UKRI AI CDT in Dependable and Deployable AI for Robotics (CDT-D2AIR)

Baoyan Si

Project
Reinforcement Learning-Based Magnetic Field Control for Ultrasound-Guided Medical Nanorobot Swarms

Summary
My research interest/project summary: My research focuses on developing a hybrid perception and control framework for medical nanorobot swarms using ultrasound imaging, where real-time ultrasound data is processed through deep learning-based segmentation to extract vessel structures and target regions, enabling state estimation in a highly noisy and partially observable biological environment. This information is then integrated with physics-informed models of blood flow and used by a reinforcement learning (RL) controller, operating externally to generate optimal magnetic field actuation signals that guide the collective behaviour of nanorobots through complex vascular networks. The system aims to achieve precise, adaptive navigation and targeted drug delivery by combining data-driven learning, fluid dynamics, and global field control, without requiring onboard computation within the nanorobots themselves.