Data-Driven Prediction of Magnetic Miniature Soft Robots States Based on Koopman Operator Theory

Magnetic miniature soft robots exhibit great potential for development in fields such as medicine and environmental exploration due to their outstanding flexibility and adaptability to different environments. However, the motion of these robots in a magnetic field is characterized by high nonlinearity, and the interaction between soft materials and their environment introduces uncertainty, making it challenging to establish traditional kinematic models. In this paper, a data-driven method based on Koopman operator theory is implemented to enable state prediction for magnetic miniature soft robots. First, a simulation environment is created to obtain motion data of the robot. Through deep learning, the nonlinear state space is mapped to a linear Koopman space, and then the extended dynamic mode decomposition (EDMD) method is used to approximate the Koopman operator for prediction. Finally, in a validation dataset with different magnetic field inputs, the prediction results are compared with the actual data to assess the deviation. The results indicate that the proposed data-driven prediction method can effectively infer the motion trends of the robot, demonstrating higher predictive performance on state variables with lower complexity. This provides new ideas and methods for the motion analysis and control of magnetic miniature soft robots.