The current thyroid ultrasound relies heavily on the experience of the sonographer and radiologist, which would be physically and cognitively exhausting. Here we report a novel fully autonomous robotic ultrasound system (FARUS) that can scan thyroid regions without human assistance. In this application, human skeleton point recognition, image segmentation, and force feedback are used to solve the problem of thyroid localization. The orientation of the ultrasound probe is adjusted via Bayesian optimization. By using FARUS, both transverse and longitudinal scanning of thyroid tissue can be performed, as well as real-time thyroid nodule detection with artificial intelligence (AI) algorithms. Experimental results on human participants demonstrate that FARUS can perform high-quality ultrasound scans in comparison with manual scans obtained by clinicians.