In-Material Computation: A Computational Metamaterial for Data-Efficient Tactile Interfaces

Embedding computational capabilities directly into the physical structure of soft materials is a central goal for developing next-generation smart sensors and human-machine interfaces. However, achieving deterministic information processing within a compliant material remains a significant design and fabrication challenge. We introduce a "computational metamaterial" that physically performs information encoding through a deterministic process termed "mechanical compilation". This structured elastomer, embedded with a sparse optical sensing network, is engineered to deterministically map complex high-dimensional spatial pressure patterns, benchmarked using 26 distinct Braille characters, into unique low-dimensional optical signals with 100% classification accuracy. The physically encoded information is of such high quality that a synergistic physics-informed machine learning (PIML) decoder maintains over 96% accuracy with an 80% reduction in training data, demonstrating a profound enhancement in data efficiency. This work pioneers a structure-driven design paradigm for computational metamaterials, shifting the computational burden from software to the material itself and paving a new path toward highly efficient, low-complexity sensing systems.