Neural Network-Based Control of a Tendon-Driven Soft Slender Manipulator

In this study, we propose a novel application case for the control of a tendon-driven soft manipulator using neural networks. The ability of neural network models to efficiently learn the nonlinear behavior of the system, coupled with their real-time usability, positions them as powerful tools for modeling and controlling soft manipulators. Our approach involves training a neural network to establish the mapping between motor angular positions and 3-D end-effector positions of a slender continuum manipulator. Various experiments are conducted to assess and validate the performance of the neural network-based manipulator modeling. Following the assessment of the manipulator’s accuracy in executing both static and dynamic tasks, we delve into its robustness with respect to variations in speed and task locations within the workspace. The results demonstrate a consistent precision across the entire operational environment, with no discernible impact from changes in speed. Finally, the investigation into the repeatability of the system shows minimal errors when executing the same trajectory multiple times. This work contributes to advancing the state-of-the-art in tendon-driven soft slender manipulator control, highlighting the effectiveness, robustness and repeatability of neural network models in addressing the challenges posed by complex and dynamic systems.