Tuning nonlinear mechanical metamaterials for fully-analog sensing and robotic control

Physical Reservoir Computing (PRC) leverages the nonlinear response of physical systems for distributed, non-electronic computation. In this paradigm, a trained, typically large linear readout extracts useful information from a rich set of physical nonlinearities. Flexible mechanical metamaterials, known for buckling-induced pattern formation and tunable wave propagation, offer an ideal platform for this approach. Inspired by PRC, we harness the geometric tunability of metamaterial networks composed of flexibly coupled rigid units to perform in-physico sensing and robotic control. We first show that the metamaterial can achieve proprioception: its global deformation can be reconstructed from sparse strain measurements. Building on this capability, we implement a fully analog tactile control strategy—akin to biological thigmotaxis—to guide a wheeled robot. In this demonstration, an optimally designed metamaterial "skin" serves simultaneously as sensor and controller: its nonlinear deformation, measured by few strain sensors and linearly processed by an analog circuit, directly drives motor voltages without digital computation. These results illustrate how flexible mechanical metamaterials can physically embody computation, enabling adaptive, electronics-free robotic intelligence.