In a cardiovascular intervention the surgeon stands beside the patient, pushing a guidewire through vessels they can only see on X-ray — and absorbing that X-ray for every minute of the procedure.
This project removes the hand from the wire. A hard-magnetic guidewire is steered from outside the body by a permanent magnet on an industrial robot arm, which bends and turns the tip through tortuous vasculature without anyone standing in the beam.
The system
Two subsystems, one operator.
Teleoperation. A single Xbox controller drives all three actuators at once — the industrial arm carrying the magnet, the feeding mechanism advancing the wire, and the distal rotary motor. Real-time filtering keeps latency low enough that a human can take over the moment something looks wrong.
Control platform. A Qt/PyQt application on the PC: GUI built in QtDesigner, with a script editor, console, live camera view and one-click macros. Procedures can be scripted in Python and replayed, which is what turns a manual rig into a repeatable one. An emergency stop halts every device through a multi-threaded control path.
Modeling the wire
Predicting where a magnetically loaded wire will go means modeling how it deforms. Constant-curvature and Kirchhoff models are not enough here, so the wire is treated as a Cosserat rod — a slender body free to bend, twist, stretch and shear.
The rod is discretised with a finite-difference scheme that couples position, orientation, curvature, and internal forces and moments. Magnetic torque enters both from a uniform field and from the dipole field of the external permanent magnet, so the simulation runs under the spatially varying field the wire actually sees.
Sweeping these simulations builds a lookup from magnet pose to tip deflection. That table is what the path-following controller steers against.
In the phantom
A transparent 3D-printed vascular phantom with multiple bifurcations stands in for patient anatomy. The complete system — controller, software, vision feedback, magnetic actuation — was evaluated on two tasks.
Following a path. The guidewire tracks a preset route through the branches and reaches the target under autonomous control.
Recovering from a wrong turn. When the wire enters the wrong branch, the system sees the deviation on camera, retracts to the previous junction, recomputes the magnet pose from the rod model, and advances again along the correct path. Catching a mistake and undoing it is the part that matters clinically.
My contribution
- The unified manual control framework — Xbox controller through the PC to the arm, feeder and distal motor — including real-time filtering and signal routing.
- The Qt/PyQt platform: GUI, scripting interface, macros, and the multi-threaded emergency stop.
- Formulating and implementing the Cosserat-rod model, its finite-difference discretisation, and the magnetic loading.
- Running the phantom experiments, and integrating vision feedback with control to demonstrate autonomous navigation and error recovery.