Sensing Intelligence as a Trainable Metamaterial Property

2026-05-13

AI Summary

This work proposes embedding perceptual intelligence directly into metamaterial bodies, departing from conventional sensing systems that delegate all signal processing to electronic and computational units while treating mechanical structures as passive components devoid of physical preprocessing capabilities. By leveraging differentiable physics-based simulation, the approach enables end-to-end backpropagation of neural network perception loss to the geometric design parameters of the metamaterial, thereby allowing the physical body to autonomously optimize its response and shape incoming stimuli. This represents the first integration of mechanical structure design into an end-to-end perceptual learning framework, challenging the long-standing hardware-software separation paradigm. Experimental results demonstrate that the optimized metamaterial bodies can improve perception accuracy by up to fivefold or reduce sensor requirements by nearly an order of magnitude.

Abstract

In biological systems, sensing is not performed by the brain alone: the body deforms, vibrates, and filters external stimuli before they are transduced into neural signals. In engineered systems, this processing burden is placed largely on electronics and computation, while the mechanical body is usually designed only for strength and stability. Here, we present sensing intelligence as a trainable property of the body. We show that the geometry of a metamaterial can be optimized to reshape external stimuli into internal signals that are easier for a neural network to interpret. Rather than hand-designing this physical preprocessing, we let the neural network train its own body for sensing by backpropagating the sensing loss to the body's design parameters through differentiable simulation. Across numerical and experimental sensing scenarios, the optimized body improves sensing accuracy by up to fivefold or reduces the number of required electronic sensors by nearly an order of magnitude.

Problem

• sensing intelligence
• metamaterial
• physical preprocessing
• trainable property
• body-embedded sensing

Innovation

• sensing intelligence
• trainable metamaterial
• differentiable simulation
• physical preprocessing
• body-brain co-design