Percutaneous Coronary Intervention (PCI) relies on manual, fluoroscopy-guided steering of a guidewire throughthe coronary vasculature. This procedure is technically demanding, exposes both patient and operator to ionizing radiation, and outcomes depend heavily on operator skill and experience. Robotic assistance for guidewire navigation has been explored extensively, spanning classical model-based planning, reinforcement learning, and imitation learning approaches. However, most existing work either validates purely in simulation or relies on continuous fluoroscopic feedback during execution, which does not address the radiation-exposure motivation that partly drives interest in automation in the first place.
This project proposes to design, implement, and validate a vision-guided, closed-loop control system for an existing PCI actuator (translation and rotation) operating on 3D-printed vascular phantoms. Vessel geometry is captured once via camera imaging at the start of a run; a classical, model-based planning algorithm analyses and segments the image, computes a target path, and a closed-loop controller uses ongoing camera-based tip tracking to correct for actuation and model uncertainty during execution.
