Toward Perception‐Native Electronic Skin: Bio‐Inspired In‐/Near‐Sensor and Neuromorphic Computing for Humanoid Robots
TL;DR
These point toward a perception‐native electronic skin, whose stimulus‐to‐perception pathway is built into the skin, not a remote processor—a foundation for humanoids that act by touch, not just sight.
Abstract
Humanoid robots enter human environments, where manipulating objects with dexterity, keeping balance while walking, and staying safe under unexpected contact depend on touch; unlike vision, touch must cover the whole body. Individual sensors already rival human sensitivity, but no single design fits every region: each needs its own mix of density, sensitivity, and modality, but no whole‐body skin yet integrates them. This review traces the problem across the device, system, and intelligence levels, and finds the obstacle to be less the device than the system. Dense tactile streams are read out, relayed, and closed into control loops, but interconnect proliferation and von Neumann data movement keep them from closing in milliseconds. Array‐level fixes relieve the wiring burden only by shifting it onto computation, leaving a trade‐off between how richly a region can sense and how fast it can act. Resolving this without sacrificing latency demands bio‐inspired sensor–computing integration—near‐sensor, in‐sensor, and neuromorphic schemes emulating the body's receptors and the synapses and neurons relaying their signals, concentrating computation where it is needed. These point toward a perception‐native electronic skin, whose stimulus‐to‐perception pathway is built into the skin, not a remote processor—a foundation for humanoids that act by touch, not just sight.