Deep learning-based structural design and mechanical properties analysis of pneumatic actuators with tunable multistability

Pneumatic actuators have attracted increasing attention owing to their simple fabrication, large deformation capability, and high flexibility. However, most existing pneumatic actuators must be continuously pressurized to sustain tensile and bending deformations, which hinders precise control of deformation modes. In this study, a tunable multi-stable pneumatic actuator is developed based on the mechanical characteristics of Kresling origami and B-spline curved beams. Deformation modes with multiple degrees of freedom are achieved. First, the steady state characteristics of the Kresling origami are analyzed using a truss model, and its force-displacement relationship is experimentally obtained. Then, based on an optimization design framework, a deep learning (DL) neural network model is established to predict the mechanical properties of the B-spline curved beam. To evaluate the influence of geometric parameters on the mechanical properties of the curved beam, Shapley additive explanations (SHAP) analysis is performed. Finally, the design paradigm of a pneumatic arm and gripper based on the proposed actuator is demonstrated. The tunable multi-stability of the actuator under negative pressure and its capability to grip objects with diverse shapes and masses are demonstrated. This study provides a reference for the design of soft robots exhibiting multiple deformation modes, reconfigurability, sequential deployment, and versatility.