Data-driven static position modeling of a parallel tendon-actuated soft continuum robot using artificial neural networks

Abstract Parallel soft continuum robots combine the compliance and adaptability of soft structures with the structural advantages of parallel mechanisms, offering promising capabilities for complex robotic applications. However, their highly nonlinear behavior presents significant challenges for accurate and computationally efficient modeling. This study presents a data-driven approach for static end-effector position modeling of a parallel tendon-actuated soft continuum robot using an artificial neural network (ANN). A physics-based model of the robot is developed in the SoRoSim MATLAB framework using the Geometric Variable-Strain (GVS) formulation, with gravity incorporated into the static simulations. Simulation data are then used to train the ANN to learn the nonlinear mapping between the tendon actuation forces and the Cartesian position of the end effector. The trained ANN demonstrates close agreement with the SoRoSim-generated positions over the investigated input domain while providing considerably faster position predictions than direct SoRoSim evaluations. This computational advantage highlights the potential of the proposed data-driven model for applications requiring repeated and rapid model evaluations, particularly in future control-oriented implementations. The presented results establish the ANN as an efficient surrogate for the considered SoRoSim-based static position model, while the current evaluation remains simulation-based.