Data-driven design of shape-programmable magnetic soft materials

Abstract

Magnetically responsive soft materials with spatially-encoded magnetic and material properties enable versatile shape morphing for applications ranging from soft medical robots to biointerfaces. Although high-resolution encoding of 3D magnetic and material properties create a vast design space, their intrinsic coupling makes trial-and-error based design exploration infeasible. Here, we introduce a data-driven strategy that uses stochastic design alterations guided by a predictive neural network, combined with cost-efficient simulations, to optimize distributed magnetization profile and morphology of magnetic soft materials for desired shape-morphing and robotic behaviors. Our approach uncovers non-intuitive 2D designs that morph into complex 2D/3D structures and optimizes morphological behaviors, such as maximizing rotation or minimizing volume. We further demonstrate enhanced jumping performance over an intuitive reference design and showcase fabrication- and scale-agnostic, inherently 3D, multi-material soft structures for robotic tasks including traversing and jumping. This generic, data-driven framework enables efficient exploration of design space of stimuli-responsive soft materials, providing functional shape morphing and behavior for the next generation of soft robots and devices.