Active Sensing for Autonomous Robotic Ultrasound: Probe Orientation Control Using Dense Surface Perception
Abstract: Ultrasound imaging quality strongly depends on probe orientation and operator expertise, limiting consistency across clinical settings. This work presents an active-sensing robotic ultrasound system that autonomously maintains optimal probe alignment using dense surface perception. The proposed end-effector, A-SEE2.0, integrates dual RGB-D cameras to estimate local surface geometry and enforce normal probe contact without relying on preoperative models. By tightly coupling perception with real-time robot control, the system adapts to complex, uneven anatomies. Experimental validation on phantom and mannequin surfaces demonstrates robust normal alignment, while in-vivo forearm studies show image quality comparable to expert manual scanning. This approach highlights the potential of perception-driven control for scalable and operator-independent robotic ultrasound.
Advisor: Professor Haichong Zhang
Committee: Professor Giovanni Pittiglio, Professor Connor McCann