AI dairy feeding needs to count the leftovers

A milking robot can record the concentrate it dispenses without knowing whether the cow eats it. Leftovers may be eaten by the next cow, so a delivery record can describe the wrong animal’s intake.[1]

University of Wisconsin guidance explains the existing machinery: feed tables assign allowances, while calibrated auger rotations dispense them. That controls delivery; it does not measure consumption.[1]

A 2026 paper by Kelechi Akwataghibe and colleagues addresses the gap with a 3D camera and a neural network. Images of the feed bin before and after each visit allow the model to estimate concentrate remaining and consumed.[2]

The team reports a 28-day study of 48 lactating Holstein cows covering 5,081 feeding events. Its estimates varied between cows and between visits by the same cow. This concerns concentrate at the milking station, not the animal’s entire diet.[2]

Virginia Tech’s 27 October account of Self-learning Dairy describes the next decision: matching protein supply to each cow. The four-year, $1.15 million USDA-funded project uses robotic milking and feeding records to investigate individual nutritional responses.[3]

Amino acids supply the building blocks of milk protein, while energy powers their assembly. Excess dietary protein contributes nitrogen to manure and urine. The project aims to improve production and nitrogen-use efficiency by reallocating protein among cows with different needs. These are development goals, rather than measured results of a newly launched service.[3]

The wider ration still matters. Wisconsin’s guidance explains that feed offered at the bunk supplies most daily nutrients; concentrate eaten in the robot complements it.[1]

Suppose a cow is offered 1 kg of concentrate and eats 700 g. A model that logs the full kilogram as intake would associate her milk output with 300 g of feed she never ate.[2][3]

Report an error or suggest a correction