Data-Driven Correction Kinematic Modeling and Control for Continuum Robots

To address the issues of high-precision modeling and real-time reconstruction for the nonlinear large-deformation kinematics of continuum robots, a variable curvature kinematic model for continuum robots was proposed, which combines a knowledge-based analytical kinematic model with a data-driven machine learning model. A Gaussian process regression (GPR) model was used to modify the piecewise constant curvature (PCC) kinematic model with a small amount of experimental data, improving the prediction and open-loop control accuracy of the robot's end pose. A linear variable curvature (LVC) model was adopted to describe the variable-curvature overall configuration of the continuum robot. The configuration parameters were then obtained by solving an inverse kinematics optimization problem based on the corrected feature-point poses, enabling high-precision real-time visualization. A remote operating system for a compact cable-driven continuum robot was used for control experiments and configuration prediction. The results indicate that the proposed method can significantly improve the accuracy of configuration calculation and trajectory planning while maintaining high computational efficiency.