Data-Driven Intelligent Shape Estimation and Control of Continuum Robotic Arms With Force Interaction

Soft manipulators have attracted significant attention in recent years due to their high maneuverability and ability to adapt to complex environments. However, modeling, shape estimation, and precise control of these systems remain challenging. This paper presents a novel intelligent approach for shape estimation and control of tendon-driven continuum robotic arms in the presence of force interaction with environment based on a large set of experimental tests. To this end, using data obtained from two orthogonal cameras, a vision-based method capable of reconstructing the 3D position of the robot backbone is developed. Based on the experimental dataset that includes interaction forces, tendon tensions, and backbone-disk positions, a mapping is established between system variables (i.e. positions and applied forces) and the considered outputs (i.e. configuration parameters). These parameters are then fed into a Physics-Informed Neural Network (PINN) for accurate estimation of the manipulator shape. Additionally, this paper presents a comprehensive experimental evaluation of advanced model-based control strategies for continuum soft robots. Leveraging the PINN model, four control methods, pure feedforward, enhanced PD, hybrid, and adaptive PD, are implemented and experimentally validated in a point-to-point trajectory adjustment. Obtained results demonstrate high accuracy in estimating the deformation behavior of the soft manipulator under interacting force conditions, supporting future developments in real-time control.