolia, now part of Turkey, shows that people there were using milk in the seventh millennium BCE. A study of more than 2,200 pots from the Near East and southeastern Europe calls this the earliest direct evidence of milk use (Evershed and colleagues, Nature, 2008).

ebi.ac.uk

36

Milestones achieved across 6 domains.

10

To watch for: announced, not yet confirmed outside.

12

Open deliverables. Nobody has shipped them.

In one minute

Where it started

People in what is now Turkey were using milk by the seventh millennium BCE, the age of the oldest milk fat found in cooking pots. For most of the time since, animals were milked by hand and the milk was used close to home. From the late 1800s, the pulsating milking machine, the Babcock fat test, silage, milk trains and later refrigerated bulk tanks let farms sell milk to distant cities and be paid for what was in it.

Where it stands

FAO counts 930 million tonnes of milk produced in 2022, from about 150 million households. Cattle give 81% of the world's milk and buffalo 15%. India produces 22% of it, and its village cooperatives collect milk every day from 16.9 million member farmers. Large farms elsewhere milk by robot, choose bulls by their DNA and pipe manure into digesters that capture its gas. New problems keep arriving: bird flu reached US dairy cows in 2024, and nitrogen and methane from manure now limit what farms may do in the Netherlands and California.

What is open

Many everyday checks on a dairy are still done by eye or by hand: reading the mastitis paddle test, scoring calf dung, shaking feed through sieves. Many of the newest claims, from batch-milking robots and solar milk chillers to AI herd assistants and the methane cuts credited to digesters, rest on the maker's or funder's own numbers. Phone cameras, open vision models, cheap sensors and free satellite data now let a small team build these checks and publish results that anyone can verify.

Domains

AchievedWatch forOpen

Figureheads

1 person

Advances to build with

1 research note

Open in dairy

3 of 12 open deliverables

Open · Breeding and herd health

Phone-read mastitis paddle test

An offline phone app that reads the California Mastitis Test (CMT), the cow-side test in which a little milk from each quarter of the udder is mixed with a reagent on a paddle and the gel that forms is judged by eye. The app photographs the paddle and returns a score and an estimated somatic cell count (white blood cells in milk, which rise with infection) for each quarter, for cows, buffalo or goats.

Built with what exists

Any Android phone and a standard CMT paddle and reagent; Label Studio for labeling; an Ultralytics YOLO11 classifier, or a small vision model from Liquid AI's LFM3 family fine-tuned on one graphics chip (see the Note "One GPU to fine-tune"); LiteRT to run the model on the phone; a milk-recording lab for reference cell counts.

Credit · unclaimed

Open · Feed

TMR particle check by photo

A phone method that measures how long the pieces in a total mixed ration are. Spread a sample on a dark tray with a printed scale, take a photo, and get the shares that would have landed on each sieve of the Penn State Particle Separator, the farm's standard tool. Run it on fresh feed and on the leftovers to show how much the cows sorted their ration.

Built with what exists

A phone camera; Meta's Segment Anything Model 2 (SAM 2) to outline each piece; OpenCV to measure lengths; a Penn State Particle Separator and kitchen scale for reference results; Python.

Credit · unclaimed

Open · Milking

Open cup-attaching test rig

A bench test anyone can copy, for the hardest step in robot milking: putting four cups on four teats. A silicone udder on a frame sets the teats at random positions and angles. A desktop robot arm finds them with a depth camera and attaches teat cups under vacuum, with water in place of milk. Release the designs, code, recordings and a scoring sheet, so research and commercial systems can be compared on the same rig.

Built with what exists

A desktop six-axis arm such as the UFactory xArm 6, programmed through its open Python SDK; a RealSense D405 depth camera; Ultralytics YOLO to find teats in images; teats cast in platinum-cure silicone from 3D-printed molds; standard teat cups, liners and a portable bucket milking unit for vacuum. A learned policy can be trained from recorded attempts with Hugging Face's LeRobot library (see the Note "No download").

Credit · unclaimed

Where it sits