Adaptive neural controller with predefined boundary constraints for 3D printed cable-driven soft actuators

It is difficult to build dynamic models for 3D-printed soft actuators because of their material and structural flexibility and the complex intrinsic and extrinsic interactions encountered in human-centric or complex non-structural environments. Soft actuators require the control error during motion to be limited. However, existing control methods with predefined boundary constraints are designed for rigid actuators and are not directly applicable to soft actuators. In this paper, an adaptive neural controller based on a quasi-static model is proposed. The quasi-static model of the soft actuator is used to determine how the viscoelasticity of the flexible material influences the neural network, enabling the neural network to identify a better fit to the dynamic model of the soft actuator. Finally, experimental results verify that the proposed controller constrains the tracking error of the soft actuator to within the predefined boundary.