AUTOMATED CATTLE INJECTIONS: A ROBOTIC AND DIGITAL TWIN APPROACH TO FEEDLOT EFFICIENCY

Emilio Hurtado, Chandra Suryadevara, Jayden Walker, Gerard Tayag, Hardeep Ryait

Keywords: Robotic Vaccination; Digital Twin; Artificial Intelligence; Precision Livestock Farming; Cattle Feedlots; NVIDIA Isaac Sim; Agricultural Automation

TL;DR: Simulation-first robotic cattle vaccination: a digital twin in NVIDIA Isaac with artificial intelligence and PPO control achieves sub-centimeter targeting, de-risking feedlot automation.

Abstract: Robotics adoption in feedlots is slowed by cost and risk around unproven systems. We present a simulation-first pipeline for automated cattle vaccination built as a digital twin in NVIDIA Isaac Sim. The system integrates an AI vision module (fine-tuned Detectron2) to segment the injection region from simulated RGB-D input and a Proximal Policy Optimization (PPO) controller to guide a robotic manipulator to the target. We validated the framework on a Franka Emika Panda arm and then adapted it to a custom 6-DOF manipulator. Policies were trained under 5 s, 10 s, and 15 s task durations and evaluated over 50 episodes each. The agent achieved sub-centimeter final position error across all conditions; the 10 s policy yielded the lowest median error (0.456 cm), while the 15 s policy was most consistent (median absolute deviation 0.063 cm). These results demonstrate precise, reliable targeting and show how time-horizon-specific policies provide operational flexibility. Overall, the digital-twin approach de-risks development, accelerates sim-to-real transfer, and offers a practical path toward scalable, safe automation in Precision Livestock Farming.

Submission Number: 57