Static Modeling of A Soft Robotic Manipulator Using Neural Networks and Computer Vision Techniques

This research explores the static modeling of a planar soft robotic arm composed of two parallel silicon-based actuators using neural networks. Soft robots, characterized by their pliable structure and pneumatic actuation, present challenges in traditional control and modeling due to their nonlinear behavior and material flexibility. Conventional methods often fail to accurately predict their motion, especially under varying internal pressures and structural imperfections. In contrast, neural network-based approaches offer robust solutions by effectively capturing these complexities through data-driven techniques. Utilizing a Levenberg-Marquardt algorithm for training, this study develops a neural network model that predicts the 2D spatial position of the robotic arm's end-effector based on input pressure variables. Experimental data, obtained via image processing techniques for precise position tracking, informs the model's training and validation phases. Results demonstrate high accuracy in predicting end-effector positions, validated by low mean squared errors and high coefficients of determination. This approach not only advances static modeling capabilities for soft robotic systems but also lays groundwork for future dynamic modeling and real-time control applications.