Pigs and cattle cannot tell a farmer when they are stressed or in pain, so researchers are building AI systems that read their emotions from body posture and movement alone. Human observers are too slow and expensive to monitor every animal on a large farm continuously, and people often misinterpret animal behaviour. This project tackles that gap by combining ethology—the science of animal behaviour—with deep learning models that analyse visual cues such as stance, gait, and ear position. The team will first study how humans themselves infer emotion from body language, then train neural networks to recognise those same signals in pigs and cattle. Crucially, they will model emotions not as simple labels like “happy” or “afraid” but as positions in a multi-dimensional space, allowing the AI to detect subtle shifts in emotional state over time. If successful, the system could enable continuous, automated welfare monitoring on farms, flagging distress earlier than human observation allows. That would make timely interventions possible—adjusting feed, enrichment, or housing—and reduce the environmental impact of poor welfare. The work is both fundamental and applied: it advances understanding of how animals express emotion while building a scalable tool for the livestock industry.
View original technical description
Context: Automated assessment of animal emotion is gaining attention as a potential means for improving farmed animal welfare and if realised could lead to step-changes in the ethical treatment of farm animals. By supplementing human observations, which are often labour-intensive and impractical, this approach offers a more efficient alternative. Advances in 'smart farming' suggest that AI models being explored for human emotion recognition could be adapted for animals by incorporating species-typical behaviours. Understanding these behaviours is key to developing advanced machine learning (ML) algorithms for detecting animal emotions. Humans naturally infer emotions in other humans and animals through body posture and movement - a skill likely evolved to anticipate behaviours (friend/foe). These cues form an ‘emotional body language’, but it remains unclear which features contribute to our perception of emotions across species and contexts. This project will investigate how body postures and movements inform emotion assessment in animals, applying this understanding to design and train AI models. Challenges: Unlike humans, animals cannot self-report their state, making recognition of their emotional states complex. Challenges include interpreting emotional cues, addressing species differences, accounting for environmental factors, and overcoming observer bias. We address these by co-developing an autonomous system that analyses animals’ body postures and movement to infer emotions. Ethology, the study of animal behaviour, provides tools for characterising behaviour in ways that allow inferences about emotions to be made, while AI focuses on identifying and modelling the behavioural markers. Aims/Objectives: We aim to determine whether animal emotions can be identified and tracked by a machine through visual analysis of behaviour (appearance/movement). By varying environmental conditions and observing behavioural responses, we will use ML to map these to emotional states. Instead of using basic emotion labels or a simple two-axis chart, we will model animal emotions in a rich multi-dimensional space created by a neural network. This can reveal new emotional patterns and show changes in emotional states as movements in this space, offering fresh insights for studying animal behaviour. Our aims are both fundamental and applied: we will use a ML framework to better understand what is spatially and temporally salient when recognising animal emotion and then deploy this knowledge to realise a machine able to infer emotion from behaviour. Studying pigs and cattle will reveal how well machine vision models transfer across species. Using transfer learning, we’ll test whether models trained on one species can accurately detect emotions in another. This will lay the groundwork for a scalable, biologically informed system for automated emotion recognition across species and contexts. Potential Applications/Benefits: This project offers potential to transform animal emotion recognition, enabling automated systems to detect emotions more accurately and continuously than human observers. For ethologists, it will inform upon the nuances of behavioural expression of emotion, how this varies within and between individuals and across species, allowing for the testing of alternative conceptual models of emotional states. For computer scientists, it incorporates insights from ethology and human observation - such as learning through observing dynamic, uncertain real-world data, as humans and animals do - into the modelling of complex emotional patterns, helping to realise new kinds of models. Welfare assessments on farms face major technical barriers to widespread adoption. This project aims to drive system-level change by enabling better monitoring, enrichment, and timely interventions to improve welfare and reduce environmental impact.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know