Programmable Hydrodynamic Invisibility Enabled by Machine-Learning-Guided Metamaterials

Manipulation of fluid transport in porous media underpins a broad range of technological applications and natural processes. Hydrodynamic invisibility enables flow manipulation without disturbing the external flow field, with representative devices including cloaks, concentrators, rotators, and camouflage. However, most existing devices are static and thus fail when background permeability varies. Here we present a machine-learning-guided metamaterial strategy for programmable hydrodynamic invisibility, using a cloak as a model system. A tunable-permeability shell maintains cloaking across a wide range of background permeabilities and enables hydrodynamic camouflage by matching prescribed exterior reference flows. An inverse-design framework rapidly maps target permeability responses to manufacturable geometries, and experiments validate high-fidelity performance under high, medium, and low background permeabilities. These results indicate that programmable hydrodynamic metamaterials provide a scalable, general strategy for robust, on-demand manipulation of fluid transport in porous media, with potential applications in separation science, microfluidics, flow-network engineering, and soft-matter biomechanics and poroelasticity.