Full-workspace deformation prediction of magnetic soft continuum robots based on State-Dependent Switching Physics-Informed Neural Network

Abstract

Abstract Magnetic soft continuum robots (MSCRs) exhibit significant potential for applications in complex and constrained environments, such as minimally invasive surgical interventions and catheter navigation, owing to their unique advantages of remote actuation, high compliance, and adaptability. However, accurately describing the physical evolution mechanism of MSCR transitioning from small to large deformations is challenging due to the nonlinear behaviours caused by magneto-mechanical coupling effects and large deformations. To address this challenge, this paper proposes a State-Dependent Switching Physics-Informed Neural Network (SDS-PINN) for achieving high-precision Prediction of the nonlinear deformation of MSCR across the whole deflection domain (0$^{\circ }180^{\circ }$). This method is based on constructing multiphysics coupling models for small and large deformation stages. It dynamically adjusts physical constraints using a state-dependent switching (SDS) mechanism, enabling adaptive switching and continuous modelling between different deformation modes. The effectiveness of the proposed SDS-PINN approach is validated through numerical simulations and comparative experimental studies. Results show that SDS-PINN outperforms traditional PINN in different deformation modes, with a reduction in free-end displacement Prediction error by over 99% in large deformation scenarios. This study proposes a unified and physically interpretable framework for the deformation modelling and control of MSCRs, which significantly enhances prediction accuracy and stability under complex working conditions.