Farm-Scale Autonomous Welfare Monitoring in Precision Livestock: A Systematic Review of Robotics and Multimodal AI with an Emphasis on the Lab-to-Farm Deployment Gap

Francois Gonothi Toure, Mounir Boukadoum, Abdoulaye Banire Diallo

Keywords: precision livestock farming, dairy cattle, animal welfare monitoring, autonomous navigation, computer vision, multimodal AI, edge computing, technology adoption, systematic review, welfare representation learning

TL;DR: We systematically reviewed autonomous cattle monitoring and found that the key challenge is no longer developing foundational technology, but ensuring it can be reliably and scalably deployed on real farms.

Abstract: While breakthroughs in autonomous robotics and multimodal artificial intelligence (AI) promise continuous, real-time monitoring for precision livestock farming, their practical on-farm application faces significant limitations, revealing a critical lab-to-farm deployment gap. This deployment gap is rooted in fundamental challenges relevant to the embodied AI community: poor model generalization, sim-to-real fragility, and the absence of standardized validation benchmarks. The primary objective of this systematic review is to highlight state-of-the-art knowledge on these technologies to understand and bridge this gap, proposing a path forward that benefits both agricultural practice and on-farm research. From a pool of over 900 articles reviewed on autonomous navigation and AI-driven analytics, we systematically selected 33 studies to propose recommendations for adopting farm-scale autonomous monitoring in precision livestock. Our review reveals that while foundational technologies are well-established, research remains fragmented and often limited to laboratory simulations or small, single-farm field trials. Based on these findings, we propose a deployment-oriented roadmap with recommendations for developing integrated, robust, and scalable systems. Furthermore, we pinpoint a critical deficiency, that is, the lack of a standardized learning representation (ontology/schema) for collected welfare insights. This deficiency prevents the creation of reproducible datasets for the embodied AI community, hindering the development of truly robust and generalizable models for livestock welfare.

Primary Area: applications to robotics, autonomy, planning

Submission Number: 23543