Design and Shape Control of Robotic Morphing Interface With Reprogrammable Stiffness Based on Machine Learning

Deformable organisms in nature inspire the design of shape-shifting robots, including soft robots, bionic robots and physical human-robot interfaces. However, to achieve multi-objective shape imitation and multi-form transformation, shape-shifting robots often require complex actuation systems, control strategies, and inverse design algorithms. In this letter, we propose a robotic morphing interface with reprogrammable stiffness (RoMI-RS) based on machine learning. RoMI-RS uses a circular elastic bilayer as the base, which can produce isotropic deformation under pneumatic actuation. By repeatedly attaching and detaching high-stiffness limiting layers to the surface of the base, the stiffness distribution can be reprogrammed, guiding anisotropic deformation. Thus, without changing the base material or actuation mechanism, RoMI-RS can precisely mimic various static shapes and dynamic movements. To address the nonlinear coupling of soft materials and pneumatic actuation, we employed a data-driven approach to inversely design limiting layer arrangements (i.e., the stiffness distribution of RoMI-RS) in the form of images. Hence, our proposed pneumatic RoMI-RS not only responds quickly and deforms reversibly but also allows users to intuitively and rapidly reconfigure target shapes. We also demonstrate the applications of RoMI-RS in shape-shifting robotics, particularly in soft grippers and physical human-robot interfaces, verifying its deformation flexibility and adaptability.