Abstract

Low vaccination efficiency and high labor demand constrain large-scale sheep farming. This study developed an automated vaccination robot integrating posture constraint, vision-based injection-site localization, and automated needle handling in a unified “constrain-localize-inject” workflow. The system comprises a hindquarter posture-correction frame, a bilateral neck-restraint mechanism with pressure feedback, and a vision-guided injection unit with automatic needle disposal. A Kendryte K210 edge-computing module enabled real-time localization of the cervical muscle region. Finite element analysis of the restraint column under a 100 N lateral load showed a maximum von Mises stress of 1.832 MPa, well below the 220.6 MPa yield strength of plain carbon steel, and a maximum displacement of 0.003 mm. Functional trials on 20 life-size artificial sheep models were completed without failure. A 500-cycle bench endurance test achieved a 95.8% overall success rate, although performance declined from 100% to 88% across successive test sets. The vision model achieved 95.86% precision, 92.67% recall, and 94.15% mAP@0.5, with a localization error of 5.6 ± 2.1 mm and an inference latency of 38.6 ms per frame. Laying the foundation for subsequent live animal experiments.

IPC Classification

Keywords