Physics-Informed Neural Networks for Quasi-Static Tip Position Control of Magnetic Catheters

Magnetic catheters offer promise in biomedical applications, but accurate, real-time tip prediction under electromagnetic actuation is difficult. We propose a physics-informed neural network (PINN) that integrates data-driven learning with physical constraints—force/torque balance, energy minimization, and shape bounds—to estimate equilibrium configurations under eight-coil actuation. Trained on 10,000 first-principles simulations, the model achieves a mean tip error of 2.1mm in 3D and stays below 3mm in 95% of cases across 0–3A inputs. Physics-based losses cut extreme errors by over 60% versus purely data-driven models, improving generalization from limited data. The framework enables scalable, interpretable digital twins for precision control of soft robots in clinical and industrial settings.