Improved animal welfare and greater effi­ciency through AI and robotics

Picture a humanoid robot ambling through poultry houses checking condi­tions and completing general tasks, maybe even with its own robotic dog at its side. From a distant field, a farmer asks his phone, “Hey McDonald, how are my chickens doing today?” The simple prompt returns a rundown of key infor­ma­tion ranging from temper­a­ture read­ings to bird behavior indi­cating a need for inter­ven­tion such as a change in feed or a vet visit. 

This future isn’t as science fiction or distant as it sounds. Georgia Tech Research Insti­tute senior research scien­tist and robotics branch chief, Colin Usher, says the compo­nents of this reality are coming together at a pace that is fast accel­er­ating. Cost is falling. In 2025 a Chinese company offered a bipedal humanoid robot for sale for less than $6,000. Power needed to run onboard computers is shrinking while compu­ta­tion power is increasing. An internet connec­tion can flash data from barn sensors to data centers providing near-instant results for calcu­la­tions that used to take days. 

Data instead of gut feeling

The use of large language models (LLMs) will likely allow producers to interact with tech­nology using normal language and basic prompts, Usher says. Data gath­ering and inter­pre­ta­tion are crit­ical fronts to advancing func­tional autonomy in live­stock produc­tion. 

We have data. We want infor­ma­tion. How can we tie these points together to tell a story and make a deci­sion that will have an impact down the road.

Suzanne Leonard

“We have data. We want infor­ma­tion. How can we tie these points together to tell a story and make a deci­sion that will have an impact down the road,” says Suzanne Leonard, North Carolina State Univ. Exten­sion specialist. Infor­ma­tion also has to be econom­i­cally valid. Indi­vidual bird feed conver­sion isn’t likely to be valu­able for commer­cial producers. Sensors to eval­uate envi­ron­mental condi­tions in the pig barn to opti­mize health and decrease mortality is a much easier sell. 

AI detects prob­lems at an early stage

Guoming Li, Univ. Of Georgia poultry engi­neer, sees near-term poten­tial for computer vision moni­toring of gait scores and other behav­iors in broiler flocks. AI and deep learning models can gather and analyze behavior metrics in week two that can predict gait scores in week seven. “We can use early predic­tion to make nutri­tion or other correc­tive strate­gies to improve gait, which will possibly improve welfare and produc­tivity later,” Li says. 

We can use early predic­tion to make nutri­tion or other correc­tive strate­gies to improve gait, which will possibly improve welfare and produc­tivity later.

Guoming Li

Computer vision levels the board for expe­ri­ence. A veteran producer may notice subtle behav­ioral signs a fresh hire may not. “Sensors and compu­ta­tion provide more objec­tive, accu­rate, and consis­tent eval­u­a­tion,” Li says. Tech can also liter­ally be an extra set of eyes. NCSU scien­tists are seeking funding to advance a camera system to monitor farrowing. 

Instead of a human constantly checking 50 farrowing sows, the moni­toring system keeps track. If a sow is doing her job and deliv­ering every 10-15 minutes, let nature take its course. But if the system deter­mines it’s been too long between piglets, it will alert for human inter­ven­tion. “Tech­nolo­gies like these can help humans opti­mize how we spend our time,” Leonard says. 

Robots take over routine tasks in the barn

One such tech­nology nearing commer­cial­iza­tion is a Georgia Tech robot that collects floor eggs in broiler breeder houses. It’s an impor­tant, but simple and time-consuming task. If you don’t pick up the eggs, more eggs get laid on the floor, Usher says. It’s simple labor, but labor and time are a commodity in short supply. 

A lot of work has been done on robots working with humans, not a lot has been done on robots working with animals.

Colin Usher

Usher notes the impor­tance of secondary func­tions for industry-wide adop­tion of robotics. The egg collector, for example, could also collect mortal­i­ties, haze chickens to nests, encourage new chicks to feed and water, and monitor barn condi­tions. “The robot is using AI algo­rithms to look at the scene as it’s driving around. Why can’t the camera system also eval­uate chickens, feeders, drinkers, and more?” he says. Inter­ac­tion between animals and robots is a chal­lenge. The egg collecting robot had to be trained to ‘play chicken’ to clear a path. “A lot of work has been done on robots working with humans, not a lot has been done on robots working with animals,” he says.