A fully autonomous robotic ultrasound system for thyroid scanning has been developed to address the limitations of traditional ultrasound (US) methods, which rely heavily on the expertise of sonographers and radiologists. This system, named FARUS, enables autonomous scanning of the thyroid without human assistance, identifies malignant nodules, and provides data for the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) calculation. The system uses human skeleton point recognition, reinforcement learning, and force feedback to locate thyroid targets, dynamically adjusts the ultrasound probe orientation via Bayesian optimization, and employs deep learning techniques for real-time segmentation of the thyroid gland and potential nodules. The system was tested on human participants and demonstrated high-quality scans comparable to those obtained by clinicians. It also has the potential to detect thyroid nodules and provide data for ACR TI-RADS calculation. The system consists of a six-degree-of-freedom UR3 manipulator with a linear US probe, a US probe fixture, and a six-axis force/torque sensor. The system automatically scans the thyroid in four phases: thyroid searching, in-plane scanning, out-of-plane scanning, and multi-view scanning. The system uses reinforcement learning to adjust the probe's movement until the thyroid gland is accurately located and Bayesian optimization to optimize the probe orientation. The system also uses deep learning techniques for real-time segmentation of the thyroid gland and potential nodules. The system was tested on 19 patients and demonstrated high-quality scans with accurate nodule detection and classification. The system was also compared with manual scans and showed similar performance in terms of image quality, centering error, orientation error, and image entropy. The system was found to be safe and effective, with participants reporting no adverse reactions. The system has the potential to be used in clinical settings for rapid screening and diagnosis of thyroid nodules. The system is expected to be used in specialized hospitals, clinics, and remote areas, providing an early warning of thyroid nodule development. The system operates on an autonomous scanning mode, which minimizes the risk of transmitting infectious diseases between patients and healthcare providers. The system has shown promising results in detecting thyroid nodules and providing data for ACR TI-RADS calculation. However, the system needs further improvements, especially for small-scale and low-contrast nodules. The system also needs to incorporate video streams for better performance.A fully autonomous robotic ultrasound system for thyroid scanning has been developed to address the limitations of traditional ultrasound (US) methods, which rely heavily on the expertise of sonographers and radiologists. This system, named FARUS, enables autonomous scanning of the thyroid without human assistance, identifies malignant nodules, and provides data for the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS) calculation. The system uses human skeleton point recognition, reinforcement learning, and force feedback to locate thyroid targets, dynamically adjusts the ultrasound probe orientation via Bayesian optimization, and employs deep learning techniques for real-time segmentation of the thyroid gland and potential nodules. The system was tested on human participants and demonstrated high-quality scans comparable to those obtained by clinicians. It also has the potential to detect thyroid nodules and provide data for ACR TI-RADS calculation. The system consists of a six-degree-of-freedom UR3 manipulator with a linear US probe, a US probe fixture, and a six-axis force/torque sensor. The system automatically scans the thyroid in four phases: thyroid searching, in-plane scanning, out-of-plane scanning, and multi-view scanning. The system uses reinforcement learning to adjust the probe's movement until the thyroid gland is accurately located and Bayesian optimization to optimize the probe orientation. The system also uses deep learning techniques for real-time segmentation of the thyroid gland and potential nodules. The system was tested on 19 patients and demonstrated high-quality scans with accurate nodule detection and classification. The system was also compared with manual scans and showed similar performance in terms of image quality, centering error, orientation error, and image entropy. The system was found to be safe and effective, with participants reporting no adverse reactions. The system has the potential to be used in clinical settings for rapid screening and diagnosis of thyroid nodules. The system is expected to be used in specialized hospitals, clinics, and remote areas, providing an early warning of thyroid nodule development. The system operates on an autonomous scanning mode, which minimizes the risk of transmitting infectious diseases between patients and healthcare providers. The system has shown promising results in detecting thyroid nodules and providing data for ACR TI-RADS calculation. However, the system needs further improvements, especially for small-scale and low-contrast nodules. The system also needs to incorporate video streams for better performance.