7/20/2026
Researchers in the Department of Industrial & Enterprise Systems Engineering have developed Behav2Need, a new AI framework that goes beyond recognizing animal behavior to infer the underlying needs driving those actions. By combining animal behavior science, multimodal sensing and causal AI, the system helps explain why animals behave the way they do, enabling more accurate welfare monitoring and earlier interventions on smart farms.
Written by
Researchers at the University of Illinois Urbana-Champaign, in collaboration with the University of Michigan and the U.S. Department of Agriculture, have developed a new artificial intelligence framework that goes beyond recognizing animal behavior to infer the underlying needs driving those actions. The technology, called Behav2Need, could help farmers better monitor animal welfare and intervene before health problems become severe.
The study, led by doctoral researcher Ruiqing Wang and assistant professor Yiwen Dong in the Department of Industrial & Enterprise Systems Engineering, addresses a longstanding challenge in animal monitoring. While existing technologies can identify what animals are doing — such as sleeping, nursing or moving — they often cannot explain why those behaviors occur.
"Most existing animal-monitoring systems focus on identifying what the animal is doing while providing limited insight into why the behavior occurs," Wang said. "Our goal was to move beyond behavior recognition by connecting observed behaviors with the underlying needs and environmental conditions that drive them."
Presented at the 2026 ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys), the research combines principles from ethology with multimodal sensing and causal artificial intelligence to better understand animal welfare.
Moving Beyond Behavior Recognition
Animals cannot directly communicate internal states such as hunger, fatigue, thermal discomfort or social needs, making it difficult for farmers and animal-care professionals to identify welfare concerns before they become serious.
Drawing from principles of ethology, the researchers developed Behav2Need to model how internal needs and external conditions interact to produce observable behaviors. The framework combines video cameras, floor-mounted vibration sensors and a Dynamic Causal Bayesian Network to estimate latent needs — such as energy demand, thermal comfort and social dependence — that cannot be directly observed.
"Animals can't tell us when they're hungry, cold or fatigued," Wang said. "By inferring those latent needs from behavior and environmental cues, we can better understand what they're experiencing and provide more meaningful support."
Rather than treating behaviors as isolated events, Behav2Need analyzes behavioral patterns, environmental conditions and multimodal sensor data to infer the hidden factors influencing behavior over time. The result is a more interpretable understanding of animal welfare and the environmental conditions that shape it.
Tested in a Commercial Swine Facility
To evaluate the framework, the research team deployed Behav2Need at the U.S. Meat Animal Research Center in Clay Center, Nebraska. The study monitored 82 piglets across eight farrowing crates over approximately 960 hours using synchronized video and vibration sensing.
Behav2Need achieved 70.6% accuracy in recognizing piglet behaviors and 71.3% accuracy in forecasting future behaviors, outperforming existing machine learning and temporal modeling approaches. Researchers also found that explicitly modeling hidden needs and behavior duration significantly improved predictive performance.
Beyond improving prediction accuracy, the framework revealed interpretable relationships between animal needs and behavior. For example, sleeping behavior was associated with stronger thermal and social needs, while nursing behavior showed high social dependence. Active behaviors corresponded to moderate energy demands and lower social dependence.
The framework also captured meaningful dependencies among contextual factors, sensor signals, and animal behavior. Heat-lamp usage, sow posture, piglet location, movement intensity, and vibration patterns each contributed differently to the model’s behavioral predictions.
"Simply identifying behaviors such as sleeping, nursing or being active tells us what is happening," Wang said. "Understanding the needs behind those behaviors helps explain why they emerge under specific conditions, making it possible to detect potential welfare concerns earlier and support more targeted interventions."
Looking Beyond Smart Farms
Pre-weaning mortality remains a significant challenge in swine production, with nearly one in five piglets failing to survive until weaning. By uncovering the hidden motivations behind animal behavior, Behav2Need offers a new pathway for proactive monitoring and decision-making in livestock environments.
While the current study focused on piglets, the researchers envision broader applications beyond agriculture. The same behavior-to-need framework could eventually be adapted for human-centered smart environments, where everyday ambient sensing may help infer and track changing needs without relying on continuous self-reporting.
The research team included Ruiqing Wang and Yiwen Dong of the University of Illinois Urbana-Champaign; Jiale Zhang and Sungmin Lee of the University of Michigan; and Jeremy Miles and Gary Rohrer of the U.S. Meat Animal Research Center.
The paper, "Behav2Need: Ethology-Driven Animal Need Understanding via Multimodal Sensing on Smart Farms," was presented at the 13th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys 2026).
Yiwen Dong is an Illinois Grainger Engineering assistant professor in the Department of Industrial and Enterprise Systems Engineering with affiliations in the Department of Electrical and Computer Engineering and the Department of Civil and Environmental Engineering